Speculation: GPU Performance Comparisons of 2020 *Spawn*

Status
Not open for further replies.
And they're still doing it.
Those capacities just don't overlap.
If hypothetical N10 was on 14LPP, it'll be scaled down in favour of sIODs or smaller client APUs all the same.
"every single time" still is a broad and untrue generalization. One-Liners.
 
The commentary is an accurate near-term profile of AMD’s market position right now, and I don’t think it is “getting ahead of ourselves” in any way.

Sure, revenue diversification matters in long term with AMD itself as the testament, but so as maximising RoI aka best bang for the bucks. With the finite amount of TSMC production capacity (after fulfilling client deliverables), why would they prioritise their traditionally weak discrete GPU products, over driving further their momentum on server and consumer CPU/APU that are the bigger money pies?

IMO it is not about giving up, but picking fights for the bigger fishes to fry. It is not life and death for them to deprioritise dGPU production/launch at this very moment. That and also Nvidia not being on TSMC 7nm gives them plenty of leeway, if they truly have a great product stack in waiting.

You’re right of course but I certainly didn’t get all of that nuance from the one liner I responded too.
 
Pentium M still existed back then, and now it does not.
IDC is washed up.

Intel Austin is nice but they do LITTLEs and other additive SoCs a-la Snow Ridge so that's no AMD comp.
Yeah, AMD is in a way better position than they were back then. Simply, they are better off than they were back then, and Intel is much worse. Them not having the Pentium M uarch to rely is one important difference, then there’s their fabs floundering while TSMC is executing flawlessly among others.

Having as many wafers dedicated to CPUs as possible is a smart choice on AMD’s part right now.
I wonder how many people said that back in the Athlon heyday.
No one who was paying attention to Banias/Dothan.
 
Napkin math time:

5700xt is rated 225W. Take a 50% performance improvement per watt, you get 225W / 1.5 = 150W (ballpark). Double it and you get a 300W and 80CU gpu, effectively double the performance of 5700xt. This is very rough. 5700xt is around 70-80% of an RTX2080 in terms of performance, so you're probably in the range of 140-160% of an RTX2080 at 300W, give or take. So you can kind of play with it and see how much power they'd need to surpass RTX3080, or how much lower they are if they target sub 300W. Will be interesting to see.
 
I know it's napkin math, but napkin questions ... Is the 225W for the entire board or just the GPU chip? Would they need to double the power for the memory?
 
Yah I can improve that. Tdp for a 5700xt is only 180W. More power to the rest of the board than I thought. 45W for the board.

You could do (180 / 1.5 x 2) + 45 = 285W for 200% 5700xt (80 CU) at 50% performance per watt improvement. Very rough estimate. That’s somewhere 140-160% of a 2080ti. 5700xt was like 70 or 80% of a 2080ti. Like, I ascribe no accuracy to this. It’s just to get and idea.
 
Napkin math time:

5700xt is rated 225W. Take a 50% performance improvement per watt, you get 225W / 1.5 = 150W (ballpark). Double it and you get a 300W and 80CU gpu, effectively double the performance of 5700xt. This is very rough. 5700xt is around 70-80% of an RTX2080 in terms of performance, so you're probably in the range of 140-160% of an RTX2080 at 300W, give or take. So you can kind of play with it and see how much power they'd need to surpass RTX3080, or how much lower they are if they target sub 300W. Will be interesting to see.
5700xt is about 85% the performance of RTX2080. A straight doubling of 5700xt performance puts it 40% ahead of RTX2080ti.
 
Adding ray tracing hardware to Navi will eat up some of the performance gains. I don't see a way to assess this though.
Ya i dont think AMD will achieve this level of performance. I was just noting he had low balled the performance of the 5700xt. Something in between 3070 and 3080 for 499$ is my guess.
 
Status
Not open for further replies.
Back
Top