[HN Gopher] Scaleway launches RISC-V servers
       ___________________________________________________________________
        
       Scaleway launches RISC-V servers
        
       Author : enz
       Score  : 70 points
       Date   : 2024-03-03 11:01 UTC (11 hours ago)
        
 (HTM) web link (labs.scaleway.com)
 (TXT) w3m dump (labs.scaleway.com)
        
       | MaxBarraclough wrote:
       | Interesting step, their claim of _world 's first RISC-V servers
       | available in the cloud_ is true as far as I know.
       | 
       | I see they list 3 officially supported GNU/Linux distros: Debian,
       | Ubuntu, Alpine. I wonder how mature these are on RISC-V at this
       | point and whether they're ready for production server usage.
        
         | brucehoult wrote:
         | Much more ready for server usage than desktop usage.
         | 
         | Linux is Linux. It's hard to go wrong when you're not mucking
         | about with graphics cards and first-person games and things
         | like that.
        
           | MaxBarraclough wrote:
           | For server workloads there can still be plenty of non-trivial
           | architecture-specific dependencies besides the core OS
           | itself, such as the JVM and other JIT-based interpreters.
           | 
           | Python 'pip' packages, which often have native code
           | dependencies, may also lack official RISC-V builds.
        
             | brucehoult wrote:
             | Access to servers such as these, for pennies per hour, will
             | accelerate authors of those packages porting them to
             | RISC-V.
             | 
             | It's a lot lower barrier to entry than buying the
             | equivalent board outright for $180.
             | 
             | You can also just do `docker run -it riscv64/ubuntu` on
             | your x86 or Mac, at zero incremental cost.
        
         | yjftsjthsd-h wrote:
         | Nitpick: Alpine is a Linux distro, but not a GNU/Linux distro;
         | it uses musl libc and busybox coreutils.
        
           | williadc wrote:
           | This is the first time I've seen the correction go the other
           | way.
        
           | MaxBarraclough wrote:
           | Interesting, thanks.
        
           | arcanemachiner wrote:
           | Or as I've recently taken to calling it, BusyBox plus Linux.
        
         | cpswan wrote:
         | Debian support for RISC-V is presently only in Sid (testing
         | release). So not production ready.
         | 
         | When I ask people what it will take to get RISC-V into stable
         | releases the answer is always 'better access to test hardware'.
         | Hopefully this is a step along the way.
        
       | my123 wrote:
       | TH1520 with the XuanTie C910 core. Avoid. That particular
       | platform is really incredibly slow, Cortex-A5x level. And with
       | RVV 0.7.1 instead of 1.0 too.
        
         | nevi-me wrote:
         | I was going to try the server out until I saw this comment.
         | Which alternative board/computer would you recommend one buy?
         | 
         | I'm learning RISC-V assembly with a friend, it'd be great for
         | us to have a machine to tinker with (we'll use ESP-32-Cx for
         | 32-bit).
        
           | b9b10eb736 wrote:
           | Still way better than a VisionFive 2, and more than enough to
           | get your hands in the thing (especially with the hourly
           | pricing). You have to acknowledge that the RISC-V ecosystem
           | isn't as mature as that of ARM64 and AMD64. You don't get the
           | best performance/price ratio with RISC-V (yet). Many
           | extensions were ratified so recently that hardware doesn't
           | even exist (yet).
           | 
           | As far as I'm concerned, the LicheePi 4A from Sipeed which is
           | also equipped with a TH1520 offers a decent value for a
           | RISC-V SBC. If you want more performance than that, it's
           | going to cost you a lot more.
        
             | brucehoult wrote:
             | Right now, you can't get more performance than the TH1520
             | (C910 cores) and JH7110 (U74 cores) at ANY price. The only
             | thing you can do is get more of the same cores i.e. SG2042
             | with 64x C910 cores.
             | 
             | The same goes for emulation. QEMU on my i9-13900HX laptop
             | is about the same speed on each core, but there I have 24
             | cores (32 threads).
             | 
             | I just compiled gcc 14 natively on both LicheePi 4A (the
             | same as this hosting) and docker/QEMU:
             | 
             | i9-13900HX (8 P + 16 E cores) docker running riscv64/ubuntu
             | image                   real    101m44.472s         user
             | 1695m26.395s         sys     24m36.671s
             | 
             | LicheePi 4A (TH1520 SoC, 4x C910 cores)
             | real    422m41.367s         user    1430m56.638s
             | sys     70m17.994s
             | 
             | The LicheePi actually used less CPU time, but the i9 won
             | with more cores.
        
