[HN Gopher] AMD records its highest server market share in decades
___________________________________________________________________
AMD records its highest server market share in decades
Author : giuliomagnifico
Score : 214 points
Date : 2024-08-12 13:40 UTC (9 hours ago)
(HTM) web link (www.tomshardware.com)
(TXT) w3m dump (www.tomshardware.com)
| jmakov wrote:
| Will be interesting to see how Intel recovers if at all.
| Actually, is there today (or since 1 or 2 years) any reason to go
| with Intel on desktop or server?
| ISV_Damocles wrote:
| I personally only use AMD (excepting one test machine), but
| Intel does have the best single-thread performance[1] so if you
| have some crufty code that you can't parallelize in any way,
| it'll work best with Intel.
|
| [1]: https://www.tomshardware.com/reviews/cpu-
| hierarchy,4312.html...
| Rohansi wrote:
| Unless your workloads are not very cache optimized like most
| games, then AMD's 3D V-cache CPUs take the lead.
| adrian_b wrote:
| The new Zen 5 has a much better single-thread performance
| than any available Intel CPU.
|
| For instance a slow 5.5 GHz Zen 5 matches or exceeds a 6.0
| GHz Raptor Lake in single-thread performance. The faster Zen
| 5 models, which will be launched in a couple of days, will
| beat easily any Intel.
|
| Nevertheless, in a couple of months Intel will launch Arrow
| Lake S, which will have a very close single-thread
| performance to Zen 5, perhaps very slightly higher.
|
| Because Arrow Lake S will be completely made by TSMC, it will
| have a much better energy efficiency than the older Intel
| CPUs and also than AMD Zen 5, because it will use a superior
| TSMC "3 nm" process. On the other hand, it is expected to
| have the same maximum clock frequency as AMD Zen 5, so its
| single thread performance will no longer be helped by a
| higher clock frequency, like in Raptor Lake.
| listic wrote:
| > in a couple of months Intel will launch Arrow Lake S,
| which will have a very close single-thread performance to
| Zen 5
|
| Will they? Intel Innovation event was postponed "until
| 2025"[1], so I assumed there is not going to be any big
| launch like that in 2024, anymore? Arrow Lake S was
| supposed to debut at Intel Innovation event in September
| [2]
|
| [1] https://www.intel.com/content/www/us/en/events/on-
| event-seri...
|
| [2] https://videocardz.com/newz/intel-says-raptor-lake-
| microcode...
| adrian_b wrote:
| The Intel Innovation event was canceled to save money.
| This has nothing to do with the actual launch of future
| products, which are expected to bring more money. Intel
| can make a cheap on-line product launch, like most such
| launches after COVID.
|
| Since the chips of Arrow Lake S are made at TSMC and
| Intel does only packaging and testing, like also for
| Lunar Lake, there will be no manufacturing problems.
|
| The only thing that could delay the Arrow Lake S launch
| would be a bug so severe that it would require a new set
| of masks. For now there is no information about something
| like this.
| etempleton wrote:
| For some specific workloads. Many people in the Plex community
| will point to Quicksync as being superior for media encoding
| than anything that AMD offers.
|
| The next couple of years will be interesting. If Intel can pull
| off 18a they will very likely retake the lead, however if 18a
| underwhelms Intel will be in real trouble. There is also the
| possibility that AMD chooses Intel as a fab in the future,
| which would feel a bit like when Sonic showed up on a Nintendo
| console (for some reason it just feels wrong).
| darknavi wrote:
| I was going to say Quicksync. I just spec'd and bought a new
| server build and the igpus on a few year old Intel chips is
| hard to beat for price and tdp.
| epolanski wrote:
| I think Intel still takes the crown when it comes to maximizing
| every single fps.
|
| Also, for a while, it was more competitive in the mid range
| than AMD's equivalent from a price/performance point of view.
| tracker1 wrote:
| This is where I think AMD really messed up on the current
| releases... If they'd called the 9600X a 9400 and the 9700X a
| 9600X, and charged like $100 less, I think the reception
| would have been much stronger. As it stands, it's more about
| wait and see. You can use PBO and get like 10-15% more
| performance at double the energy, but even then compared to
| the rest of the market, it just comes up short.
|
| The 9800X3D and 9950X(3D) options will really carry this
| cycle if they're good and the pricing adapts appropriately.
| I'm not holding my breath. I've been holding on with my 5950X
| since it's release and likely going to continue unless the
| 9950X(3D) is compelling enough. Not to mention the DDR5
| memory issues with larger sizes or more sticks. 96gb is
| probably enough, but at what cost.
| adrian_b wrote:
| The 9600X and 9700X are not a useful upgrade for gamers,
| but they are excellent for anyone else.
|
| Their energy efficiency is much better that for any
| previous x86 CPUs and for those who use applications that
| benefit from AVX-512 the desktop Zen 5 brings an increase
| in throughput higher than for any new CPU of the last five
| years (the last time when a desktop CPU had a double
| throughput over its predecessors was in 2019, with Zen 2
| over Coffee Lake Refresh). Also in the applications like
| Web browsers or MS Office, which prefer single-thread
| performance, they beat even the top Raptor Lake Refresh of
| 6.0 GHz, which is much more expensive and it consumes a
| power several times greater.
|
| Moreover, as shown by TechPowerUp, the performance of Zen 5
| under Windows is suboptimal in comparison with Intel in the
| programs that use a small number of active threads, like
| the games, because the Windows scheduler uses a policy that
| favors power savings over performance, even if that is a
| bad choice for a desktop CPU. That means that the scheduler
| prefers to make both threads of a core active, while
| keeping idle the other cores, even if the right policy
| (which is used on Intel) is to begin to use the second
| threads of the cores only after all the cores have one
| active thread.
|
| This should be easy to correct in the Windows scheduler,
| which does the right thing for Intel, where first one
| thread is made active on each P-core, the all the E-cores
| are made active, and only if more active threads are
| required the second threads of the P-cores are made active
| too.
| tracker1 wrote:
| I think it would depend on the cost of electricity.. as
| the performance for most is similar to the prior gen,
| which costs significantly less.
| Alupis wrote:
| Electricity costs is perhaps the last thing a pc gamer
| considers when choosing a new CPU. That's if they
| consider it at all...
|
| That would be like asking how many MPG's a rebuilt 1969
| Ford Mustang Boss 429 gets... it was not built for
| efficiency - it was built for performance.
