[HN Gopher] Alibaba cloud FPGA: the $200 Kintex UltraScale+
___________________________________________________________________
Alibaba cloud FPGA: the $200 Kintex UltraScale+
Author : signa11
Score : 235 points
Date : 2025-10-04 06:49 UTC (16 hours ago)
(HTM) web link (essenceia.github.io)
(TXT) w3m dump (essenceia.github.io)
| mati365 wrote:
| What's special about this?
| hmry wrote:
| These chips usually cost $1k-$2k when bought new individually
| from the manufacturer
| h33t-l4x0r wrote:
| Are we talking about physical products or cloud services?
| oefrha wrote:
| There's zero cloud service in the article, only physical
| boards decommissioned by a hyperscaler. Asking about cloud
| services tells me you only read the title.
| h33t-l4x0r wrote:
| I read the title and was confused why the article did not
| match it.
| debugnik wrote:
| Cloud should have been title-cased (they make this
| mistake almost every time in the article), they bought an
| FPGA decommissioned by Alibaba Cloud, the subsidiary.
| RossBencina wrote:
| Not only that, but you can develop for this particular model
| with the no-cost Vivado license.
| stephen_g wrote:
| Yeah Xilinx's pricing curve is incredibly steep. Probably
| makes sense since they're used in things like defence (radar,
| electronic warfare) so they can sell them in low quantities
| for huge prices and people will still buy them.
|
| But with decent volume they get pretty cheap - for example
| there are video converters they have FPGAs in them that you
| can buy for less than you can buy the FPGA part in single qty
| - which means the manufacturer of the device is getting the
| FPGA for 5-10% of list price.
| hmry wrote:
| Wanted to buy a GOWIN FPGA recently, like on Sipeed's dev
| boards. Mouser wanted $37 for it, more than twice what the
| whole devkit costs.
|
| Meanwhile the list price from the manufacturer is $3, but
| it's completely impossible to order anything as an
| individual. Not sure why they designed a hobbyist beginner-
| friendly dev board around a part from a company that makes
| all their products impossible to buy...
| phendrenad2 wrote:
| So people are going to desolder the chip? This doesn't seem
| very hobbyist-doable. Are there people out there who build
| products based on scrapped FPGAs from ebay?
| duskwuff wrote:
| For what it's worth: if you have a FT2232H adapter hanging around
| (and if you don't, you should consider getting one), those
| adapters can easily be flashed to be compatible with Vivado:
|
| https://docs.amd.com/r/en-US/ug908-vivado-programming-debugg...
| NoiseBert69 wrote:
| At work we always have 20-30x FT2232H on stock.
|
| They are super useful - they speak GPIO, I2C, SPI, parallel
| FIFO, etc.
|
| And their pyFTDI Python lib is great.
| Joel_Mckay wrote:
| Just be careful as the counterfeit FTDI chips can have bad
| internal regulators etc. (Usually those small Red UART and
| GPIO breakout hobby PCB.)
|
| Usually better to use the vendor supplied JTAG tools, or
| build your own with known good silicon. Recovering FPGA
| hardware is always a mixed bag of what IP and documentation
| is available to bring up the board. Looks like nice silicon
| though. =3
| NoiseBert69 wrote:
| Not all FTDI series are faked. Only the most popular FT232
| as far as I know.
|
| Same for ST. Only the STM32F103 is a well known bad actor
| in the market.
| bloggie wrote:
| In my experience the f103 clones fix silicon bugs and
| some even have improved peripherals (like i2c)
| NoiseBert69 wrote:
| Problem is that they still using the ST HAL which is not
| aware of the fixes in the background.
|
| 90% of the standard stuff will work. But as soon as you
| dig into the exotic modes bad things will happen.
| duskwuff wrote:
| > Only the most popular FT232 as far as I know.
|
| Specifically, the FT232RL - a simple USB-to-serial chip.
| (And many of the clones are actually quite high quality.)
| As far as I'm aware, the FT2232H - a more complex USB to
| various-serial-and-parallel-protocols chip - hasn't had a
| counterfeiting problem, although I believe WCH has made
| some compatible parts.
