AMD: RDNA 3 Speculation, Rumours and Discussion

[Esrever]

AMD never said anything about any of that stuff, you guy are reading way too much into it lol. All they said was they have an efficient arch and will improve it next gen.

Also the quote is "AMD RDNA™ 3 and Radeon™ graphics as a true leader in efficiency.", not the true leader, just a leader..

 

[hurleybird]

If they know they will beat Nvidia in performance/watt then they must know both Nvidia's performance and watts.

I know I'm shorter than Michael Jordan, yet I don't know his height offhand and couldn't tell you how much taller he is. Let's be serious here. Anyone with a bit of common sense should be able to tell by now that, given the likely ranges in performance and power consumption for both companies, the probability that Nvidia ends up more, or even as efficient this generation is not much higher than a rounding error. AMD knows exactly where they'll end up, has a better idea than we do where Nvidia will end up, and despite their general ultra-conservatism is already all but preannouncing their efficiency win.

 

[DavidGraham]

AMD knows exactly where they'll end up, has a better idea than we do where Nvidia will end up, and despite their general ultra-conservatism is already all but preannouncing their efficiency win.

You need to ask yourself this: why do AMD feels the urge to talk about performance per watt this day? What's forcing them to do any kind of response, to say anything? They've never done this kind of thing before.

Also,

https://twitter.com/x/status/1571891879280533507

 

[trinibwoy]

I know I'm shorter than Michael Jordan, yet I don't know his height offhand and couldn't tell you how much taller he is.

Lol that’s a pretty bad analogy. You have lots of evidence that Jordan is taller than you.

Let's be serious here. Anyone with a bit of common sense should be able to tell by now that, given the likely ranges in performance and power consumption for both companies, the probability that Nvidia ends up more, or even as efficient this generation is not much higher than a rounding error. AMD knows exactly where they'll end up, has a better idea than we do where Nvidia will end up, and despite their general ultra-conservatism is already all but preannouncing their efficiency win.

There have been generations where Nvidia has taken the power efficiency crown and generations where AMD had it. If anything the odds are in Nvidia’s favor this time around coming off a supposedly much inferior process. There’s nothing about RDNA 3 or Ada that we know so far that would imply it’s a slam dunk power efficiency win for either side.

According to TPU the 3060 Ti and 6700xt offer similar performance while the 3060 Ti is more power efficient. So clearly any claims AMD is making about efficiency leadership must be based on something they know about Ada. Guessing based on Ampere won’t cut it.

 

[xpea]

Nvidia is taking power consumption to unprecedented levels

To your short memory, AMD Radeon HD 7970 X2 in 2012 and R9 295X2 in 2014 say hello with their 500W TDP

 

[troyan]

How can they claim leader in perf/watt when a 3090 with 350W runs circle around a 6950XT with 335W? Using outdated games without raytracing is basically the Nokia moment from then.

BTW: Lovelace will be nearly twice as efficient. How exactly can AMD think this blog post is good marketing for them?!

 

[Leoneazzurro5]

The statement was only about the AMD focus in producing very efficient products. One can read what it wants from that - and in the past a lot of wrong readings were done on such statements. In any case, it may indeed be that Nvidia is ahead in pure performance. The leakers "who knew" like Kepler, also told in some posts that top cards cards are quite close in pure performance (in a 10% range). So the fight will come down more to other factors: price, price/performance, availability, and some may look also at power efficiency.

 

[Leoneazzurro5]

According to TPU the 3060 Ti and 6700xt offer similar performance while the 3060 Ti is more power efficient. So clearly any claims AMD is making about efficiency leadership must be based on something they know about Ada. Guessing based on Ampere won’t cut it.

Never knew that you should use the worst efficiency SKU compared to a lower clocked midrange part from the competition for establishing the relative power efficiency (raster, RT is clearly in Nvidia field). We could have used the 6800 and the 3070Ti, maybe, who knows. And of course there is also a process difference. So all in all it's not easy to compare different architectures in perf/w.

