SiFive Is Launching The Most Compelling RISC-V Development Board Yet

[rene]

There's a post on Reddit & video of one of these boards with an RX480 that suggests our open source graphics drivers are running on it.

https://www.reddit.com/r/Amd/comments/jkh98q/radeon_gpu_linux_kernel_drivers_are_apparently/

I didn't know that either

yep, the beauty of the open source AMD drivers others and I reoccuringly speak for in contrast to the binary only just novidio forces their fanboys down the throat, ...

[PerformanceExpert]

The 2 1/2 year old 1.5 GHz HiFive Unleashed is already faster than a Pi 3.

If this has come in at 2 GHz as anticipated (they don't seem to have said) then it should be similar performance to an Odroid C4 (quad 2.0 GHz A55) or Pi 4 (lower MHz but higher IPC).

That's not remotely possible, the Unleashed is a simple 1-issue core while Cortex-A53 is 2-way - this shows the Unleased is about halfway between a Pi 2 and Pi 3 in performance.

The new version is 2-way in-order so should get close to A53 on integer, but behind on floating point and SIMD. There is no chance it will get near the performance of a Pi 4 since that uses the 3-way out of order Cortex-A72 which is significantly faster.

[bridgman]

Couldn’t amd license them an old GCN gpu ip for cheap, limited cu.
why don’t you release new sub 75w gpu?

Unfortunately because of iGPU progress the market for small low power dGPUs is in that awkward "really small but not zero" spot and looks to stay there for quite a while, so it's hard to justify a lot of ongoing investment. We already make Polaris-based cards (RX 540 and RX 550 aka Lexa) that run without a power connector near laptop power levels.

That market is already at the point where supply chain challenges are dominating, so I doubt that anyone would want to license an existing design and take on that pain when it's easier to just buy the chips in volume for a good price.

I do miss the good old days when humans set prices instead of computers and older stock gradually had its price reduced to move it out rather than having scarcity-based algorithms charging $700 for what used to be a sub-$100 card in the hope that someone would buy it.

[willmore]

Is there an "arduino-like" board with a Risc-V microcontroller?

https://longan.sipeed.com/en/ maybe?

[ayumu]

Is there an "arduino-like" board with a Risc-V microcontroller?

There's quite a few. Some of them are dirt cheap.

Try a "RISC-V" search on aliexpress or the like.

The GD32V is particularly interesting. It's a RISC-V CPU core with STM32-like peripherals.

[stiiixy]

Unfortunately because of iGPU progress the market for small low power dGPUs is in that awkward "really small but not zero" spot and looks to stay there for quite a while, so it's hard to justify a lot of ongoing investment. We already make Polaris-based cards (RX 540 and RX 550 aka Lexa) that run without a power connector near laptop power levels.

That market is already at the point where supply chain challenges are dominating, so I doubt that anyone would want to license an existing design and take on that pain when it's easier to just buy the chips in volume for a good price.

I do miss the good old days when humans set prices instead of computers and older stock gradually had its price reduced to move it out rather than having scarcity-based algorithms charging $700 for what used to be a sub-$100 card in the hope that someone would buy it.

Well, that algorithm approach to pricing explains my reasoning for not buying older gen hardware from the usual suspects. I thought I was going nut's, or the world was!

Thank you very much for explaining. I alsoiss thoser ended xard's, but as you say, the 540/550 will probably have to bridge the gap for the time being.

[esbeeb]

Not only would I call this a "developer board", but I would call it a "low-level developer board." I think every build maintainer in the Debian project should be furnished with one of these, somehow or other (even if it's SiFive themselves paying for it, out of pocket, calculating that this expenditure will be a huge help to boosting their own ecosystem). Not to mention anyone in the linux world, who has lower-level, Video-related expertise, maybe mainlining some kernel drivers for commodity, easily-available PCIe video cards, etc.

IMHO, this board is especially suited for getting a bunch of new bugs discovered, and fixed, having to do with various low-level toolchains being used in Debian package building, as well as lower-level libraries, and getting anything related to video working: X.org or Wayland support, and maybe even Vulkan.

https://wiki.debian.org/RISC-V#Toolc...reaming_status

My prediction is that the next board SiFive makes down the road, having more power and cheaper cost (as in, decent desktop-grade performance on par at least with a Raspberry Pi 4), will more realistically get into the hands of a much larger number of Debian Developers who debug/maintain packages that are much higher level (GUI apps, and desktop environments), as opposed to the lower-level toolchains/libraries, and Video-related infrastructure.

