Const-me

48 ms·

571.
by Const-me 3y ago
nVidia has a 20 GB GPU with the same chip as 4070Ti, the model is RTX 4000 SFF. One issue is price, it costs almost twice as much. Another one is memory bandwidth, RTX 4000 SFF only delivers 320 GB/second. That is much slower than 4070
572.
by Const-me 3y ago
ROCm is not the only option, compute shaders are very reliable on all GPUs. And thanks to the Valve’s work on DXVK 2.0, modern Linux runs Windows D3D11 software just fine. Here’s an example https://github.com/Const-me/W
573.
by Const-me 3y ago
> attempt or willingness to try to solve the problem first on their own… indication of where the problem comes from, why it might be interesting None of that is required to ask questions on stackoverflow. For details, read “How do I ask
574.
by Const-me 3y ago
> asking about the best way of doing X To be fair, English is not my native language. But when I see a question “what’s the best way of doing X”, when it doesn’t have any criteria for what’s the best would be, and no other ways of doing
575.
by Const-me 3y ago
> It doesn't show any effort from your side, you just pose the question You probably assumed I asked that question? If so, the assumption was incorrect, I answered it. > without explaining what you attempted, why it worked/d
576.
by Const-me 3y ago
Not GP, but here's 3 questions I managed to recover: https://stackoverflow.com/questions/57064879/finding-coordin... https://stackoverflow.com/questions/57323981/realtime-video-...
577.
by Const-me 3y ago
Moderation has ruined the site for me. Stupid people don't realize they are stupid. When they can't understand a question they conclude there's something wrong with the question, not with them. And since they believe the ques
578.
by Const-me 3y ago
The file formats are produced on another end of the RS232 cable. That particular oscilloscope can send screenshots over that cable, in various formats.
579.
by Const-me 3y ago
Here’s a cool project which could potentially help with these handwritten notes they asking for: https://www.youtube.com/watch?v=cQO2XTP7QDw
580.
by Const-me 3y ago
I don’t think any of these efficiency claims are true. https://www.uops.info/ has data for Alder Lake from 2021. For example, some performance-critical code I have bottlenecks on FMA instructions like vfmadd132pd. Fast core
581.
by Const-me 3y ago
> Not that having a ton of E-cores isn't cool Personally, I don’t find that cool, and I think I will stay with AMD CPU for my next desktop. I want all cores to be fast, and I have enough parallelizable work for them. Also I’d like t
582.
by Const-me 3y ago
I don’t believe that’s the reason. I think it’s project management priorities. It’s possible to write efficient code to this day, examples of very complicated software implemented that way are triple-A videogames. However, it’s relatively h
583.
by Const-me 3y ago
I have once made something remotely similar, to interop between C++ and C#: https://github.com/Const-me/ComLightInterop I took different approach. Because I only needed to support these two languages, there’s no separa
584.
by Const-me 3y ago
Good article, but couple remarks. > most hardware seemingly just runs workgroups in a serial order The hardware runs them in parallel, but it’s complicated. The nVidia GPU I’m currently using has 32-wide SIMD, which means groups of 32 th
585.
by Const-me 3y ago
It seems the device embeds a conventional ARM Linux. If I were developing embedded software for a device like that, I would have used asp.net core runtime. It’s much easier to create secure web apps using C#, compared to cgi-bin written i
586.
by Const-me 3y ago
I’m not sure heterogeneous memory is such a huge deal, due to performance numbers. PCI express is relatively slow in terms of bandwidth, and especially latency. To compensate, GPUs have dedicated piece of hardware to asynchronously copy blo
587.
by Const-me 3y ago
I think it’s money. nVidia is selling same chips on different boards for different prices. An example from the current generation, $800 GeForce 4070 Ti, and $2500 L4 are based on the same chip. The difference between these two cards is amou
588.
by Const-me 3y ago
You don’t need these buggy AMD-specific technologies to leverage the AMD’s GPGPU hardware. Compute shaders are used by videogames for a decade now. For example, GTA5 from 2013 already used compute shaders to render stuff. At least in my exp
589.
by Const-me 3y ago
Gamers don’t care about FP64 performance, and it seems nVidia is using that for market segmentation. The FP64 performance for RTX 4090 is 1.142 TFlops, for RTX 3090 Ti 0.524 TFlops. AMD doesn’t do that, FP64 performance is consistently bett
590.
by Const-me 3y ago
CUDA was a great technology before 2018. Then in 2018, they prohibited usage of GeForce GPUs in data centers: https://www.datacenterdynamics.com/en/news/nvidia-updates-ge... And during 5 years which followed, they
591.
by Const-me 3y ago
For a starting point, I wrote that article couple years ago: http://const.me/articles/simd/simd.pdf I don’t recommend assembly. Intrinsics are typically good enough performance wise, and writing correct assembly i
592.
by Const-me 3y ago
Good idea, and it can be used for both prologue and epilogue pieces. Updated that gist. However, this is only relevant for very small inputs. For longer inputs the vectorized portion of the function gonna dominate the performance.
593.
by Const-me 3y ago
I don’t have any experience with that library. Still, based on the documentation you have linked, I’m not sure it could possibly generate some code similar to my version. I could be wrong but I don’t see APIs which aggregate or accumulate t
594.
by Const-me 3y ago
I’m probably an optimization expert, and I would solve that problem completely differently. On my computer, the initial C version runs at 389 MB / second. I haven’t tested the assembly versions, but if they deliver the same 6.2x speedu
595.
by Const-me 3y ago
Looks very cool, but why? Unlike many other applications of SDFs, for this particular one a traditional rasterization seems more appropriate. The polyhedron being rendered only has 32 vertices and 60 triangles. The indexed mesh gonna take 7
596.
by Const-me 3y ago
Their "Table 1" needs 2 more rows: purchase price, and typical hourly rent price. The article mentions cost efficiency couple times, yet fails to say anything substantial about that.
597.
by Const-me 3y ago
About micro-optimizations, I don’t know AVX512 but with AVX2 you could compute 4 nodes in parallel, an example in C++, untested https://gist.github.com/Const-me/90a52f291c1fcb06142307facdb... Here’s another idea how to
598.
by Const-me 3y ago
I wonder why won’t they fix the performance by changing the API specification of these collections? I don’t think they care about backward compatibility. For example, in C++/20 they decided UTF-8 string literals (available for nearly a
599.
by Const-me 3y ago
That reserve() method only reserves memory for the hash table inside the container. It does help with performance to an extent, but the memory for the items stored in that map is still allocated one by one, as new entries are inserted into
600.
by Const-me 3y ago
Adding `map.reserve( vec.size() );` indeed helped, but still slower than sorting these entries. Specifically, the time for the hash map is 740 μs this way. I think the main reason for the observed behavior is memory management. The sorted v

More ›