Const-me

38 ms·

541.
by Const-me 3y ago
That particular branch is well-predicted for long arrays. You probably only gonna have a single misprediction for the complete binary search function, the performance consequence of that branch is negligible. People (and compilers, too) don
542.
by Const-me 3y ago
> as it's not a very stateful algorithm It might be stateless, but it depends on many things unknown at compile time. One of them is the input data being processed. Binary search is exactly that, compiler don’t know at which positio
543.
by Const-me 3y ago
I don't have any expectations, but there're reasons for Vulkan to be faster. It's a mature technology used my millions of people every day. Unlike GPGPU compute, for videogames performance directly affects usability. For thes
544.
by Const-me 3y ago
Have you tested Vulkan API on the 7900 XTX? Was it faster or slower than ROCm?
545.
by Const-me 3y ago
> They still go over the various techniques used, their consequences, etc., so the processor chosen as an example is unimportant. They also include various techniques not used anymore, without mentioning that’s the case. I did a search f
546.
by Const-me 3y ago
I’m not related to academia. I don’t design CPUs. I don’t write operating systems and I don’t care about these side channel attacks. I simply write user-mode software, and I want my code to be fast. The academic research used or written by
547.
by Const-me 3y ago
I share the OP’s opinion that a lot of available information is incorrect. It seems the industry is moving faster than the academics who write books and university courses can update these sources. Big endian CPUs, CPU architectures other t
548.
by Const-me 3y ago
As soon as dogs will start causing noticeable changes to the society, they will be banned and dog ownership criminalized. By the way, last week Russians signed a new law which makes collaboration with an unregistered foreign non-profit orga
549.
by Const-me 3y ago
Yes, indeed. And it doesn’t mean promoting or discouraging dogs helps removing dictators from power.
550.
by Const-me 3y ago
Russia and China let their citizens use whatever software they please, be it commercial, pirated, or free. By itself, software is harmless to these autocrats. These regimes care about telecom, not software. Both Russia and China restrict us
551.
by Const-me 3y ago
Absolutely, it better be a decimal. These numbers in .NET are decimal floating-points, they can represent all these 12.34% without any loss of precision.
552.
by Const-me 3y ago
Authoritarian regimes are actually correlated with the popularity of free software. Russia and China both have government-mandated forks of Linux. Chinese companies like Linaro, Huawei and Alibaba contribute to Linux kernel very actively. W
553.
by Const-me 3y ago
I agree in theory, but in practice I’m not sure that’s likely to happen. I think parents and peers have more influence on the children compared to the education system. For this reason, the real problem here isn’t “how to improve education”
554.
by Const-me 3y ago
The C# version is slightly better, though. Binary floating-point numbers are less than ideal for keeping discount percentage because a value like 12.3% can’t be represented exactly.
555.
by Const-me 3y ago
I have a theory that hardware vendors are making Christmas tree-inspired products trying to segment the market between gamers and professionals. Gaming hardware is typically pretty good in terms of performance/dollar. However, similar
556.
by Const-me 3y ago
I agree the hate was always there, but this was equally true even before any internets. I don’t think it’s possible to do anything about that. Arresting all these people, or censoring their speech, or banning them from the internet is IMO l
557.
by Const-me 3y ago
The problem is specific to advertisements on the internet, and modern social media enabled by them. Ad companies have strong financial incentive to maximize user engagement. Over time, they discovered the best way to achieve that is promoti
558.
by Const-me 3y ago
This reminds me of the old pandemic of 1346-1353. Many people died back then. Traditional manorial system, which was based on serfdom, became less viable. Land owners discovered they have to compete for workers, instead of just exploiting t
559.
by Const-me 3y ago
Not enough bandwidth. My current GPU has 484 GB/second memory bandwidth. It would require 7 channels of DDR5-8400 memory (the fastest one currently defined by these specs), and GPUs aren’t yet large enough to fit 7 slots of SO-DIMM.
560.
by Const-me 3y ago
VRAM runs at high frequencies, delivers insane throughput, but the latency is not great. GPUs don’t need low memory latency. They have very high degree of parallelism. GPU cores switch to other threads instead of waiting for data from memor
561.
by Const-me 3y ago
> departure from standards There were no new HTML standards between 4.0.1 in 2000, and 5.0 in 2014. > enforcing its own vision of the web thanks to its enormous market share Modern web is only possible because Microsoft implemented no
562.
by Const-me 3y ago
Interestingly, I actually do. When I write bash scripts which will go to git and probably stay there for a while, I use long form of the switches, like --long-option-name instead of -lon. They make the script more readable when after a year
563.
by Const-me 3y ago
> Lock-free techniques are not "obviously better than locks". On GPUs, they are. GPUs don't have any locks, but they can be emulated on top of these global memory atomics. Because count of active threads is often thousands
564.
by Const-me 3y ago
> expensive data scientists Different people estimate compute costs of ChatGPT to be between $100k and $700k per day. Compared to these numbers, data scientists aren't that expensive. > just run standard CUDA stuff I doubt data s
565.
by Const-me 3y ago
> 32-bit only atomics, I don't think its a very serious solution in practice Yeah, I think I encountered that while porting a hash map from CUDA to HLSL. However, I'm not sure that's necessarily a huge deal. Probably not a
566.
by Const-me 3y ago
> undercut Nvidia so deeply that it makes sense for the big players to hire a team At least for some use cases, I think that has already happened. nVidia forbids using GeForce GPUs in data centers (in the EULA of the drivers), AMD allows
567.
by Const-me 3y ago
The power efficiency of the RTX 4000 is awesome, but it costs performance. The 4070 Ti runs at 2.3 GHz base clock / 2.6 GHz boost, the RTX 4000 SFF only runs at 1.3 GHz base / 1.6 GHz boost. For this reason, despite the chip is th
568.
by Const-me 3y ago
These things are nice to have, but you don’t actually need them. It only takes 1-2 pages of HLSL to implement efficient matrix multiplication. It’s not rocket science, the area is well researched and there’re many good articles on how to im
569.
by Const-me 3y ago
> If people could develop AI stuff on consumer cards Technically people can already do that, by leveraging compute shaders in D3D, Vulkan, or WebGPU. I’m certain it’s possible to implement D3D or Vulkan backend for PyTorch, TensorFlow, a
570.
by Const-me 3y ago
> driver bugs linked to OpenCL and ROCm CUDA is the default tech for GPGPU in HPC or AI applications, for more than a decade now. By now, people have found most of these driver bugs, and nVidia has fixed them. Similarly, compute shaders

More ›