| treprinum wrote:
| N100 & N305 have no competition in their own space IMO. For the
| rest I don't see any advantage outside momentum and "nobody is
| going to be fired for using IBM" approach.
| hangonhn wrote:
| This is fine for Intel actually. What really matters for the
| fortunes of Intel is their foundry service now. Intel could
| very well make those chips for NVidia if they can figure out
| the foundry service.
| foobiekr wrote:
| Strategically, in global terms from a US-centric point of view,
| it would be very good if Intel could recover, as having the
| entire world reliant on TSMC would be bad for everyone. Also
| there really isn't any other org like Intel at the moment in
| the broader sense in a few ways (mostly advanced research).
|
| That said, Intel really has nothing to fall back on.
| Internally, it is very much like many of the other big
| companies - run by executive whim papered over with
| justifications that are tissue thin. There are a lot of
| companies like this - Cisco, Intel, whatever the hell is left
| of IBM, Nokia, Lattice, NetApp, etc. - that will probably die
| this decade if they fail to reboot their executive and
| managerial cultures.
|
| I have a lot of friends who have been in the great parts of
| Intel (the CPU design teams and, believe it or not,
| fabrication) and the crap parts (everything else, especially
| networking and storage [what's left of it], except maybe the
| perenially suffering ICC team). Pat has failed to make things
| materially better because the system that is the body of Intel
| is resilient to disruptions of the sort he is trying. A massive
| replacement and flattening of culture is required there and he
| has not been able to execute on it.
|
| Turnover is coming for all the companies I list above. If I had
| to go out on a limb and suggest a longterm survivor from the
| list above, it might be Lattice. Otherwise everyone listed is
| driving hard straight to the ground internally, whether the
| market sees it or not.
| Miraste wrote:
| IBM has been heading toward the ground for thirty years, and
| they never seem to hit it. Whatever problems they have, they
| are both diversified and deeply entrenched in all kinds of
| government and S&P 500 systems and processes, to the point
| that their performance barely matters. Intel doesn't have
| anything like that to paper over their deficiencies.
| robotnikman wrote:
| Plenty of places still running AS/400 on crazy high support
| contracts, and no plans of moving on.
| marcosdumay wrote:
| > crazy high
|
| Not high enough that it can't go higher.
|
| Maybe somebody should make a betting site around when the
| next order-of-magnitude increase on the rent of AS/400
| hardware will come. With 6 months granularity.
|
| IMO, the people without a plan are crazy. But yeah,
| there's plenty of them.
| llm_nerd wrote:
| At some price points and for some scenarios, or if a particular
| pre-built has everything you need for the right price and
| happens to have Intel, sure there are plenty of situations
| where Intel is the right choice.
|
| It is amazing to see how Intel has faltered. This was all laid
| in stone a decade ago when Intel was under delusions that its
| biggest competitor was itself, and the only thing it cared
| about was ensuring that their lower power and lower cost
| devices didn't compete with their cash-cow high end products.
| They optimized for the present and destroyed their future.
| wmf wrote:
| Intel could come back once all their chips are on 3 nm later
| this year.
| scrapcode wrote:
| Is there any indication at all that this is due to businesses
| bringing their hardware back on-prem? Do you think "history
| repeats itself" will hold true in the on-prem v. cloud realm?
| wongarsu wrote:
| Anecdotally it seems like the rose-tinted glasses are coming
| off in terms of the amount of money and work saved by moving
| towards the cloud. Of course those aren't the only reasons to
| move to the cloud. But a lot of companies moved to the cloud
| due to promises and expectations that didn't pan out, so at
| least some of them moving back is only natural. Especially with
| many new solutions around private cloud and serverless allowing
| them to bring the "good parts" of the cloud to hardware they
| own or rent.
|
| That's just the normal hype cycle playing out.
| jeffbee wrote:
| Why do you believe that could be a related phenomenon? How much
| server market is going to non-cloud vs. cloud? Some analysts
| estimate the global server market last year at $95 to $150
| billion. AWS spent $48 billion on capital last year. Google
| bought $32 billion in capital, most of which is "on-prem" from
| their perspective. A switch in the machine-of-the-day from
| Intel to AMD at one of these large buyers could easily cause
| the slight change in market share mentioned in the article.
|
| Overall I think a lot of people continue to underestimate the
| size of the heavy hitters on the buy side of the server market.
| scrapcode wrote:
| Why wouldn't it be a viable reason? Every customer that AWS
| and Google have has the potential to make that decision and I
| believe it gets more appetizing with every data breach /
| funnel failure / etc.
| jeffbee wrote:
| Every indicator we have is that movement to the cloud is a
| massive tidal wave, and all we have on the retreat to on-
| prem is a handful of anecdotes and vibes. It's really up to
| the proponent of such an argument to provide some evidence
| for the extraordinary claim that return to on-prem is
| signifiant enough to influence server market share (or even
| establish that such revanchists would prefer AMD more than
| cloud buyers do).
| bluGill wrote:
| I doubt there will be a move to on-prem so much as places
| that already knew for their workload the cloud didn't
| make sense will be a little more careful and so not move.
| For many servers the cloud makes sense though, and so I
| think we will see places move more and more as they see
| the value of the cloud.
|
| For a large company with accountants they can do the
| accounting math, and a rack or two of servers is
| expensive - it need real estate, trained employees,
| managers for those employees, HVAC, power, backup
| servers, backup power, .... This adds up faster than on-
| prem advocates think. Many of those costs are things that
| work better in data center sized warehouse than a smaller
| company and so cloud can often be the a lower price once
| you really add up the costs. This is particularly true if
| you can share cloud resources in some way.
| marcosdumay wrote:
| > the retreat to on-prem
|
| On-prem only makes sense for really large organizations,
| exactly the ones that can negotiate with cloud providers.
|
| What I do expect to see eventually is a large movement
| towards commodity IaaS providers. And those tend to be
| middle-sized and not design their own hardware, so they
| act like the on-premises market. (But I can't say I'm
| seeing any movement here either.)
| keyringlight wrote:
| One thing I wonder about is whether it's the long term effects
| of things like spectre/meltdown coming home to roost. I can
| only speculate as I'm not in the server industry, but IIRC with
| the mitigations that reduced performance intel benefited a lot
| with companies buying more servers to make up the shortfall
| while AMD Zen was still proving itself to what seems like a
| conservative market. Has the "no one ever got fired for buying
| X" reputation shifted?