|
| The STM32F103(C8) situation is pretty funny too. Most of
| the companies making clones have long since moved on from
| outright counterfeits and have started manufacturing
| their own more-or-less-compatible XYZ32F103
| microcontrollers (e.g. GD32F103, AT32F103, APM32F103,
| CH32F103, HK32F103, etc).
| afandian wrote:
| I've always been curious why there's no generic USB
| serial chip. It seems like the most obvious thing in the
| world. I've noticed ESP32 boards tend to come with one of
| two proprietary chips, with their own drivers. Do you
| know why this isn't all standard?
| taneq wrote:
| Also beware that FTDI has bricked swathes of gear that
| (often inadvertently) used counterfeit chips, via driver
| updates. So if you buy gear that uses an FTDI chip then if
| you're unlucky want it turns out to be counterfeit, your
| thing might get bricked. :/
| matt-p wrote:
| Good, FR. Then you can send it back to who you bought it
| from. All vendors should do this.
| taneq wrote:
| The issue is (as I understand it) that some early supply
| chain distributors either unwittingly bought shipments of
| counterfeit parts, or otherwise ended up stocking them as
| genuine. Then legitimate hardware manufacturers used
| them. And let's be upfront here, any part using FTDI USB
| interfaces is low volume hardware that's either indie or,
| like, niche scientific gear. These are not chips you use
| when you have a 20-person hardware procurement department
| with the time to do a deep dive into every shipment of
| every component. These are chips you use when you're used
| to serial ports and have had USB foisted upon you and
| just want to sell the thing you built.
|
| Sending it back only punishes the OEM you bought the
| weird niche thing off, and they likely had no indication
| that the chips they paid genuine prices for were
| counterfeit. If they try to get a refund from Alibaba or
| DigiKey or whoever, they'll politely be told to fuck off,
| if they get a response at all.
|
| I totally understand why FTDI chose to take this path,
| but I also think it was a poor choice overall and I now
| avoid FTDI products (which back in the day I happily
| used, before all this) just in case I get caught up in
| the BS.
| matt-p wrote:
| Cisco use these chips inside their switches, they are all
| over the place. I still think this is good, I buy them
| from RS, any authorised distributor will not tell you to
| 'fuck off', if they do please forward it to FTDI. If
| you're buying off Alibaba or eBay then that's on you.
| Sorry.
| ThrowawayR2 wrote:
| Penalizing FTDI for the OEM's negligence in not properly
| securing their supply chain? That's absurd. It is 100%
| the responsibility of the OEM to vet their suppliers to
| ensure that they know what's going into their product.
| poizan42 wrote:
| Importantly not just counterfeit chips but also
| completely legal clone chips. For context, for a long
| time Windows did not have a driver for the CDC serial
| communication class out of the box. But FTDI had made
| good Windows drivers, and their interface was easy to
| clone. So for a time there was a market for making 3rd-
| party chips implementing FTDI's interface and copying
| their USB id to be recognized by their driver[0]. Of
| course some unscrupulous sellers would inevitably try to
| pass those of as genuine FTDI chips. FTDI was probably
| not happy with either of them, with the clone chips
| manufacturer piggybacking of their hard work. But they
| could only do anything about the counterfeits, so they
| kinda ignored the existence of legal clones and in all
| their public communication only talked about taking steps
| to shut down counterfeit chips (most likely well knowing
| that their were also destroying chips of legitimate
| competitors).
|
| [0] Clearly allowed at least in the US under the decision
| in Sega v. Accolade
| Joel_Mckay wrote:
| Many Chip VID:PID listed in
|
| > lsusb
|
| or
|
| C:\> Get-PnpDevice -PresentOnly | Where-Object {
| $_.InstanceId -match '^USB' } | Format-List
|
| are considered copyrighted IP in North America. Many of
| the paid device codes are often not like other
| Communication Device Class drivers or free generic HID
| devices listed in the USB standard.