               | b9b10eb736 wrote:
               | Thanks for the clarification, you're completely right. I
               | didn't realize the SG2042 used the same core. So,
               | Scaleway's offer really does seem to make sense!
        
               | camel-cdr wrote:
               | Arguably OpenXiangShan should be more powerful, but
               | verilog simulation isn't really that usable in practice,
               | and it has no working vector support yet. (the dev branch
               | currently hangs after a few minutes of simulating)
        
         | drmpeg wrote:
         | On the RISC-V IRC channel, it was shown that there was some bug
         | with the Milk-V Pioneer processor, which is a C920 core. I'm
         | not sure if the C910 has the same issue. The super slow
         | benchmark was:
         | 
         | stress-ng --verbose --metrics --aggressive --atomic 64
         | --timeout 600
         | 
         | If you try this, adjust the number (64) to match the number of
         | cores on your system.
        
           | my123 wrote:
           | It's not just about a given microbenchmark - that platform
           | just sucks tbh. Hopefully something better appears.
        
             | drmpeg wrote:
             | Real world performance reported: glibc build on Unmatched
             | ~7:30, on Pioneer ~16:20 hours.
        
         | brucehoult wrote:
         | > Avoid. That particular platform is really incredibly slow,
         | Cortex-A5x level.
         | 
         | If you avoid it then you won't be using RISC-V at all, right
         | now.
         | 
         | Along with the JH7110, it is the fastest RISC-V CPU you can buy
         | off the shelf today. That will probably change late in this
         | year, but for now you can't do significantly better.
         | 
         | Overall the two (and the SG2042, same cores as the TH1520 but
         | 64 of them, plus L3 cache) are very similar in speed. The C910
         | can be 30% to 50% faster on some microbenchmarks. The JH7110 is
         | usually about 10% faster on system level real world tasks e.g.
         | building software. Not enough to notice unless you sit both
         | side by side.
         | 
         | > And with RVV 0.7.1 instead of 1.0 too
         | 
         | Which matters much less than it used to, as gcc 14 can compile
         | C code with RVV intrinsics to either.
         | 
         | Just today I took someone's RVV 1.0 test code, which they'd
         | only been able to run on simulators, and compiled it and ran it
         | on my TH1520 LicheePi 4A, with only Makefile changes
         | (`-march=rv64gc_xtheadvector` instead of `-march=rv64gcv`).
         | 
         | https://www.reddit.com/r/RISCV/comments/1b57gib/comment/kt4t...
        
           | b9b10eb736 wrote:
           | Fully agreed.
           | 
           | > The JH7110 is usually about 10% faster on system level real
           | world tasks e.g. building software.
           | 
           | Pretty sure the U74 CPU from the JH7110 is way worse than the
           | C910 from the TH1520 on pretty much all aspects. So my guess
           | is that your metric is mostly explained by the fact that the
           | JH7110 has a PCIe bus which allows plugging in an SSD rather
           | than an eMMC or SD Card. But such SSD also has a cost. I
           | think this gives some perspectives.
        
       | Kab1r wrote:
       | eMMC for cloud seems to me like a dangerous proposition. As far
       | as I know, they do not hold up well to write intensive workloads.
       | I suppose you will need to use networked logging.
        
       | tux3 wrote:
       | Scaleway was also early in providing Arm servers, it's cool to
       | see them keep bringing new stuff
       | 
       | It's maybe almost _too_ early, given the performance profile of
       | these and the current state of software compat, but the
       | trajectory is definitely there for RISC-V. Wouldn 't be surprised
       | to see these compete strongly against ARM servers a few years
       | from now.
        
         | crq-yml wrote:
         | This one is probably more for Scaleway to gain experience than
         | for customers to use seriously.
         | 
         | Of course, that excuse could be true for customers too.
        
           | brucehoult wrote:
           | The performance is about the same as the first Graviton
           | servers in late 2018.
        
         | Hamuko wrote:
         | Do they still offer ARM? I remember them killing the ARM bare
         | metal back in the day.
        
           | tempest_ wrote:
           | Still seem to offer ampere based vcpus
           | 
           | https://labs.scaleway.com/en/amp2-instances/
        
             | c0balt wrote:
             | They are discontinued as of the end of January 2024. (See
             | the FAQ section). Their only ARM appears to be Cost
             | Optimized Arm: https://www.scaleway.com/en/cost-optimized-
             | instances-based-o...
        
       | freeqaz wrote:
       | I couldn't find the value, but how much are they charging per
       | month for these boxes? I see they have 4 CPU cores and 16gb of
       | RAM.
        