| treprinum wrote:
| Still wondering how AMD has only ~25% of desktop market share
| given recent Intel issues.
| bugbuddy wrote:
| Because Intel is still top dog when it comes to single threaded
| performance even against Zen 5.
| treprinum wrote:
| Yeah, but that's at the cost of a 50% probability of a CPU
| failure within a year. Moreover, it doesn't hold true in
| games vs X3D CPUs.
| bugbuddy wrote:
| That's hyperbole. There's no data to back such a claim.
| belval wrote:
| Well there are several reasons:
|
| - Intel is still competitive for gaming and productivity
| (albeit at a somewhat higher power consumption)
|
| - AMD does not have a good budget offering (competing with i3
| at ~120$)
|
| - People don't refresh their desktop as often as a laptop,
| especially for gaming where realistically an i7-7700k from 2017
| coupled with a modern GPU will comfortably output 60fps+ at
| 1440p.
| Drew_ wrote:
| > i7-7700k from 2017 coupled with a modern GPU will
| comfortably output 60fps+ at 1440p
|
| I can say by first hand this is not true for any modern MP
| game.
|
| In general, I hate these "it does X FPS at Y resolution"
| claims. They're all so reductive and usually provably false
| with at least a few examples.
| Zambyte wrote:
| On the contrary I can say first hand that it is true (3700X
| which is two years newer, but on benchamrks it is a toss up
| between the two). What modern GPU at you using?
| WarOnPrivacy wrote:
| I generally rule out all Intel CPUs between the 4770 and
| 11th Gen.
|
| The exception is when I need dirt-cheap + lower-ish power
| consumption (which isn't gaming)
| Ekaros wrote:
| An now 13th and 14th Gen... Before it is absolutely
| certain issues are fixed... So that leaves 12th?
| baq wrote:
| 12700k here, absolute unit of a CPU.
| Zambyte wrote:
| Your preference does not change whether or not it can
| achieve > 60fps at 1440p in modern multiplayer games when
| paired with a modern GPU
| Drew_ wrote:
| 3700X's 8-cores is a generational leap above the 7700k's
| quad core for modern games. They aren't comparable at
| all.
|
| I ran a GTX 1080 and then an RTX 3080. The performance
| was not very good in modern games designed for current
| gen consoles like pretty much any BR game for example.
| Some games got high FPS at times but with low minimum
| FPS.
| abhinavk wrote:
| A big chunk of gamers just play games like Valorant, CS,
| Fifa and CoD which usually run much better.
| Macha wrote:
| I would be shocked if a 7700K with a modern GPU does not
| get 60fps at 1440p in Rainbow Six, Rocket League, LoL,
| Dota, CS, Fortnite, etc.
| jajko wrote:
| I game on archaic i5-4460 (4 cores, max 3.2ghz), paired
| with rtx 2070 super and 16gb I can run literally
| everything new in 50-100 fps range, coupled with 34"
| 1440p VRR display not much more to desire for single
| player. Running this maybe 6 years with only graphic card
| and corresponding psu change.
|
| Ie Cyberpunk 2077 everything max apart from rtx stuff and
| its rarely below 80fps, dlss on good looks. Baldurs gate
| 3 slightly less but still smooth. God of war same.
| Usually first think I install is some HD texture pack, so
| far never any performance hit to speak of. Literally why
| upgrade?
|
| Consoles and their weak performance are effectively
| throttling PC games last few years (and decade before),
| not much reason to invest into beefy expensive power
| hungry noisy setup for most gaming, just to have few
| extra shiny surface reflections?
| Drew_ wrote:
| These are all old games so I wouldn't be surprised either
| replygirl wrote:
| you'd be amazed what runs at 60fps in 4k if you simply turn
| down the settings
| ineptech wrote:
| It so happens that toms recently tested this exact question
| (how much performance do you lose pairing a modern card
| with an older gpu, compared to the same card with a modern
| cpu). Full results at [0] but the short answer is that a
| $900 RTX 4080 gpu with a 2017 cpu will generally do 60+fps
| at 1440p in most games, but as low as 55 in a few.
|
| 0: https://www.tomshardware.com/pc-components/gpus/cpu-vs-
| gpu-u...
| Drew_ wrote:
| Not super surprising for single player games since
| they're usually much easier on the CPU than multiplayer.
| I was not getting minimum 60 FPS in Warzone for example.
| belval wrote:
| > not true for any modern MP game
|
| Pretty big claim
|
| > at least a few examples
|
| Reasonable claim
|
| Comfortably doesn't mean everything will be at 60fps, it
| means that most things will be at 60fps so someone with a
| 7700k will not be feeling pressure to change their CPU (the
| entire point of the thread).
| saltcured wrote:
| Aren't there Ryzen models as the budget offering that
| overlaps with i3?
| graton wrote:
| Prebuilt PC sales is likely that cause. I'm sure AMD has a much
| higher share for those who are building their computers. But
| that is most likely an overall small share of the market of
| computers sold.
| adrian_b wrote:
| The Amazon statistics show much higher sales of AMD CPUs than
| of Intel CPUs, so this supports your belief about DIY
| computers.
| jtriangle wrote:
| Most computers sold are prebuilts, and most computers sold
| are for commercial use, ie office bees, so video game
| oriented PC's are a relatively small corner of the market.
| jeffbee wrote:
| Intel sells a complete package with highly integrated laptop
| things such as wifi (CNVio2) and webcam (IPU6) for a very low
| price, and they have things that fleet buyers want like vPRO.
| treprinum wrote:
| The article had a separate "mobile" category for that where
| AMD had <20% market share.
| jeffbee wrote:
| While true, people still want things like Wi-Fi and vPro in
| a desktop. And USB4 and what not.
| pathless wrote:
| Intel has had very good marketing for decades, and they've also
| surrendered to marginless sales just for the sake of appearing
| dominant. AMD's strategy of late seems to be for higher
| margins, but lower share.
| sweca wrote:
| I just bought an AMD chip for my new server build for this
| reason. But then I found out about the Sinkclose
| vulnerability...
| WarOnPrivacy wrote:
| Before a Sinkclose package can be deployed, it needs some
| multi-stage exploits to be successful. I think it's value is
| against exclusive targets.
|
| I've had more than 0 clients in my career that might draw
| that kind of attention but not a lot more.