[vying1]

Are there any realistic MIPS options for running Linux?

The fastest is a Loongson 3A4000. 4x 2.0ghz MIPS64EL. You can buy Micro ATX boards on Taobao with PCI-E and DDR4 etc.

[brucehoult]

That's not remotely possible, the Unleashed is a simple 1-issue core while Cortex-A53 is 2-way - this shows the Unleased is about halfway between a Pi 2 and Pi 3 in performance.

The new version is 2-way in-order so should get close to A53 on integer, but behind on floating point and SIMD. There is no chance it will get near the performance of a Pi 4 since that uses the 3-way out of order Cortex-A72 which is significantly faster.

I have both a Pi 3 and an Unleashed and I've run a lot of the same code on both of them. Do know how they perform, both on my code and in general feel.

Your link shows five benchmarks. The Unleashed performs about like the Pi 2 on the two encryption tests. Unsurprising as it doesn't have specialized instructions for that as it's a minor part of most people's workloads. On two other benchmarks the Unleashed is the outright winner, and on the 5th it is close to the Pi 3.

I think this demonstrates that my claim is close enough to reality.

You don't seem to be aware that a dual-issue CPU doesn't execute two instructions in the same clock cycle anywhere near to 100% of the time. The A53 (and A7) are fairly crude in this, and probably don't even achieve 40%. The Unleashed also has much more and much faster RAM than the Pi 3.

The A55 and the U74 have much better dual issue implementations with some minor out-of-order features -- technically: they have both an "early ALU" pipe stage and a "late ALU" pipe stage, which allows them to dispatch two dependent instructions in the same clock cycle in many cases. Yes, the A72 is 3-way but again this is peak and it doesn't achieve that all the time -- the average is probably around 2, at best.

There are plenty of benchmarks showing 2.0 GHz A55 (e.g. Odroid C4) performing very close to 1.5 GHz A72.

[PerformanceExpert]

Your link shows five benchmarks. The Unleashed performs about like the Pi 2 on the two encryption tests. Unsurprising as it doesn't have specialized instructions for that as it's a minor part of most people's workloads. On two other benchmarks the Unleashed is the outright winner, and on the 5th it is close to the Pi 3.

Neither OpenSSL nor Blowfish are accelerated using special instructions. Before you declare victory, those results are for a Pi 3 at 1.2GHz rather than 1.4GHz and use 32-bit Arm code. AArch64 is significantly faster (eg. SSL score is 3 times better), so the gap with Pi 3 is even larger than the results indicated.

You don't seem to be aware that a dual-issue CPU doesn't execute two instructions in the same clock cycle anywhere near to 100% of the time. The A53 (and A7) are fairly crude in this, and probably don't even achieve 40%. The Unleashed also has much more and much faster RAM than the Pi 3.

My point was that 2-way beats 1-way at the same frequency, not that 2-way implies a 2x speedup (which it doesn't of course). However, from experience I know Cortex-A53 has pretty good dual-issue.

The A55 and the U74 have much better dual issue implementations with some minor out-of-order features -- technically: they have both an "early ALU" pipe stage and a "late ALU" pipe stage, which allows them to dispatch two dependent instructions in the same clock cycle in many cases. Yes, the A72 is 3-way but again this is peak and it doesn't achieve that all the time -- the average is probably around 2, at best.

Cortex-A55 supports dual issue of load+store (which U74 doesn't do), has low latency private L2 cache (which U74 doesn't), has 2 FP pipelines (U74 has 1) and SIMD (U74 doesn't). So comparing U74 to A55 is a gross exaggeration of the capabilities of U74 - it's more similar to Cortex-A7 with Neon removed.

There are plenty of benchmarks showing 2.0 GHz A55 (e.g. Odroid C4) performing very close to 1.5 GHz A72.

Well my link above shows that the Pi 4 is more than twice as fast as Pi 3 using AArch64. Cortex-A55 is ~20% faster per clock than A53, so there is still a huge gap between A55 and A72.

I'd be surprised if the Unmatched can beat a Pi 3, but Pi 4 is simply out of its league. Maybe the U84 will do better but there hasn't been any news for 12 months. It seems to me RISC-V people are better at generating hype than creating actual designs that run fast...