|
| Early next year Windows 11 will be locking out unsigned
| drivers from the OS. Good for keeping the kernel more
| consistent, but could be bad for hobby hardware drivers.
|
| Personally, I think it is ludicrous, but for 83% of users
| they have no other option than pay $6.32 for a $$0.43
| chip. =3
| matt-p wrote:
| The copycats are free to write thier own drivers though?
| They could even open source them. Why should they get to
| steal them from FDTI?
|
| I do agree the cost of the chips are insane however,
| they're ultimately worth paying for because of the driver
| support and 'just works' factor.
| Joel_Mckay wrote:
| > copycats are free to write thier own drivers though?
|
| Sure, anyone paying for a Microsoft certified hardware
| driver, USB PID allocation, and cert fees every year...
| Everyone else is not allowed in the OS we paid for...
|
| It is a decades old serial port emulation chip...
| Racketeering with a computer should still be illegal. =3
| poizan42 wrote:
| > are considered copyrighted IP in North America.
|
| Yes they may be, but in the Sega v. Accolade case I
| mentioned earlier it was decided that copying of both
| copyrighted and trademarked material for interoperability
| purposes is fair use. In that case Accolade had to
| literally embed "SEGA" inside their ROM and send the text
| to the Genesis III in order for it to boot their games.
| The Ninth Circuit court ruled that this constituted fair
| use.
| CamperBob2 wrote:
| _Early next year Windows 11 will be locking out unsigned
| drivers from the OS._
|
| That's been the case since Windows 7 for 64-bit drivers,
| hasn't it? What's changing with 11?
| hxorr wrote:
| The issue was, if I recall correctly, that the Microsoft
| CDC driver was quite buggy. This led to the situation
| where CDC device vendors were unwilling to produce CDC
| compatible devices, because the default Microsoft driver
| would be used and, of course, the blame for any bugginess
| gets pinned on the hardware vendor.
|
| Hence the current situation of many CDC-type devices that
| are intentionally incompatible enough with the standard,
| in order to require the device vendor's own drivers which
| they can ensure meet their quality expectations
| Joel_Mckay wrote:
| In general, the Microchip CDC driver was free to use.
|
| "The CDC class has been implemented in Windows (since
| Windows 98), macOS and most Linux distributions. Since
| Windows 10, no extra information has to be given in the
| start-up sequence as the operating system now has a
| generic driver that will be used for CDC. To support
| older versions of Windows (Windows 7 and earlier),
| provide an .inf file that associates with the correct
| driver."
|
| However, these will not support the high-speed bit-banged
| JTAG mode FTDI chips can perform with the right driver.
| =3
| bschwindHN wrote:
| I _do_ happen to have one laying around from another project,
| weirdly enough. I've never done FPGA work though, when you say
| it's compatible with Vivado, what does that mean in practice? I
| can configure an AMD FPGA, or something more?
| ddlsmurf wrote:
| I think they mean it can be flashed into a programer/jtag
| interface compatible with vivado. If you want to get started
| with fpga, may I recommend 1bitsquared.com 's products
| robinsonb5 wrote:
| Vivado offers some embedded debugging and logic analyzer
| infrastructure, which allows you to tap into signals within
| the design and monitor them from the PC. (For instance, the
| Memory Interface Generator will report its initialisation
| status to the PC and if something failed it will tell you
| what). This is done over JTAG, and Vivado will only support
| that when using a select few JTAG dongles.
|
| If you're happy to create your own debugging infrastructure
| (which isn't that hard - almost all FPGAs have at least two
| dedicated IR scancodes specifically for user applications)
| then you can use any JTAG dongle supported by OpenOCD.
| jevinskie wrote:
| Vivado, along with expensive adaptors, supports some JTAG
| dongles that are just FT2232 with an EEPROM that contains USB
| descriptors that Vivado recognizes and treats as a (some low
| cost onboard adaptor for Digilent boards, IIRC) supported
| adapter. https://gist.github.com/rikka0w0/24b58b54473227502fa
| 0334bbe7...
|
| If you like FT2232, you'll love the Tigard board that takes a
| FT2232H and adds level shifters and is basically as tricked
| out as you can make a FTDI adapter.