| Rinzler89 wrote:
| Intel N100 NUCs with 16GB RAM and 512GB SSDs can be had for
| less than $200 on Amazon. What does AMD have in that price
| point?
|
| NUCs with AMD APUs are nearly double the price. AMD is top dog
| in the high performing PCs but not everyone buys high
| performance PCs.
| tracker1 wrote:
| There aren't really any AMD options as cheap as the
| N100/95/97 mini-pc options. That said, there are some truly
| great options in the sub-$600 mini-pc segment. The Beelink
| SER8 with an 8845hs is pretty great to say the least. Bought
| one a few weeks ago for HTPC/Emulation usage running
| ChimeraOS, it's done a very good job. Though I might replace
| the BT/WiFi controller for something better supported. Only
| hung up once, but still annoying.
|
| I wouldn't consider it "high performance" but it's definitely
| good performance and far better than any Intel mini-pc
| options I've looked into.
| jauntywundrkind wrote:
| Nice to see Beelink has some bios updates available. Many
| of these mini-PC.folks don't seem to offer any updates!
|
| Here's a random sample of 1: https://dr.bee-
| link.cn/?dir=uploads%2FSER%2FBIOS%2FSER6%20MA...
| theropost wrote:
| This is true, though an AMD 5700U mini PC might go for $300,
| or less if you can find a bargain. However, it will hands
| down dominate a N100. 2.5 to 3x faster, not to mention the
| APU (vega 7) for video/gaming capabilities. I was always
| Intel, but AMD has really stepped up in the last few years.
| Their APU line up is top notch, and pretty incredible for the
| mini PC market (and laptops). The newer gen are great too,
| but for the best bargains, then Zen 3 lineup is the way to go
| MostlyStable wrote:
| My lab is trying to purchase a new office computer with heavy
| enough workloads to justify a mid-to-high tier CPU. I
| recommended that we buy an AMD machine primarily because of the
| current intel issues.
|
| The first problem I ran into is that the majority of companies
| selling pre-built computers do not offer an AMD option (and
| pre-built is a must. We are not allowed to use grant money to
| buy all the parts and assemble ourselves, which would have been
| my preferred option). And when they do, they are often limited
| in selection and might only be older generation or budget-tier
| CPUs.
|
| I did eventually manage to find an AMD option that fit our
| needs and my boss forwarded to billing for approval. We got
| significant push back with IT recommending various intel
| machines that they claimed were "better" (although also several
| hundred dollars more expensive), but were also available
| through the universities preferred sales partners.
|
| We could probably fight and get them to allow us to purchase
| what we want (the fact that they have to "allow" it when 100%
| of our funding is non-university dollars is a separate,
| ridiculous issue), but honestly, at this point we have probably
| already spent more staff time on this than the real world
| differences, even including the micro-code problems, are worth.
|
| I'd bet that a lot of corporate contexts, and even home
| enthusiast customers, operate on the same momentum: they have
| always bought intel and so they will continue to buy intel.
| Even in the absence of some strong belief that intel is
| "better".
| f1shy wrote:
| Do I remember correctly that intel was paying companies for
| not offering AMD? Like Dell?
| dijit wrote:
| You do. 15 years ago though.
|
| https://www.intel.com/pressroom/legal/docs/NY_AG_v._Intel_C
| O...
| Alupis wrote:
| The effects live on, unfortunately. Justice was not
| served as an outcome of that case.
|
| Today, people's mindshare is still largely with Intel,
| mostly due to nearly 2 decades of associated brand new
| computers, and performance with Intel, since they were
| the only offered option 90% of the time.
|
| Think about your typical non-technical person shopping
| for a computer. "I want Intel, they've always done well
| for me".
| pdimitar wrote:
| Don't be naive. This only means that they wisened up and
| made sure they will not be caught for 15 years now.
| vlovich123 wrote:
| Why not start your own company (even non profit) that offers
| prebuilt machines by buying the parts you want and assembling
| them?
| dijit wrote:
| Doing it for his current employer would likely open him up
| to liability regarding embezzlement.
|
| But it's not a bad idea, the margins for PC building is
| really low for the general public though.
| vlovich123 wrote:
| I don't see how a non-profit where the cost of the
| machine is the cost of the parts only could even
| realistically be embezzlement.
| ineptech wrote:
| It's not that it would be embezzlement, it's that it
| pattern matches to embezzlement. People with purchasing
| authority generally can't choose vendors they are
| involved with because that's such a popular way to steal
| money.
| bluGill wrote:
| That is a very common way embezzle money. Well not the PC
| business, generally you pick something that looks higher
| margin and/or harder to get into, but really anything
| will work so long as it looks like a legitimate business
| that does something real. The scam is easy: buy something
| cheap, and charge a lot of $$$ for labor or support
| costs. It is very easy to hide overhead that can be go
| back to you.
|
| You also typically are not directly involved in the
| business - that is both too much work and too obvious:
| instead you get your brother-in-law/uncle/cousin/... to
| start the business so it looks independent. Then the
| relative does expensive family vacations, football games
| and other such things that you happen to be invited on
| along with all your friends in common. (that is you need
| to make it look like money is never given to you, but
| because it is paid for you save a lot of money).
|
| The other common thing to do is a charity. If a
| politician's family is in any way involved with a charity
| you should assume it is a scam - the charity might
| otherwise do good work but the family member is there as
| a way to hide bribes. (again, the money never reaches the
| politician, but it reaches family members who the
| politician wants to do well). This of course makes life
| hard for honest family members of politicians -
| everything they do looks like the politician (not them!)
| asking for a bribe.
| renewiltord wrote:
| You need to go to an integrator and get an Epyc-based
| machine. You can do $7k ish per.
| etempleton wrote:
| I think there is a lot of trust of Intel. It is the only
| processor company most people know.
|
| For IT departments I think a lot of it is that they want you
| to have processors they know work with their software stack.
| In addition, Intel has their own software stack that works
| with their processors that include features such as Remote
| Desktop and security. Intel, like Microsoft, has become very
| good at catering to the Corporate IT crowd.
| pdimitar wrote:
| > _For IT departments I think a lot of it is that they want
| you to have processors they know work with their software
| stack_
|
| I thought the world solved this problem with compilers like
| 30 years ago.
|
| Also you're vastly overestimating the knowledge of the
| average computer users. Many of them can't even tell the
| difference between the CPU and GPU.
| aljgz wrote:
| Sorry, even you overestimated. A lot of users think CPU
| is the case.
