simne

36 ms·

451.
by simne 3y ago
> you still can train them on a single GPU with 24GB video memory It depends, on what target. For pure science (or for enjoy), I could train GPT-4 class model on C64, but this method will not fit on concurrent market, where need fast che
452.
by simne 3y ago
For about tech questions you asked. You asked right questions, but you missing context. What really main bottlenecks of NN hardware are neither number crunching, nor memory. Real bottleneck is that GPT-2 is may be last LLM for which was pos
453.
by simne 3y ago
Exists one important thing, many people don't aware of. When some good smart team (business or not it is not much important), focus on some task and have corresponding resources, it really could make things, impossible for universal te
454.
by simne 3y ago
> But they claim they are going to achieve a 10^6x Classics of management, to ask people more then they could, and they will do most possible, so I don't bother much on such claims. And also this is teambuilding bs, to motivate peop
455.
by simne 3y ago
And it is extremely hard to find person, who prefer the people who were deep in the trenches. What I see most time, managers buy bright cover, very few bother to look deeper. And it's easy to understand - as said wisdom, "No one f
456.
by simne 3y ago
> I know people who spend more than I spent just on graphics cards every year Could you make any assumption, how many such people exists? If this number enough to achieve some financial sustainability for manufacturer?
457.
by simne 3y ago
> how is thermodynamic linear algebra different from monte carlo algorithms You may think of it as case of hardware for generate ideal noise on steroids. And in typical computing system it is serious enough problem, that in some first Xe
458.
by simne 3y ago
Easy. I made tests on desktop core i7-7700 with 64G DDR4-2400. And I've tested 13B..30B..70B models on it, and you may imagine, how easy to manage how many CPU cores used. Answer is - it is really works, but slow (about 0.5..1 tokens p
459.
by simne 3y ago
It is known for electronics designers, that specialized circuits outperforms GPUs for few times. Before appear Tensor cores, GPUs was about 4 times worse (speed, power consumption). With Tensor cores, GPUs become better, but they still need
460.
by simne 3y ago
It is known for electronics designers, that specialized circuits outperforms GPUs for few times. Before appear Tensor cores, GPUs was about 4 times worse (speed, power consumption). With Tensor cores, GPUs become better, but they still need
461.
by simne 3y ago
Pullman Company will disagree with you.
462.
by simne 3y ago
Two answers, each 100%. It is known very long time, that specialized AI chips are more efficient than GPUs, even when we talk about Tensor cores, because GPU is not only Tensor cores, they MUST include ramdac, all other video stuff, and net
463.
by simne 3y ago
It is international division of labor. Silly thing, exUSSR become late into semiconductor market, so all sub-micron deals was under Asian manufacturers, but nobody want to work with older tech. Rumors said, in 2000s, Russians made near all
464.
by simne 3y ago
Well, I visited page, because IBM mentioned. Nothing new found. It is known very long time, that IBM hardware is reliable and you must buy it to run their software (to be honest in this case I'm not sure what IBM software I could run o
465.
by simne 3y ago
You first plasma physicist in this topic. I'm electronics engineer, but learn plasma physics also, so I also see very obvious mistakes. But I also have ballistics knowledge and know about Sun wind and light pressure on particles, and I
466.
by simne 3y ago
Question to all here: anybody (except me) learn plasma physics and understand what author write about? Only I see so obvious mistakes in physics? 1. Debye length DECREASE is harmful, but he talks about INCREASE. 2. When aluminium vapor beco
467.
by simne 3y ago
> The study in question HAS NOT been peer reviewed You first seen this!
468.
by simne 3y ago
> the resulted compounds will still be in the atmosphere Why you think so?
469.
by simne 3y ago
Your words so novice, and you make huge number of mistakes, so I need spend much resources to talk with you. And I have no feeling, you value my time. Please answer question(s): why these efforts valuable for me? Or why these efforts valuab
470.
by simne 3y ago
Well, I seen many times, when one people lie to other people. You are very special case, you lie to yourself. Sure, in democracy you have right to do so, as you wish.
471.
by simne 3y ago
You hit. Exists just two views on high education (middle education considered just mandatory). First view, that high education must be right (now some countries already formalized guaranteed free high education), and second view, that high
472.
by simne 3y ago
Greetings from Ukraine! Country at war, after decided to move from Soviet world to West. When Ukraine is different, but we have very same troubles - we learned that Universities should make us new technologies, and why I said about war - ne
473.
by simne 3y ago
BTW revision:2 also said much. For real commercial chip it is fortune to see revision 5, but usual for market something more than 10. This is because chip manufacture is iterative process, you spent at least 3 revisions to achieve predictab
474.
by simne 3y ago
You comparing apples with carrots. In democracy you sure have rights for such comparing but I have rights to say you are wrong. > I dont see how ISA capable of ~5MB/s write speed is a bottleneck when updating 320x200@60 takes less t
475.
by simne 3y ago
Привіт, Слава Україні! Героям Слава! Well, first, I know few people in Pyatigorsk, and they said me, Beshtau is in strict translation from Tatar "five mountains" or in Russian "5 gor = Pyatigorsk" :) I now cannot say you
476.
by simne 3y ago
To boost credit rating. When you have none credits, your credit rating is ZERO, and it is considered bad. So you could pay for example 10$ (yes, small amount) from zero percent credit, and regularly pay it and this will effectively raise cr
477.
by simne 3y ago
PS production cost is not tied to market price in market economy. This is very typical mistake to think them tied.
478.
by simne 3y ago
What you want to say with all of these stuff? ISA was not an option for VIDEO card at early 90s. VESA (VLB) was not an option, because it was LOCAL bus, closely tied to CPU architecture, so 486 VLB was not compatible with machines with ANY
479.
by simne 3y ago
Greetings from Ukraine, country at war now, and one of main foundation whale of USSR at not too far past. I'm not too old, but I born in USSR and seen from inside, how it steal western technologies and where it stuck. We now chosen dem
480.
by simne 3y ago
> making ISA video cards instead Sorry, I'll disappoint you. I've experience with these times PCs, with integrated video and with discrete, and must say, this was killer difference - just same card on ISA bus was magnitudes slo

More ›