| https://github.com/tigard-tools/tigard
| NoiseBert69 wrote:
| Cool stuff.
|
| Reminds me of recycling used Mining- or LED-Controllers as they
| can be bought on Ali and have quite capable FPGAs.
| matesz wrote:
| Reminds me how Alibaba innovated in database space with a cluster
| of FPGAs doing LSM compaction with custom MySQL storage engine -
| https://dl.acm.org/doi/pdf/10.1145/3299869.3314041
|
| Basically they achieved massive improvements in throughput and
| latencies vs RocksDB, for some operations even an order of
| magnitude.
|
| They even offered this as a service, but eventually pulled out.
| Not sure if they are still using it, but I was surprised that
| pretty much nobody else, as far as I am aware, is using any
| hardware acceleration for repetitive database workloads.
| donavanm wrote:
| AWS was using in house hardware query acceleration in redshift
| a few years back https://aws.amazon.com/blogs/aws/new-aqua-
| advanced-query-acc.... I dont know how much attention its
| getting (or performance its bringing) since AWS started
| focusing more in P&L the past year or two.
| foft wrote:
| If you are interested in FPGA PCIe or PCI cards there are also a
| lot of Gidel boards available second hand. ProcSpark/ProcStar
| etc. The official software is proprietary and they often contain
| multiple FPGA devices so there is again an exercise working out
| the pinouts etc if you want to simply use them with Quartus. I
| got one with a massive Stratix IV on.
|
| A Kintel UltraScale+ is quite a prize though, really nice write
| up.
| jevinskie wrote:
| The top-level Verilog module and Vivado .xdc file (contains pin
| mappings, timing constraints, etc) from Gidel for the HawkEye
| 20G-48. [0] No SDK from Gidel though.
|
| My back burner project for them is to create a PCIe TLP
| sniffer/MiTM/device emulator by hooking up two together via 10
| GbE for relaying TLPs with one of the remaining 10 GbE
| connection going to a host PC for the sniffed/injected TLPs.
| The Aria 10 FPGA PCIe hard IP allows for either root or
| endpoint mode so I "just" need to draw the rest of the owl,
| avoiding any Quartus IP modules that would make the setup non-
| transparent.
|
| I'm not sure what using a 10 Gbit link for PCIe will be like
| with faster devices but fail0verflow got away with TLP proxying
| with 115200 baud UART. [1]
|
| [0]: https://www.ebay.com/itm/335904285904 [1]:
| https://fail0verflow.com/media/33c3-slides/#/11
| haunter wrote:
| >The board also came with a travel case
|
| But why? Are these boards frequently removed and physically moved
| around in data centers?
|
| Or was that an eBay seller thing to protect it?
| a012 wrote:
| > But why?
|
| To bring them to the field, because they're _field_ programable
| gate array
|
| /s
| brookst wrote:
| That's why the trowel, I suppose.
| random_duck wrote:
| Lol, nice one.
| rkagerer wrote:
| Lots still available on eBay
| https://www.ebay.com/itm/167626831054
|
| I guess there aren't any GPIO's wired up to headers or such? Any
| suggestions on a project for it that's a good fit?
| lnsru wrote:
| That's really nice! Cheap hardware was gone during covid years
| completely. And now beautiful Kintex for 200$ again. I was
| super interested in these when I was younger. 2x SFP with PCIe
| reminds me on NetFPGA project. This card is perfect for it!!
| djoldman wrote:
| """
|
| dmesg log :
|
| [ 0.388790] pci 0000:00:00.0: [14e4:2712] type 01 class 0x060400
|
| [ 0.388817] pci 0000:00:00.0: PME# supported from D0 D3hot
|
| [ 0.389752] pci 0000:00:00.0: bridge configuration invalid ([bus
| 00-00]), reconfiguring
|
| [ 0.495733] brcm-pcie 1000110000.pcie: link up, 5.0 GT/s PCIe x1
| (!SSC)
|
| [ 0.495759] pci 0000:01:00.0: [dabc:1017] type 00 class 0x020000
|
| ...