 

[Bondrewd]

Also if they offer hardware ML acceleration with RDNA3, which is rumored to be the case.

Nope.
Also never.

Let's not pretend that Nvidia release only 4090 type GPUs shall we?

Everything below AD102 is all flavours of anemic and gets shredded by N3x parts.
Especially AD106/107, those babies gonna feel the pain much like GTX260 did all those eons ago.

 

[arandomguy]

That chart is just an updated version of the same campaign they ran back in May -

https://twitter.com/x/status/1526264896731680769

Interestingly it also appears their lead is going down in most categories.

 

[troyan]

Nope.
Also never.
Everything below AD102 is all flavours of anemic and gets shredded by N3x parts.
Especially AD106/107, those babies gonna feel the pain much like GTX260 did all those eons ago.

Okay, i take the bait: AD106 will have ~20 billion transistors at 200mm^2. How exactly is AMD "shredding" that with 6nm?

 

[Bondrewd]

AD106 will have ~20 billion transistors at 200mm^2.

Meaningless numbers that don't even account for cost delta.

How exactly is AMD "shredding" that with 6nm?

With magic, of course.

 

[trinibwoy]

Never knew that you should use the worst efficiency SKU compared to a lower clocked midrange part from the competition for establishing the relative power efficiency (raster, RT is clearly in Nvidia field). We could have used the 6800 and the 3070Ti, maybe, who knows. And of course there is also a process difference. So all in all it's not easy to compare different architectures in perf/w.

You could do that but it wouldn’t change the conclusion. You could also use the 3080 Ti and 6900xt. Btw the 6500xt is the least efficient RDNA2 SKU by some margin.

 

[Leoneazzurro5]

You could do that but it wouldn’t change the conclusion. You could also use the 3080 Ti and 6900xt. Btw the 6500xt is the least efficient RDNA2 SKU by some margin.

Of course the conclusion would change, in TPU the 6800 has higher average FPS than a 3070 Ti and at the same time it has lower power consumption.
This only to say that comparing a single point is simply misleading, you now cite the 6500XT but if we take the 6600XT we would have a complete different story.

 

[trinibwoy]

Of course the conclusion would change, in TPU the 6800 has higher average FPS than a 3070 Ti and at the same time it has lower power consumption.
This only to say that comparing a single point is simply misleading, you now cite the 6500XT but if we take the 6600XT we would have a complete different story.

Exactly. And we don’t know which data point AMD is using when they talk about efficiency. All we know is that in the 8nm vs 7nm battle they traded blows depending on performance level.

 

[DegustatoR]

Everything below AD102 is all flavours of anemic and gets shredded by N3x parts.
Especially AD106/107, those babies gonna feel the pain much like GTX260 did all those eons ago.

Ah, yes. The fabled "shredding" (which has been going on for about two years now if you still trust some user here) is going to get "shreddier".

 

[Bondrewd]

Ah, yes. The fabled "shredding" (which has been going on for about two years now if you still trust some user here) is going to get "shreddier".

Of course, that's always how it goes with AMD.
Shit gets nicer and crazier!

 

[Leoneazzurro5]

Exactly. And we don’t know which data point AMD is using when they talk about efficiency. All we know is that in the 8nm vs 7nm battle they traded blows depending on performance level.

But yesterday's piece (which started the convo) was not about comparing itself with the competition, instead it was a comparison between RDNA2 and 3.
And yes, also that is influenced by the SKU choice of course. Both GPU makers appear to have more efficient architecture compared to the past, btw.

 

[DegustatoR]

Of course, that's always how it goes with AMD.
Shit gets nicer and crazier!

Let's see if anyone will notice anything this time at least.

 

[DavidGraham]

instead it was a comparison between RDNA2 and 3.

It was also about the competition, mentioning NVIDIA by name several times, and then posting an old modified slide about efficiency compared to competition.