nhaehnle

18 ms·

121.
by nhaehnle 9y ago
AMD GPUs actually have identical performance for int32 and fp32, except for full 32-bit integer multiplies and divisions. I think that's a big part of why cryptocurrency miners like them so much.
122.
by nhaehnle 9y ago
Of course glDrawElements() has to go into the driver, because it needs to do hardware-specific work. This is obviously true in OpenGL(ES), but even if you were to implement it via a translation layer to Vulkan, you still have to call the va
123.
by nhaehnle 9y ago
> The current vertex is a globally set variable (as you'll notice once you try to set textures and colors, becayse glVertex3f just sets a global variable), etc. Bad example. glVertex3f is the one function (among the various glColo
124.
by nhaehnle 9y ago
Fair enough. The first one, for the memory comparison, is indeed from 2015. The second one isn't -- 2015 desktop/server GPUs didn't have good half-float operations yet, as that hadn't really been a market. However, the f
125.
by nhaehnle 9y ago
Sure, but read the slide. They have 64k multiply-accumulate units, running at 700 MHz. That means ~46T/s multiply-accumulates, which means ~92T/s individual arithmetic ops. It's a standard way to measure this. I think it&#x27
126.
by nhaehnle 9y ago
I really don't get how they came up with those numbers comparing CPUs to GPUs. They claim to have 3.5x as much on-chip memory as a GPU, but the R9 Fury X has 16.7 MiB of register memory compared to their 28MiB. And then of course the
127.
by nhaehnle 9y ago
That's not really a response to matt4077's argument, which is that what a "reasonable person" may think can change based on changed circumstances such as development of new technological possibilities.
128.
by nhaehnle 9y ago
Humans can't agree on what constitutes hate speech. I mean, there are general sentiments that most reasonable people agree on, but it gets hairy very quickly in the details. The real problem here is that by letting an AI decide wha
129.
by nhaehnle 9y ago
Even if you don't have a compositor you still have at least a window manager, which has much the same issues (high latency, ICCCM).
130.
by nhaehnle 9y ago
People over-emphasize a lot of the problems X has. A lot of the real original problems like high latency were evolved away. And while backwards compatibility can be annoying, that's just what you get for trying to do serious software e
131.
by nhaehnle 9y ago
That's actually a cool explanation for why some people might think this is correct. Let's teach them that 1 is an exponent, too ;)
132.
by nhaehnle 9y ago
> These communities are built on strong network effects – where, according to Metcalfe’s law, the value of the network is exponentially increased by the number of users who participate. It's not exponential! Heck, the footnote is ci
133.
by nhaehnle 9y ago
People are already stacking multiple layers today for memory, although the layers are always manufactured separately and then bonded together in a separate step. I wouldn't be surprised to see more of that in the future, think caches o
134.
by nhaehnle 9y ago
You haven't really addressed gh02t's point. The original (professed) intent of patents was to protect individual inventors. If some other manufacturer finds a cheaper way to manufacture the same thing, they can get a patent on the
135.
by nhaehnle 9y ago
Mildly amusing observation: You're defending the original article, yet at the same time you're perpetuating sexist assumptions (to be clear, I mean the assumption that men are inherently worse than women at certain tasks). This is
136.
by nhaehnle 9y ago
Now refactor everything so that the purpose-built shaders are actually generated from the ubershader simply by hard-coding the various decision points, possibly using preprocessor tricks? Seems like a natural next step that should be able t
137.
by nhaehnle 9y ago
You should be happy to hear that while AMD still has a proprietary driver for some outlier workstation cases, the officially recommended "default" driver is the fully open-source stack.
138.
by nhaehnle 9y ago
You can enable the IOMMU to prevent a PCI device from doing arbitrary DMA.
139.
by nhaehnle 9y ago
Something equivalent to deferred loads is a part of AMD's GCN architecture, which has existed for many years now. (It's not exactly a deferred load; rather, there's an explicit wait instruction that can block a thread until a
140.
by nhaehnle 9y ago
My pet theory is that if everything goes as well as it can go, people thousands of years from now will call the period we are currently living through the singularity. I mean, if you look at the definition of singularity that is implied by
141.
by nhaehnle 9y ago
Usually, people who point out that money is a human-made construct are not trying to say that money is useless. Rather, they're trying to encourage you to question how money is constructed, likely because they think that making some
142.
by nhaehnle 9y ago
I'm honestly uncertain. My best guess right now is: If you know what you're doing, it's possible that you might do memory management better in Vulkan. Then again, it's possible that by trying to exploit the freedom Vul
143.
by nhaehnle 9y ago
Eh, I would say Vulkan decreases reliability. Precisely because Vulkan drivers are much thinner and do no error checking, it becomes very likely that you will end up writing code that works on one system but fails on the next. Granted, Open
144.
by nhaehnle 9y ago
This is neat. There's a minor race condition at startup because scanhostdir calls scanportdir before watchdir. Reversing the order of calls would close the gap and shouldn't have any adverse effects.
145.
by nhaehnle 9y ago
I understand that somebody may not want to raise such a matter personally. The real problem is that OP wasn't even given the details of the accusation. Because even assuming that the accusations are correct, there is likely no malice i
146.
by nhaehnle 9y ago
To be fair, the PSP is involved quite intimately in the various features for securing VMs. With the Zen processors, it should in principle be possible to send an encrypted VM to a cloud provider and be certain that this is the VM that runs,
147.
by nhaehnle 9y ago
Since the title mentioned GPU-friendliness: GPUs are built all around scatter/gather memory accesses. The actual performance of memory accesses of course still depends on locality. The first load in a parallel binary search will be fas
148.
by nhaehnle 9y ago
Traditionally not really, unfortunately. The problem is that negative results usually don't allow you to tell a story. The solution to this is a publication process where you register studies before you do them, in exchange for guara
149.
by nhaehnle 9y ago
So, the one really interesting (new to me) part of the article is that they mention a study according to which boys are born more emotionally sensitive than girls. If this is true, then society is doing something that is clearly against nat
150.
by nhaehnle 9y ago
Performing studies with low N makes it more likely for spurious effects to pass the various statistical effects. Therefore, it makes it more likely to obtain a result that can be published (despite being wrong).

More ›