|
| Since most people might not be intimately as familiar with PCIe
| terminology, allow me to quickly document what is going on here.
|
| 0000:00:00.0: is the identifier ...
|
| [14e4:2712]: is the device's [vendor id:device id], these vendor
| id identifiers are assigned by the PCI standard body to hardware
| vendors. Vendors are then free to define there own vendor id's.
|
| The full list of official vendor id's and released device id can
| be found : https://admin.pci-ids.ucw.cz/read/PC/14e4 or in the
| linux kernel code...
|
| """
|
| Wouldn't it be really nice if dmesg had more info about devices?
| random_duck wrote:
| Agreed, it would be so nice. But I can also understand they
| might want to keep dmesg as learn as possible.
| synergy20 wrote:
| just received a kv260 board, i thought vivado is free to use?
| brandonpelfrey wrote:
| You can typically get a free license for some boards,
| especially evaluation boards.
| 15155 wrote:
| Everything is free if you continually renew your trial every 30
| days.
| sweetjuly wrote:
| Vivado has free and paid tiers. The free tiers ("WebPack")
| support their a myriad of their smaller devices (which includes
| the Kria boards). The larger devices (Virtex, etc.), however,
| are generally only supported by the paid versions of Vivado.
|
| It's inconvenient for hobbyists, sure, but for enterprise uses
| the cost of Vivado for a team is largely inconsequential (which
| I suspect is why they get away with this).
| duskwuff wrote:
| The licensing scheme is not terribly robust, either; there
| are a number of cracked (full) licenses floating around if
| you know where to look. Given that AMD makes most of their
| money on hardware sales, and the software is only really
| useful in conjunction with that hardware, I suspect they
| don't care very much.
| HarHarVeryFunny wrote:
| Is anyone aware of any examples of FPGAs used to implement neural
| nets or other connectionist architectures?
|
| What's exciting about FPGAs is that they make "custom" chips
| affordable for colleges and even individuals... At this point in
| time, with the AI world fixated on LLMs, it seems that there's an
| opportunity for small groups/individuals to use FPGAs to
| experiment with more animal-like ground-up approaches to
| intelligence.
| le-mark wrote:
| Why the downvotes? My understanding ai is not a use for fpgas
| because models require a lot of ram (preferably fast ram) and
| fpga themselves don't have a lot of it. An fpga will have pins
| you can attach fast ram, but then board layout and layering
| traces becomes very non trivial. Hence gfx cards (and a lot of
| them) are more suited.
| throwaway48476 wrote:
| FPGA will never be as fast as an ASIC. NV data center chips
| cant do graphics anymore, just compute.
| HarHarVeryFunny wrote:
| Sure - not as fast or flexible, but relatively cheap and
| reprogrammable - good for experimentation, maybe for
| building a real-world proof-of-concept of a learning
| architecture ?
| random_duck wrote:
| Funny you say that, if you look at the git repo for the
| blog there are a bunch of asic screenshots currently being
| committed.
| HarHarVeryFunny wrote:
| FPGA certainly not ideal for an experimental AI architecture,
| but if data is just flowing thru it (sensory input in, motor
| control out) then I don't see RAM as an issue. Could be an
| interesting option for experimenting on a mobile robot.
| Aurornis wrote:
| > Is anyone aware of any examples of FPGAs used to implement
| neural nets or other connectionist architectures?
|
| Do you mean as a specialized processor for high speed
| inference, or as if the neural net including weights was
| directly synthesized into the FPGA architecture?
|
| Neither use case would be as powerful as other options. FPGAs
| don't provide infinite leeway for implementing completely
| custom logic. They have a finite number of blocks that can be
| reconfigured in certain ways as synthesized by the software.
| This architecture isn't a good fit for the large LLMs we think
| about which need a lot of high bandwidth memory access. You can
| connect an FPGA to high speed memory, but the current crop of
| GPUs are going to be much better at the job.
