The title says AMD but it seems like they just wanted to talk about NVDA. Was that actually the report on the AMD processors? "It's a chip for AI like NVDA makes, but they already announced it in January so they gave some more details on it". Thanks for the info?

I bet on AMD, each year they do a survey and come up with good tech and keep in mind with a very keen QA team.

@paxdriver

Жыл бұрын

You have a short memory lol AMD has been great since zen 2 only 4 or 5 years ago for CPU only. Since then they've been great at GPUs only for gaming, not productivity and not compute but their server chips based on zen have been added to their elite product stack. AMD might make good GPUs in the future, but as far as useful, productive work, amd has not been hitting it out of the park, they've just been beating Intel thanks to TSMC and match Nvidia for just gaming - again, thanks to TSMC. Theyre doing well, but Intel has done nothing and Nvidia has sat back and done nothing for 3 generations now happy to reap the rewards of 3 year old technology. It's not hard to catch up on chip designs when nobody else is doing anything productive on the engineering front in half a decade. Especially since AMD doesn't even need to manufacture anything. You give them way too much credit just for not being incompetent imho lol

@praveenspike

Жыл бұрын

@@paxdriver @paxdriver well graphics and processors are 2 different components, 1 company does both while the other anyone can guess. So comparing Intel, nvidia and AMD at the same time becomes irrelevant. If Intel and nvidia were both the same like AMD then your point becomes valid.

@paxdriver

Жыл бұрын

@@praveenspike Graphics are vector and matrix operations just like AI. They are the same thing. Acceleration is just hardwired circuitry to make more complex matrix / vector operations more quickly. But they're not different, they are the same thing graphics and machine learning.

The problem with AI cores in RDNA3 and CDNA3 are half slower than Nvidia tensor cores. Amd ai engine only can do 8X INT8 but nvidia gpu can do INT8 calculation up to 16X and 32x(with sparsity). Amd also lack of software support plus on their presentation they only do single demo.😢

@duladrop4252

Жыл бұрын

Just want to correct you on this, their MI300X is a beast not even two H100 Hopper can even beat one of MI300X yeah it can do 5.2K TFlops compare to the 3.9KTFlops of Two H100 of Nvidia, and has a beefy memory of 192GB of HBM3 again its Twice of Two H100 combine memory.

This guy doesn't know anything about AI or accelerators.. Are they tensor or vector focused, or generic matmul accumulators, or simply HBM equipped, are they FPGA or ASIC or do they perform well under graphics rendering workloads too? He's literally said nothing at all about the new processor other than its like Nvidia... Nvidia's what? Their A100's or their CUDA series? Yahoo is beating you on tech reporting Bloomberg, get in the game.

@kusumayogi7956

Жыл бұрын

The problem with AI cores in RDNA3 and CDNA3 are half slower than Nvidia tensor cores. Amd ai engine only can do 8X INT8 but nvidia gpu can do INT8 calculation up to 16X and 32x(with sparsity). Amd also lack of software support plus on their presentation they only do single demo.😢

@paxdriver

Жыл бұрын

@@kusumayogi7956 It's almost entirely the software and driver sid eof things. They could easily develop chiplets with more advanced operations but the reason they don't is because they haven't got a basic API working yet and they'd be putting the cart before the horse if they over developed the hardware too soon. The API needs to catch up first, Nvidia has a huge head start on that front.

Rubbish. 😂😂😂