NVIDIA CEO Isn’t Intimidated by AMD’s New Multi-Die GPU, Laughs at the Concept of an “NVIDIA Killer”

Tsing

The FPS Review
Staff member
Joined
May 6, 2019
Messages
12,595
Points
113
nvidia-jensen-huang-kitchen-geforce-rtx-3080-1024x576.jpg
Image: NVIDIA



AMD recently revealed its Instinct MI200 Series Accelerators, HPC GPUs that are generating a lot of buzz for being the world’s first multi-die GPUs capable of what is said to be a 4.9x improvement in performance versus competing products. Despite its promise of up to 47.9 TFLOPS of FP64 performance and 3.2 TB/s of memory bandwidth, NVIDIA CEO Jensen Huang doesn’t seem phased by red team’s achievement, however. Asked about the development by TheNextPlatform’s Tim Morgan in a recent interview, Huang shrugged his shoulders and suggested that the new Instinct was just another supposed “NVIDIA killer” that would fail to live up to that claim.



From a portion of Morgan’s interview with Huang regarding NVIDIA’s competition, software, and the Omniverse:



Timothy Prickett Morgan: And now there is a two-year cadence in place for GPUs, DPUs...
Click to expand...

Continue reading...


 
Grimlakin said:
Hate to say it but that's pretty accurate.
Click to expand...

Well, I mean, that's exactly what he would say whether or not he actually is concerned.

I think it depends on how they implement the multi-die tech.

If they are just doing another multi-GPU board that relies on some version on Crossfire, then I agree. It will likely be trash.

If on the other hand they have solved the difficult chiplet interconnect issues and are able to do it like with CPU chiplets, so that all the shaders show up as if they are part of the same GPU, and work together like one large GPU, this could be absolutely game changing.

Suddenly the ability to produce a bigger and badder GPU is simply matter of a combination of perf/watt, limited by cooling capacity, power draw and physical space. It becomes trivial to just add more chiplets and a bigger cooler if you need more performance.

Yields should also go through the roof. Instead of the huge monolithic 519.8 mm2 6900xt, you can have multiples of tiny chiplets each with much much higher yields, and binned to be able to clock much higher.

It should drop costs, increase flexibility and result in the potential of some pretty **** badass video cards.

That's a big IF though. The interconnects would have to be SUPER low latency or it just wouldn't work.
 
I'm just saying it is often said that the next gpu is going to be the killer of the king. Not that it's to be dismissed.
 
Nearly Every generation the next chip is better than the last, so to say the next GPU is killer is no real statement. If the next gen could pull off a better than 980ti -> 1080ti kind of upgrade they would have my attention. If they could produce enough to be readily available then they would have my interest, and my money. The longer the shortages go, the less and less interested I get.
 
This from a company that the only product that is easily purchased off the shelf is an overpriced GT 1030...... Eff em both.
 
According to AMDs presentation , it looks pretty great. I don't think we will see this any time soon on gaming though. I don't know what the Instinct cards are going to list for, but it will surely be insane.
 
 
No where in the interview does it state the Jensen laughs at the nvidia "killer".
 
Nvidia is scared. They probably have zero r&d in to what AMD has been doing since 1000 series Ryzen. Multiple chips with infinity fabric appearing as one die. If anyone has an edge on making GPUs with similar tech it's AMD.
 
 
You must log in or register to reply here.
Become a Patron!
Back
Top