AMD’s Big Navi Is Supposedly 40 to 50 Percent Faster than the RTX 2080 Ti

Wonder if they be doing chiplet style or mono die
Pretty sure it'll be a chiplet design. We did a number of stories last year alluding that was their plan. Granted, it was pure speculation then but seems like a safe bet.
 
So, AMD has been claiming an "up to 50% improvement in perf per watt".

I'm not sure where this will be coming from considering it's basically the same architecture, and on the same process node, but lets take them at their word for a thought experiment here.

Typical desktop GPU's tend to max out at 250W TDP somewhere. If AMD is able to get 50% higher perf per watt, and they cap it at 250W, we should be seeing something that is ~60% faster than a 5700xt, which based on a quick googling of benchmarks would make it trade blows with a 2080ti.

Now, as we saw with Vega, AMD has been willing to stomp all over convention in the past and go with very high TDP designs, shipping them with expensive AIO coolers. If they are willing to go up to 350W again like they did with one of the liquid cooled Vega 64 (Fronteir edition or something?) we could be talking 125% faster than the 5700 XT, which would make it 40% faster than a 2080ti.

So, based on what they have said to date, unless they are hiding something, the absolute fastest we can expect is 40% faster than 2080ti.

That is only if they both somehow succeed in pulling a rabbit out of a hat and gaining 50% more performance per watt out of the same arch on the same process node (which seems unlikely, but I am going to reserve judgment until I see what they launch) AND go with the nuclear option of releasing a massive chip with a 350W TDP as a consumer GPU.

Remember, the version of the Vega that went to 350w in the past was:
1.) Very expensive at $1,499
2.) Targeted at professional users

So, I would love to see an amazing consumer big Navi come out, but to say I am a skeptic is an understatement....

...unless they have a very well kept secret we don't know about.
Vega Frontier was a compute-heavy card. Even so, at 350W it was only 3% faster than the Vega 64 in games and 20-25% faster in compute applications. A linear increase in power does not necessarily correlate to linear increase in performance.
 
I am not predicting neither for Ampere or RNDA2 driver soundness or overall performance. Just waiting for Brent to have some fun putting them through the mud, fire, guns, IQ observations -> Heaven is coming for some. Speculating does not matter, only the real thing in your hand or at least Brent's case here.
 
So I've read its 2x 36cu for 72cu... Wonder if they be doing chiplet style or mono die. Io die and chiplet is entirely possible, though never mentioned or even suggested, ... Don't know if AMD can be this tight lipped. Now imagine multiples of 36, so could be 36, 72, 108, and 144? Nah couldn't be..
Otherwise i think 40% above 2080ti is possible and probable. But while I wish it was something so advanced as a chiplet gpu, I doubt it.

Yea nobody would do Chiplet with CPU's either... it's just dumb and would be so difficult to do... oh wait... I think you're tossing out good eggs with the bad. Lets just sit tight and see what happens.
 
Become a Patron!
Back
Top