| p_l wrote:
| It actually used to be a somewhat valid term, yes. The
| case with the computer was the Central Processing Units,
| and the rest was peripherals.
| minkles wrote:
| It's also that most IT people doing corp stuff don't even
| give a crap or even know what half the hardware is these
| days. I mean our IT department bought developers laptops
| with 8Gb of soldered in RAM and no expansion then bitched
| at the dev team when they were told to piss off
| pdimitar wrote:
| > _pre-built is a must. We are not allowed to use grant money
| to buy all the parts and assemble ourselves, which would have
| been my preferred option_
|
| Huh. Why? This seems arbitrary.
| metalliqaz wrote:
| because the bean counters track _assets_ , which means
| computers, not various parts.
| CoastalCoder wrote:
| I spent probably $20k on labor costs alone to order 4
| developer laptops for my team when I worked for the U.S.
| Navy.
|
| Procurement rules often exist for good reasons, but they're
| one of the reasons I moved back to the private sector.
| pdimitar wrote:
| Apologies for my ignorance, but can you describe it to me
| in practical terms what does it exactly mean to spend
| $20k on labor costs just to order 4 developer laptops,
| please?
| CoastalCoder wrote:
| Soliciting multiple bids, getting exceptions to normal
| computer procurement rules, navigating the other more
| mainstream rules, and some other stuff, IIRC.
|
| Happily it was a while ago and I purged most of the
| details from my brain.
| mandevil wrote:
| That's the number of hours it took to do all of the
| paperwork necessary for the acquisition, times the hourly
| rates charged to the government (as a general rule for
| USG cost-plus contracts, 1.5x the developer salary to
| cover benefits, contractor overhead, etc.) by the people
| doing the paperwork.
|
| The one I remember best from when I was working gov't
| contractors was that there were forms that certified we
| were currently complying with all current US trade
| embargoes and sanctions regimes. So we had to investigate
| and make sure that we were complying with them all- the
| State department has a searchable list of all the people
| and companies that are under various levels of sanctions.
| But that paperwork, times all of the other forms we had
| to fill out, times a competitive bidding process (with
| possibility for appeals) to make sure the government
| wasn't getting ripped off, etc.
|
| This can happen in the private industry as well: I
| remember once seeing a whole bunch of level 2 managers in
| a conference room battling it out over exactly which
| laptop to buy for ~15 developers who needed special extra
| beefy hardware for a project, and commenting to my L1
| manager that the price difference between the two laptops
| was less than the cost to the company of that meeting, so
| it would have been wiser to just buy the more expensive
| laptops and forget the argument.
| CJefferson wrote:
| Because if you assemble yourself, and it doesn't turn on,
| it's your job to debug which part is failing. If you buy
| pre-build you get a warranty for the whole thing.
|
| We had some people at work buy parts and it was fine for a
| while, then became a nightmare, unstable machines where we
| could never figure out what part was to blame. It isn't
| worth paying Dev's salaries to let them debug their faulty
| hardware.
| bluGill wrote:
| How much time is spent choosing and assembling parts? The
| less time you spend on tasks like that the more valuable
| time spent on something useful. (or so it is assumed)
| jpeloquin wrote:
| It often is arbitrary. Also, the rules as enforced by the
| awardee institution often differ from the funding agency's
| written rules. The people who are responsible for reviewing
| purchasing are only responsible for making sure no rules
| are broken, not with producing anything, so they tend to
| interpret the rules conservatively and just say "no" if
| they're not sure. The IT department also may try to block
| anything they didn't pick even if the funding agency would
| allow it. (Speaking from experience with university
| bureaucracy.)
|
| It's not out of concern for cost-benefit ratio (overhead
| from parts selection and assembly) or anything like that.
| Getting something useful done is solely the researcher's
| problem.
| acchow wrote:
| > the majority of companies selling pre-built computers do
| not offer an AMD option
|
| Don't you just need to find 3-4 sellers to do a shopping
| comparison? Don't need anything close to the "majority"
| diabllicseagull wrote:
| I've had a similar experience a few years ago. At the time,
| enthusiast Ryzen CPUs had just started outperforming Xeons
| that Dell and others were exclusively pushing. A sub thousand
| box edging out a ten thousand dollar workstation wasn't
| enough to move the needle until Threadripper workstations
| started to show up. Even then I wasn't able to push our IT to
| consider AMD as a serious replacement. Maybe that's partially
| why we still see ten Intel options for every AMD one.
| dagw wrote:
| It's really hard to buy pre-built high end desktop computers
| from big brands with the latest AMD CPUs.
| Der_Einzige wrote:
| To this day, Intel is still deeply ahead in software support
| for AI.
|
| MKL, OpenAPI, Intel Optimization for sklearn, for pytorch, etc
| (there are many more). No equivilants exist from AMD but NVIDIA
| has an equivalent for all of these (i.e. CuPy).
|
| Using AMD for AI stuff only makes sense if you never believe
| that you may want to do any kind of ML-like workload on or with
| the CPUs assistance. This is true for many folks though, but
| for those it's not true for, they will buy blue until Intel no
| longer exists.
| treprinum wrote:
| MKL runs on Ryzens just fine and newest Ryzens all have
| AVX512 which all new desktop/laptop Intels lack. "AI" on CPU
| is a gimmick, everybody does it on GPU anyway (outside 128GB
| RAM M3 Max that has some LLM use, but it's still due to
| shared memory with GPU).
| Der_Einzige wrote:
| Certain workloads aren't total "gimmicks" (i.e. small
| embeddings models), and often those CPU optimizations are
| on things that absolutely do take advantage of the CPU well
| (i.e. graph analysis) or on traditional ML algorithms like
| random forests (intel optimizations for sklearn) which are
| still important and do still run well on CPUs.
|
| Not to mention that most people have more RAM than they do
| VRAM.
|
| Running MKL on your datacenter grade AMD processor
| technically violates some license and it relies tricks like
| this - https://danieldk.eu/Posts/2020-08-31-MKL-Zen
|
| Technically, if your found to be allowing this as a cloud
| provider by Intel, they have grounds to sue you, so yes,
| _you_ can use MKL on a Ryzen at home, but you put yourself
| at risk for lawsuits by doing this in a data center at
| scale
| 0xDEADFED5 wrote:
| I'm doing fast AVX512 embeddings on Ryzen, and fast ONNX
| AVX512 reranking on Ryzen. Though I do the actual heavy
| lifting on GPU, doing all the RAG stuff in CPU is helpful.
