4 ms·
New RISC-V microprocessor can run CPU, GPU, and NPU workloads simultaneously
- joha4270 3y agoThis article looks like an ad too me. The article and the linked original source[1] is somewhat sparse on technical details, but if I divine it correctly, what they're making is essentially a GPU, with its compute units being RISC-V cores instead of some proprietary ISA. I predict this will be about as successful, as when Intel tried it a decade ago with the Xeon Phi. But lets see [1]: https://www.x-silicon.com/news/ https://www.x-silicon.com/news/
- sylware 3y agoI am always pleased to see new RISC-V based implementations. But here, people have to keep in mind: GPU is all about performance. Namely, having the GPU handling some CPU load must in no way impact significantly GPU performance.
- proc0 3y agoIsn't the distinction about parallelism? CPU has more built-in logic while GPU is about running the same computations in parallel.
- brucehoult 3y agoCPUs with length-agnostic vector ISAs -- so vector register size is an implementation detail and code for 128 bit vector registers runs unchanged (but efficiently) on 2048 bit registers -- will very soon be competitive with GPUs on GPGPU tasks and better on things needing close coupling between vector and scalar. They might even be quite good as actual GPUs, with instructions for things such as texture interpolation added.
- sylware 3y agoHow do you get the real vector register size in order to give it as a parameter to your vectorized algorithms?
- ksec 3y ago>Almost a year before Elon Musk made his company naming easy for people who can’t write or spell, a group of engineers, architects, programmers, and biz dev folks from AMD, Qualcomm, Intel, ATI Technologies, Dell, and others were working Forget about why would Dell matters here for a minute, but ATI Technologies....... I mean I do miss the good old days of S3 Verge, 3Dfx Voodoo and ATI All in Wonder or Rage 3D. But they were acquired by AMD in 2006, that is coming close to 20 years.
- kjs3 3y agoThey aren't the only 'ATI Technologies' that could be involved. For example, https://www.ati.org/ https://www.ati.org/. I dunno if that's the case here, tho. Yeah...Voodoo was awesome.
- BirAdam 3y agoI wrote about 3dfx. Crazy thing is, it was the same people who were at SGI and then Pellucid. Additionally, ATI and Nvidia can trace their technology back to SGI… fun times. https://www.abortretry.fail/p/so-powerful-its-kind-of-ridiculous https://www.abortretry.fail/p/so-powerful-its-kind-of-ridicu...
- westurner 3y agohttps://news.ycombinator.com/item?id=39518293 https://news.ycombinator.com/item?id=39518293 : > [PQ] SSL/TLS Accelerators and ML Accelerators / AI Accelerators An opportunity for a new RISC-V arch, perhaps. > With some newer architectures, the GPU(s) are directly connected to [HBM] RAM; which somewhat eliminates the CPU and Bus performance bottlenecks that ASICs and FPGAs are used to accelerate beyond. The HBM3E Wikipedia article says 1.2TB/s. Latest PCIe 7 x16 says 512 GB/s: https://news.ycombinator.com/item?id=39933959 https://news.ycombinator.com/item?id=39933959