| nickpsecurity wrote:
| https://arxiv.org/abs/2404.10076
| Unaimend wrote:
| Have a look into differentiable logic gate networks
| bschup wrote:
| Certainly, we use networks on FPGA at FNAL for inference on
| LHC/CMS high energy physics results. There are also cases
| involving microsecond control of dynamical systems.
| https://fastmachinelearning.org/hls4ml/
| https://indico.cern.ch/event/797510/contributions/3313676/at...
| https://cdrdv2-public.intel.com/833277/deploying-tinyml-fpga...
| https://ml4physicalsciences.github.io/2019/files/NeurIPS_ML4...
| brookst wrote:
| Serious competence porn. I love it.
| jhallenworld wrote:
| I just used the Raspberry-PI V to bring up the PCIe interface on
| a Lattice Certus-Pro NX "Versa" board, it's very convenient.
|
| What's cool is that the Raspberry-PI V is just fast enough to run
| modern desktop software, including (ack!) Microsoft Teams. I was
| able to give a live demo of my FPGA design by sharing the desktop
| during a conference call.
|
| My only complaint is, of course, lack of SoC documentation from
| Broadcom.
|
| Say what you will about Intel, but they do provide full chip
| documentation if you go through their NDA process. With it, I
| made a very nice FPGA PCIe environment on Xeon Ivy Bridge: it let
| me save all of the PCI config registers, then reconfigure the
| FPGA, then reprogram them and have the chip show up on the bus
| without having to reboot the server or re-enumerate. The trick is
| to temporarily set the root complex bit to disable PCIe error
| detection during the reconfiguration process.
|
| (Was using Altera Stratix-IIgx at the time).
|
| There is possibly another way to this, but haven't tried it:
|
| https://stackoverflow.com/questions/32334870/how-to-do-a-tru...
|
| But even aside from that, with the documentation you have access
| to a lot of error reporting which is extremely useful during PCIe
| bring-up.
| akamoonknight wrote:
| Ah, that's a fancy trick! I remember many a time making Linux
| real angry with my PCIe fiddling. I believe in the end we got
| Partial Reconfiguration set up, where the PCIe portion of the
| device would not be reprogrammed but we could reload the rest.
| It was definitely a trade off though in terms of layout
| planning and space reservation.
| rfoo wrote:
| That's cool. For those who happened to be in China right now,
| there's the same listing on a Chinese marketplace (idlefish) for
| only 480 CNY (~68 USD). That's... better than too good to be
| true. I'm going to have a try.
| mk_stjames wrote:
| I don't see if they found out the pinouts for the SFP connectors
| on the back. If those were documented, and the PCI-e lanes were
| documented, I could see someone having some real fun with these
| building some bespoke fiber networking gear.
|
| SFP transceivers are just that- transceivers, they will spit out
| what you put in, so you could even build some interesting bespoke
| fiber optic signaling gear with these, and a PCIe backplane, or
| just solo.
| random_duck wrote:
| The SFP connector pinnouts are in the pinnout section.
| johndoe0815 wrote:
| Is there any RAM on the board except for the on-chip block RAMs?
|
| The ebay listing mentions 128 MB, but I wonder if this can be
| trusted since some other information in the "item specifics"
| overview is clearly bogus ("7-Function Digital Multimeter
| Shield").
| epolanski wrote:
| Not gonna lie.
|
| This ai craze makes me think that:
|
| - there's not much of a moat long term in inference hardware
| (can't comment about training)
|
| - the models themselves are a commodity that can be swapped any
| day
|
| - the products are coming but not at the pace one thought. It's
| been 3 years since Chatgpt launched and we still have no decent
| service where I can inject documents and ask questions.
| YZF wrote:
| What do you mean by inject documents and ask questions? Doesn't
| Gemini let you do that, e.g. NotebookLM or Gem? (not sure
| what's the equivalent products from the other vendors)
| epolanski wrote:
| Not a great experience still, doesn't scale, not precise
| enough, still looks to be rag based, so you get the usual
| problems rag can't solve due to chunking, etc.
___________________________________________________________________
(page generated 2025-10-04 23:00 UTC)