| AI on CPU is still _mostly_ a gimmick, but as models get
| smaller and more capable it 's becoming less of a gimmick.
| WarOnPrivacy wrote:
| > Still wondering how AMD has only ~25% of desktop market share
| given recent Intel issues.
|
| I sometimes pick Intel over AMD because I don't have AMD's CPU
| naming conventions memorized - and I'm in a hurry.
| dukeyukey wrote:
| Hah, I felt that urge last time I bought a PC - I could
| easily roughly compared Intel performance vs price, but with
| AMD I had to look up half the models.
| WarOnPrivacy wrote:
| > I could easily roughly compared Intel performance vs
| price, but with AMD I had to look up half the models.
|
| That's sort of what I'm up against.
|
| AMD makes excellent products. If I have a day or two I can
| ask my son for recommendations. He knows AMD lines off the
| top of his head. If I have 10 minutes, I go with what I
| know.
| jhickok wrote:
| Like you don't have time to research the CPUs for a personal
| machine? Man I usually go overboard reading up on the
| hardware going into my own desktop.
| WarOnPrivacy wrote:
| > Like you don't have time to research the CPUs for a
| personal machine?
|
| No. It's when I have to quickly line up hardware for my
| customers.
|
| I'll have 5 projects in play and I get an email from a
| client saying they need a workstation or notebook for a new
| hire ASAP. I can skim Intel CPUs among Lenovo offerings
| fairly quickly.
|
| Back when I was a system builder, I knew every capability
| of every chip. When Intel had hobbling virtualization
| (VT/x, etc), I knew which chips were capable. In that day,
| I often chose AMD to save time because every AMD CPU was
| fully capable.
| jtriangle wrote:
| We changed our current order for workstations to AMD parts.
| They arrive today or tomorrow I think.
|
| Odds are it would have been fine, they were just i3 worker bee
| boxes, but, it's not really worth the risk when there's a
| viable alternative.
| mdasen wrote:
| Honestly, this is one of the reasons I think Intel's stock is
| likely undervalued at this point (not that I want to invest my
| money with management I don't feel confident in).
|
| Intel is facing a lot of threats and they rested on their
| laurels. At the same time, Intel is still 75-80% of the market
| - even as AMD started killing Intel in performance for several
| years. I'm less sure that Intel will hit its fab milestones
| than Intel's press releases, but I'm also not sure that AMD can
| really take advantage of that. AMD is reliant on TSMC's
| capacity and as we've seen recently, AMD often has to wait a
| year or two before it can use TSMC's latest processes. Even
| AMD's upcoming Zen 5 processors will be using TSMC's 4nm N4P
| process (with future Zen 5 processors expected to be 3nm). If
| Intel is able to get 18A out the door in 2025, Intel's
| fabrication will certainly be competitive with what AMD can get
| (and likely quite a bit better).
|
| From what I've read (and I might be wrong), it sounds like
| Intel is moving to High NA EUV soon (now-ish/2025) while TSMC
| is waiting until around 2030
| (https://www.tomshardware.com/tech-
| industry/manufacturing/tsm...). Is TSMC falling into the trap
| of letting a competitor out-invest them - the same trap that
| Intel fell into ~2012-2022? Probably not, but it does seem
| likely that the advantage TSMC had will taper away over the
| next couple years - assuming Intel isn't full of crap about
| their progress.
|
| Even if TSMC maintains a lead against Intel, can AMD take
| marketshare? Specifically, can AMD produce as many chips as
| demanded or are they somewhat limited by how much capacity they
| can get from TSMC? For example, if people want X number of AMD
| server chips, does AMD manufacture 25% of that number and 75%
| of customers buy Intel chips because that's what is available
| due to AMD's lack of fab capacity? I'm not saying that is the
| case - it's a genuine question for someone that knows more
| about the industry than I do.
|
| We do know that TSMC has limited capacity. Apple tied up TSMC's
| 3nm capacity such Apple is the only company shipping 3nm parts.
| It looks like Intel will be the next company shipping 3nm parts
| with their Lunar Lake using TSMC's N3B process this fall.
|
| I'd also note that Intel's fabs aren't unlimited either. Meteor
| Lake moved to Intel 4 in December 2023, but Emerald Rapids
| (server) remained on Intel 7 at the same time. Intel can't just
| churn out as many Intel 4 chips as they'd like.
|
| I too wonder how/why AMD is still below 25% marketshare given
| Intel's issues. I wonder if AMD has limitations on how much fab
| capacity it can get which is stifling attempts at gaining
| marketshare. At the same time, it looks like Intel is on a
| course to fix its issues - if one can believe what Intel is
| saying. If Intel is able to ship 18A chips in 2025 at volume,
| it seems like it should fix the big issue that Intel has been
| facing - inferior fabs. AMD has been able to out-do Intel in
| part because it's been able to rely on TSMC's better
| fabrication. Is this just a blip where Intel looks bad
| 2020-2024? That's a genuine question because I don't really
| know enough about the industry. Maybe someone will say that
| Intel's PR is mostly bluster and that they're still being run
| by beancounters trying to optimize for short-term profits.
| Maybe someone will say that Intel is investing in the right
| direction, but there's still a lot of risk around pulling off
| High NA EUV for 2025. Maybe someone will say that Intel is
| likely to hit their milestones and will regain the title of
| best-fab.
|
| However, given that AMD hasn't been able to really hammer
| Intel's marketshare, it feels like Intel has more staying power
| than I'd think. Again, like the parent comment, it's always
| puzzled me that AMD hasn't gained a ton of marketshare over the
| past 5 years.
| treprinum wrote:
| All these points about Intel are just waiting for an answer
| to "Is Intel capable of changing its prevailing culture or
| not?". There is a reason why Intel fell behind and it wasn't
| due to a lack of engineering geniuses inside. You might bet
| on the stock market and see. AMD obviously has a SPOF in TSMC
| and its limited capacity.
| mdasen wrote:
| Yep, that's a lot of the essence. I would point out a few
| things, though.
|
| Unlike many other cases of a company getting a bad culture,
| Intel hasn't seen its marketshare destroyed. They still
| have time to turn things around. Even if things go in a
| very bad direction, they could become a fabless company
| like AMD and compete for TSMC's capacity on equal terms
| with AMD.