         | tyingq wrote:
         | Name CPU RAM Storage Ethernet Network Price excl. VAT/hour
         | Price excl. VAT/month       EM-RV1-C4M16S128-A 4xC910 RISC-V
         | 64GCV 1.85GHZ 16 GB 128 GB 100 Mb/s 0,042EUR 15,99EUR
        
         | mikmak wrote:
         | just click the link => 15.99EUR per month or 0.042EUR per hour
        
       | theanonymousone wrote:
       | Does Docker work with RISC-V?
        
         | cpswan wrote:
         | Yes. I've been running it on a VisionFive 2 and a BeagleV-
         | Ahead.
        
         | pella wrote:
         | https://hub.docker.com/u/riscv64
         | 
         | minimal pre-build images ...
        
         | brucehoult wrote:
         | It absolutely does. I use it every day.
         | 
         | You can run your RISC-V images transparently on your x86 or Arm
         | (e.g. Apple) machines, as docker desktop automatically uses
         | QEMU. Or run them natively in docker on your RISC-V board.
         | 
         | Minimal starter image: riscv64/ubuntu
        
         | b9b10eb736 wrote:
         | I works great (with the usual requirement of having the
         | expected kernel options/modules enabled). The main issue right
         | now is the lack of image support for RISC-V. I hope this is
         | going to change if hardware starts becoming accessible.
        
       | fodkodrasz wrote:
       | 100 Mbit/s NIC? Seems a bit thin.
        
         | mistrial9 wrote:
         | great for code, terrible for lots of data. maybe not trendy but
         | neither is it fatal?
        
         | jacoblambda wrote:
         | For 4 cores and 16GB of RAM? not really. That's probably close
         | to saturation for the hardware for most workloads other than
         | raw data streaming.
         | 
         | The real sell of this hardware is that it's doing this with
         | about 5W of power draw at peak utilisation.
        
       | znpy wrote:
       | Doe these cpus have any cryptographic hardware acceleration? They
       | list gpus and npus only.
       | 
       | I have an old atom-based kimsufi by ovh, cryptography will eat
       | almost a full core when doing file transfer over ssh or any kind
       | of https...
       | 
       | No cryptographic hardware acceleration could really be the
       | Achilles' tendon for this platform.
        
         | Findecanor wrote:
         | The T-Head C910: no. It is considered a bit of an old core by
         | now. It is open source so it has been adopted a lot.
         | 
         | The RISC-V standard has extensions for both scalar and vector
         | cryptography instructions but yet only a few cores out there
         | support one or the other. Look out for CPUs based on the SiFive
         | P670 which should have the latter. When Android smartphones
         | with RISC-V show up, they will also have vector cryptography.
        
           | brucehoult wrote:
           | > T-Head C910: no. It is considered a bit of an old core by
           | now
           | 
           | Announced in July 2019, it is one of the most recent RISC-V
           | cores you can buy. Hardware takes time, usually 3-4 years to
           | go from announcement of a core to an SBC you can buy.
           | 
           | The C910 is the only RISC-V core you can buy in an SoC with
           | 64 of them (Milk-V Pioneer)
           | 
           | The only newer core you can buy is the C908, available only
           | in the K230 SoC on the CanMV-K230 board. It had some
           | advantages (it's the only board rn with Vector 1.0), but it
           | also has the HUGE disadvantages of being single core and only
           | 0.5 GB RAM, vs quad core and 16 GB RAM on this cloud
           | offering.
           | 
           | > Look out for CPUs based on the SiFive P670
           | 
           | This is not available. The first boards are expected late
           | this year.
           | 
           | It will be a huge advance, certainly, with twice the speed
           | per core, and the first SoC SG2380 with it having 16 cores.
           | 
           | But not today.
        
       | camel-cdr wrote:
       | I'll reiterate what I wrote before:
       | 
       | 042EUR/hour per hour of compute is a great deal for developers.
       | If you tests with cross compilation and qemu anyways and
       | sometimes needs test/benchmark on the native hardware, then this
       | means you can spend about 5EUR and you are probably set for
       | months if not a year.
       | 
       | Also, with the latest gcc you can finally target rvv 0.7.1, which
       | is supported by these CPUs. You just write your standardized rvv
       | 1.0 intrinsics and if you add `-march=64gcxtheadvector` gcc 14
       | will just generate the equivalent rvv 0.7.1:
       | https://godbolt.org/z/va9sfEnMW
        
         | justahuman74 wrote:
         | We probably want to encourage people to forget pre 1.0 vectors
         | exists, to ease ecosystem fragmentation
        
       ___________________________________________________________________
       (page generated 2024-03-03 23:01 UTC)