atq2119

23 ms·

391.
by atq2119 3y ago
I wouldn't be surprised if you're right, but it would be helpful to spell out acronyms like this one.
392.
by atq2119 3y ago
There have always been a number of reasons for why somebody might believe in the supernatural, the sibling comment points out one of them. Two important reasons have declined substantially over the last 200 years. First, the "God of th
393.
by atq2119 3y ago
Thanks, I had missed that development.
394.
by atq2119 3y ago
Which DX12 Ultimate features are missing from Vulkan? They may be missing from Roadmap definitions, but that wasn't my point. Vulkan is often first for innovation being accessible / possible in some form, but DX12 is usually fir
395.
by atq2119 3y ago
The Vulkan Roadmap definitions are supposed to address this problem for the 95% use case. If you want to be at the bleeding edge of technology, you're always going to have to put in some extra work. At least Vulkan makes that possible:
396.
by atq2119 3y ago
Just for the record, the "not at all" part is incorrect for the nodes I'm aware of. Correct would be "way worse", i.e. it's still getting denser, but the improvement is way worse than that of random logic.
397.
by atq2119 3y ago
I agree it's quite surprising, but in hindsight it makes some sense: The original implementation has 2 branches per iteration: one for the loop, and one to test whether the bit is set or not. The optimized implementation has 1 branch p
398.
by atq2119 3y ago
Spot on. FWIW, exponential growth as a function of a discrete variable is very common (e.g. all of algorithmic complexity), but it has to be (at least modeled as) an unbounded numeric variable. You can't have exponential growth as a
399.
by atq2119 3y ago
I don't think this is a deliberate strategy but rather the result of focusing resources in where there are currently the largest margins, which is in the big server chips. It is somewhat short-sighted though, since having support on al
400.
by atq2119 3y ago
It's important to note that the luddites didn't necessarily say "no" to technology. What they objected to was how the gains of the technology were distributed. History really does rhyme.
401.
by atq2119 3y ago
Fun fact: It was originally legally mandated that if you form a business, it just be named after you.
402.
by atq2119 3y ago
I think both can be true. By analogy, consider mental arithmetic vs. electronic calculators and spreadsheets. Being able to do mental arithmetic is an important skill, arguably the only way to have a profound and deep understanding of many
403.
by atq2119 3y ago
Agreed, though I'd add that it's quite likely that "simply don't use AI tools" will become less and less viable relatively quickly.
404.
by atq2119 3y ago
Another important ingredient of informed consent is that there are realistic, viable alternatives. Those alternatives may have downsides, but if those downsides are so large as to become crippling, they stop being realistic and viable. That
405.
by atq2119 3y ago
If human skills behind the model are a limiting factor, then OpenAI is in a pretty good position: they have a headstart, and software tends towards "winner takes all" because of the low marginal cost. Look at Google search for a g
406.
by atq2119 3y ago
Every matrix multiply can be viewed as a 1x1 convolution, just like every convolution can be viewed as an (implicit, for larger than 1x1 kernels) matrix multiply. I'm not sure this is particularly enlightening, but it's probably o
407.
by atq2119 3y ago
I would argue that all of the ideology-based advocacy for Bitcoin and other cryptocurrencies is/was advocating to solve a people problem with technology.
408.
by atq2119 3y ago
That's a choice you make, though, isn't it? If you get that many matches, be more selective. Everybody wins that way.
409.
by atq2119 3y ago
They are breaking in, though. By all accounts, MI300s are being sold as fast as they can make them.
410.
by atq2119 3y ago
AMD was founded at almost the same time as Intel. X86 didn't exist at the time. But yes, AMD was playing the "follow x86" game for a long time until they came up with x86-64, which evened the playing field in terms of archite
411.
by atq2119 3y ago
Makes you wonder why DoE labs and similar facilities don't mandate open licensing of CUDA.
412.
by atq2119 3y ago
My understanding is that there was some trademark silliness around "open compute", and AMD decided that instead of doing a full rebrand, they would stick to ROCm but pretend that it wasn't ever an acronym.
413.
by atq2119 3y ago
Perhaps. But then again, its reason for success and its charm comes from being given access to the University for free because it's being run by a university-associated team. Messing with that would almost certainly destroy the charact
414.
by atq2119 3y ago
That "signing" has always been bullshit security theater. You'll note that every other graphics API manages just fine without it. I'm glad somebody posted the "signing" algorithm out in the open.
415.
by atq2119 3y ago
Not flawlessly. There was a pretty scary bug with the radar(s?) during the moon landing of Apollo 11.
416.
by atq2119 3y ago
The underlying mathematical driver of the power of two random choices is a pretty amazing fact about the "n balls into n bins" random experiment. If the balls are thrown independently and uniformly at random, the most loaded bin w
417.
by atq2119 3y ago
I think you may be too hung up on the C(?) definition of trap representation. Clearly the OP suggestion isn't viable for default C integers anyway, and whichever language OP's suggestion is applied to is free to define the behav
418.
by atq2119 3y ago
Punitive fines are fine, but they need to be scaled by income and wealth (to discourage picking only on the weak), and the concept of "innocent until proven guilty" needs to apply. My understanding of the US problem with civil ass
419.
by atq2119 3y ago
I think the idea is that overflowing would be a guaranteed trap of some kind, i.e. defined behavior.
420.
by atq2119 3y ago
On the claim that instruction sets are undocumented: https://gpuopen.com/amd-isa-documentation/ Of course this doesn't apply to Nvidia, but MI-300 seems to be pretty viable for machine learning, so if you want thi

More ›