|
| If you look at a company like Yahoo, Google had already
| become dominant when it was trying to turn itself around.
| AMD hasn't even hit a third of the market.
|
| Not only that, but it's probable that AMD isn't capable of
| becoming even 50% of the market due to capacity
| constraints. If AMD can make 25-30% of the market's
| processors and no more due to capacity constraints, even if
| Intel processors are inferior, they're still going to be
| the bulk of sales. By contrast, Google had basically
| infinite capacity to dominate search, Facebook had
| basically infinite capacity to crowd out MySpace and other
| rivals, etc. The point is that Intel has a lot more runway
| to turn itself around if it's essentially guaranteed 70% of
| the market.
|
| In some ways, this resembles the Boeing/Airbus situation.
| While Boeing has been in a bad place recently, Airbus can't
| take advantage of it. That gives Boeing a long time to
| change its trajectory and is probably why Airbus is only
| worth around 15% more than Boeing. Even if airlines want to
| buy Airbus planes, they can't. Likewise, even if OEMs and
| datacenters want to buy AMD chips, maybe they can't because
| AMD doesn't have the capacity to make them.
|
| In some cases, you need to rapidly change the culture and
| direction of a company because you're quickly dying. In
| Intel's case, it seems likely that they could waddle along
| for a long time if AMD isn't able to quickly grab
| marketshare. I think it's easier to gradually change the
| culture over a longer period of time if you have the
| ability to stave off collapse in the meantime.
|
| For example, maybe we'll look back on Intel in a decade and
| say "yea, the change started around 2020 when Intel
| realized it had fallen behind. It continued to fall behind
| while it tried to correct things, but then turned a corner
| and in 2025 they'd regained the fab crown." Are we already
| close to that culture change showing up in the public view?
|
| I just think I don't know enough about the industry to
| really know. Maybe someone would say "Intel has already
| done the culture change and we've seen that in things like
| Intel 4 and you're really going to notice it in 2025 and
| 2026 with Intel beating TSMC to High NA EUV." Maybe someone
| would say "No, intel hasn't changed its culture and you can
| see that Intel 4 is a niche where they can't even make
| enough volume to put out a server product with it." Or
| maybe Intel is focused on the more medium-term with 18A and
| not wanting to waste resources on a less-than-stellar
| interim process.
|
| I guess my question isn't just whether Intel is capable of
| changing its culture, but whether that's already happened.
| The public perception is a lagging indicator. Most of the
| public didn't have a bad perception of Boeing 1996-2017. We
| only recognized the problems in hindsight. Likewise, we
| recognized Intel's shortcomings in hindsight. We won't
| recognize whether Intel has changed for years after it has
| already changed.
|
| So, has Intel's culture already changed (and we're going to
| see some pretty awesome stuff from them in 2025 and 2026)
| or has Intel been on the same path to mediocrity for the
| past 4 years (and they've just gotten better at press
| releases announcing that they'll be better soon)?
| Wytwwww wrote:
| > maybe they can't because AMD doesn't have the capacity
| to make them.
|
| Or Intel just cut prices on their server CPU so that they
| are sort of competitive despite much higher power usage
| to core ratio?
| acchow wrote:
| If the Generative AI growth continues, then physical real
| estate and power usage will be top priorities.
|
| The data center shortage means hyperscalers want as much
| performance as possible per server (and ideally the best
| performance/watt because cooling is also a concern). This
| why AMD has double the ASP of Intel in the server market.
| high_na_euv wrote:
| Aint high na euv for 14A?
| qwytw wrote:
| Isn't Meteor Lake made by TSMC (at least some parts like the
| GPU)? Which would explain the capacity constraints.
| bangaladore wrote:
| My next CPU will be AMD. Why I'm not upgrading yet:
|
| 1. I generally use the same CPU for 2-3 years. I have a 11700k
| right now.
|
| 2. AMD hasn't released the beefier versions of its new CPU.
| High-end consumer Intel chips have many more cores/threads (P +
| E), while AMD's top right now has 16 threads.
|
| #2 Is why I'm holding off for now. I'm waiting for the X3D
| version of their chip with ~2x the core count and more threads.
| mrweasel wrote:
| Depending on you definition of recent, because it's to recent
| to really make any significant impact.
|
| If you're thinking recent as in the last few years, then
| because the performance and power issues with Intel aren't
| large enough to make any difference for regular desktop use.
| Most will be on a 3 - 5 year upgrade cycle, so your new office
| PC just needs to be better than a five year old one, which it
| will be.
|
| What it might do it damage the used market. Prices on Intel
| based refurbished PCs needs to drop, by a lot, now that we know
| that many/most of those CPUs are damaged beyond repair and will
| continue to degrade.
| xboxnolifes wrote:
| Because, like you said, the issues are recent.
| deelowe wrote:
| Because AMD's general purpose offerings are pretty poor in
| comparison to intel (e.g. per per watt, productivity, etc). AMD
| only really shines in high end applications like gaming or high
| core count cloud computing.
| JasonSage wrote:
| I think it's going to change a lot in coming years, but it's
| early days.
|
| I feel like the writing has been on the wall for Intel's
| downward trajectory and AMD's substantial improvements for a
| few years now, but I think a lot of trust and brand loyalty has
| papered over the signs for Intel enthusiasts.
|
| On a hardware news site covering the recent news, I saw for the
| first time in Intel buyers... shock. Incredulity.
| Disappointment. It's the first time I've seen self-admitting
| Intel fans come out in numbers questioning their beliefs and
| perceptions.
|
| I think in the next 4-year cycle as home PC builds turn over
| and Intel buyers are coming back to market, there's going to be
| a large influx of AMD converts. The Intel disaster lately will
| have turned erosion into an exodus. Maybe not a monumental one.
| But the AMD numbers are going to grow seriously over the next
| half decade.
| nickdothutton wrote:
| I would be amazed if Intel has completely abandoned the kind of
| business practices they are famous for, and were fined (circa
| $1.5 billion) for WRT systems builders/retailers.
| wmf wrote:
| It usually takes 5-10 years for the evidence to come out.
| Whatever anti-competitive practices they're using now don't
| seem to be working very well.
| albertopv wrote:
| Soon migrating part of DC to Oracle Cloud (don't ask...), a
| consultancy firm will do the job, all VM will be on AMD CPUs.
| cletus wrote:
| It's hard to believe it's been roughly two decades since the
| Athlon/Opteron almost killed Intel, which would've been the last
| time AMD did so well in server market share.
|
| The short version of this story is that Intel licensed the x86
| instruction set back in the 90s to several companies including
| AMD and Cyrix. Intel didn't like this as time went on. First,
| they couldn't trademark numbers, which is why the 486 went to the
| Pentium. Second, they didn't want developers producing compatible
| chips.
|
| So Intel entered into a demonic pact with HP to develop EPIC.
| That's the architecture name. Itanium was the cip. Merced was one
| of the early code names. This was in the 90s when it wasn't clear
| if RISC or CISC would dominate. As we now know, this effort was
| years last, with huge cost overruns and by the time it shipped it
| was too expensive for too little performance.
|
| At the same time, on the consumer front we had the Megahertz
| Wars. Intel moved from the Pentium-3 to the Pentium-4 that scaled
| really well with clock speed but wasn't great with IPC. It also
| had issues with pipelines and failed branch prediction (IIRC).
| But from a marketing perspective it killed AMD (and Cyrix).
|
| Why is this important? Because 64 bit was around the corner and
| Intel wanted to move the market to EPIC. AMD said to hell with
| that and released the x84-64 instruction set (which, by the terms
| of the licensing agreemnt, Intel had a right to use as well) and
| released the Athlon series of desktop chips, followed later by
| the Opteron server chips. These were wildly successful.
|
| The Pentium-4 hit a clock ceiling of 3-4 GHz, which still pretty
| much exists today. In the 90s it was thought chips would scale up
| to 10GHz or beyond by many.
|
| What saved Intel? The Pentium-3. You see the Pentium-3 had
| morphed into a mobile platform because it was very energy
| efficient. First as Pentium-M and later as Core Duo and Core 2
| Duo. This was the Centrino platform. Some early hackers took
| Centrino boards and built desktops. The parts were hard to get if
| you weren't a laptop OEM. They probably salvaged laptops.
|
| Anyway by the mid to late 2000s, Intel had fully embraced this
| and it became the Core architecture that has evolved largely ever
| since to what we have today.
|
| But back then Intel was formidable in terms of bringing new
| smaller processes online. This was a core competency right up
| until the 10nm transition in the 2010s, which was years late. And
| TSMC (and even Samsung) came along and ate their lunch. I can't
| tell you that happened but Intel never recovered, to this day.
|
| As for AMD, after a few years they never seemed to capitalize on
| their Opteron head start. Maybe it was that Intel caught up. I'm
| not really sure. But they were in the wilderness for probably
| 10-15 years, right up until Ryzen.
|
| Intel needs to be studied for how badly they dropped the bag.
| Nowadays, their CEO seems to be reduced to quoting the Bible on
| Twitter [1].
|
| I'm glad to see AMD back. I still believe ARM is going to be a
| huge player in the coming decade.
|
| [1]: https://www.indiatoday.in/technology/news/story/intel-
| facing...
| reginald78 wrote:
| IIRC Intel never wanted to license x86. IBM required it so
| they'd have a second supplier. And Intel has been trying to
| correct that ever since.
| bluGill wrote:
| > What saved Intel? The Pentium-3.
|
| The Pentium-pro which become the Pentium-2, then Pentium-3
| then... Of course each change in name come with some
| interesting new features, but the Pentium-pro was where that
| linage started.
| lvl155 wrote:
| At this point, I am asking why AMD doesn't buy Intel. Sure,
| antitrust police will be all over it but it would be an epyc
| irony for AMD.
| foobarian wrote:
| Wait a little more and maybe we won't need to add more choices to
| AWS amd64 or arm64! :-)
| kens wrote:
| The article states that Intel has 75.9% of datacenter CPU
| shipments and AMD has 24.1%. This implies that ARM has 0% of the
| server market, which is not the case. I suspect this article
| neglected to state the important restriction to "*x86* server
| market", which makes a big difference to the conclusions.
| FuriouslyAdrift wrote:
| By unit quantities, yes... by revenue, however, they are nearly
| neck in neck
|
| "While Intel earned $3.0 billion selling 75.9% of data center
| CPUs (in terms of units), AMD earned $2.8 billion selling 24.1%
| of server CPUs (in terms of units), which signals that the
| average selling price of an AMD EPYC is considerably higher
| than the ASP of an Intel Xeon."
| Sammi wrote:
| If companies are really spending twice the money on amd chips
| as these numbers imply, then this should mean that a amd
| server chip has a 2x perceived value compared to an intel
| one.
| nicoburns wrote:
| I think there are AMD chips with twice as many cores as the
| Intel ones, so this kinda makes sense.
| Sammi wrote:
| At this point I avoid Intel for the same reason I avoid Boeing.
| GeekyBear wrote:
| I'm happy to see an AMD with plenty of income that they can
| invest in the development of future products. We've seen how
| little advancement we get when Intel doesn't have effective
| competition to spur them along.
| AtlasBarfed wrote:
| I'm no Intel fan but the last time that AMD had market
| leadership and anything which was under Hector Ruiz who is they
| basically completely stopped innovating. I don't think that'll
| happen under Lisa, but the track record of AMD and market
| leadership is they just kind of sit on it. We'll see what
| happens this time around
|
| Of course, last time Intel was still sitting on a metric pile
| of Fab technology and other stuff that was still in the
| pipeline. Intel appears a lot more starred for such potential
| innovations.
|
| The path forward appears to be radical ISA switches or
| something besides yet another x86 node shrink. But Intel's
| track record for doing anything outside of x86 processors is
| very very very very poor.
| marcosdumay wrote:
| > The path forward appears to be radical ISA switches
|
| Why up to this day people still can't make x86 clones? The
| amd64 architecture is nearly 20 years old, and lots of
| compilers target ancient processors.
|
| What kind of legal protection is there that is lasting for
| that long?
|
| (Anyway, I'm glad RISC-V is taking off.)
| mlyle wrote:
| * SSE and other architectural extensions combined with
| supporting patents
|
| * Intellectual property minefields about how to adapt every
| new micro-architectural innovation to the mess that is x86
| architecture.
|
| Finally, barriers to entry in developing a competitive
| processor are fundamentally high. Combining them with the
| risk of litigation makes it untenable.
| more_corn wrote:
| Sometimes all you gotta do to win is stay on your feet when your
| opponent stumbles.
___________________________________________________________________
(page generated 2024-08-12 23:01 UTC)