Soldato
- Joined
- 21 Jul 2005
- Posts
- 21,107
- Location
- Officially least sunny location -Ronskistats
Totally contradicts what the local experts say..
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD RX 7900XT, 90% to 130% faster than 6900XT, MCM, Q4 2022.
- Thread starter humbug
- Start date
Totally contradicts what the local experts say..
Soldato
- Joined
- 21 Apr 2007
- Posts
- 2,651
Unless (as a consumer) you are powering VR, not even the 3090 or 6900 drives those high resolution HMDs well. Its raster all the way for me, if it comes with RT then great but I'd happily trade every shader for raster over RT right now.
With 4K now becoming common (at the high end), its raster raster raster until we are comfortably at 144hz @ 4k across the majority of titles. Next gen (flagships) should be mostly there... just need that level of performance to filter down.
144 FPS at 4K? Very doubtful.
The most challenging titles get 30-40 FPS at 4K (even not taking into account features like ray tracing), so future cards would ideally offer roughly double the performance of the current gen.
I think the performance jump from RDNA2 to RDNA3 will be a bit less than from RDNA to RDNA2.
AMD has to work to power constraints and they've already hit 300w (tdp) with the 6900XT, so how powerful these cards are will probably depend on if the power limit is 400w or 500w.
The most power efficient RDNA2 GPU at 4K is 6800, which has a TDP of 250w:
Power efficiency is definitely something AMD has an advantage on Nvidia over, with RDNA2.
Assuming that 5nm EUV offers some large improvements to power consumption, perhaps AMD could offer something similar to 2 MCM dies of the 6800?
I wonder if it's true that they are going to use some odd combination of 5nm and 6nm EUV fabrication processes.
I think one of the problems with multi chip designs, is they aren't going to be as power efficient as single large die /monolithic design. But, probably cheaper and easier to produce smaller dies.
The most challenging titles get 30-40 FPS at 4K (even not taking into account features like ray tracing), so future cards would ideally offer roughly double the performance of the current gen.
I think the performance jump from RDNA2 to RDNA3 will be a bit less than from RDNA to RDNA2.
AMD has to work to power constraints and they've already hit 300w (tdp) with the 6900XT, so how powerful these cards are will probably depend on if the power limit is 400w or 500w.
The most power efficient RDNA2 GPU at 4K is 6800, which has a TDP of 250w:
Power efficiency is definitely something AMD has an advantage on Nvidia over, with RDNA2.
Assuming that 5nm EUV offers some large improvements to power consumption, perhaps AMD could offer something similar to 2 MCM dies of the 6800?
I wonder if it's true that they are going to use some odd combination of 5nm and 6nm EUV fabrication processes.
I think one of the problems with multi chip designs, is they aren't going to be as power efficient as single large die /monolithic design. But, probably cheaper and easier to produce smaller dies.
Last edited:
Why do people keep saying that AMD will have a power efficiency advantage going forward? How do you guys know?
Surely you aren't stupid enough to base it on current gen cards? Where AMD's RDNA 2 is on a very power efficient 7nm node while Nvidia are using a much less power efficient 8nm node(and it's really a 10nm node) Look at the gains AMD made in power efficiency from RDNA 1 to RDNA 2, that was mainly down to switching to from one 7nm process to a more power efficient 7nm process.
If both were using the same 7nm node which would have the advantage??
Surely you aren't stupid enough to base it on current gen cards? Where AMD's RDNA 2 is on a very power efficient 7nm node while Nvidia are using a much less power efficient 8nm node(and it's really a 10nm node) Look at the gains AMD made in power efficiency from RDNA 1 to RDNA 2, that was mainly down to switching to from one 7nm process to a more power efficient 7nm process.
If both were using the same 7nm node which would have the advantage??
Its not a 10nm node its 8nm , its the last of the DUV, before Samsung to go 7nm and EUV. If you listen to MLID Nv are using up to 600w for th high end next gen part, and AMD are coing in at nearly 40% less than that.
you aren't stupid enough to think 7nm is very power efficient are you? I joke ofc.
There's probably very little difference between the current fab. processes used by AMD and Nvidia. I seem to remember that AMD improved the transistor density for RDNA2, using a presumably optimised 7nm.
Navi 21 (7nm) transistor density: 51.5M / mm²
Ampere GA102 (8nm) transistor density : 45.1M / mm²
Not a massive difference.
So, perhaps it's down to how much each company optimises the fab. process used.
And more importantly, the design of the architecture itself, AMD is more concerned with power consumption.
There's probably very little difference between the current fab. processes used by AMD and Nvidia. I seem to remember that AMD improved the transistor density for RDNA2, using a presumably optimised 7nm.
Navi 21 (7nm) transistor density: 51.5M / mm²
Ampere GA102 (8nm) transistor density : 45.1M / mm²
Not a massive difference.
So, perhaps it's down to how much each company optimises the fab. process used.
And more importantly, the design of the architecture itself, AMD is more concerned with power consumption.
Last edited:
Samsung's 8nm node is just a tweaked 10nm. It's not a power efficient node like the 7nm node RDNA 2 is on.
And seriously MILD? Taking his word on what's coming out in the future would be also be stupid.
Soldato
- Joined
- 21 Jul 2005
- Posts
- 21,107
- Location
- Officially least sunny location -Ronskistats
They have also provided new power adapter cables with their 30 series cards and gone to the trouble of changing the PSU delivery - so this indicates on top of your info they knew this for ages, not sure how people take so long to see the bigger picture. Probably OK for mid-tier though.
And TSMC N7 is a tweak of its own 10nm DUV. It wasnt until they started 7N+ they moved to EUV. Also 6nm is a process tweak of N7 , and AMD are using that (it`ll keep costs down compared to N5 or even the rumoured N4 that Nv are going to use). You are aware of all of this?
As for MLID - he broke the news about the change in PSU spec before anyone else and voila, those said same PSU with 16 pin gpu power are now in the market. Open your eyes son.
Every product that has come out on the TSMC 7nm process has been extremely power efficient. RDNA2 , M1 Mac , Vermeer cpus. To think that the 7nm node process is not having a very large effect on AMDs power efficiency advantage is just wrong , it is a large part of it.
We just will not know which is more power efficent in the next gen until independent reviewers have the cards in their hands. What we do know is that Nv and AMD are both issuing warning about how much power these new cards will use and that is extremely worrying.
A large part of choosing your next GPU will not only have the metrics of performance and price to consider but will also have power use involved in the equation to a degree we have never seen before. If one company can give 90% of the preformance of the other whilst using 75% of the power then I am going for that card.
I really like my 6800 only using 200W and would be a big shock to have to use 600W to get the 2x performance upgrade that I look for in GPU upgrades.
We just will not know which is more power efficent in the next gen until independent reviewers have the cards in their hands. What we do know is that Nv and AMD are both issuing warning about how much power these new cards will use and that is extremely worrying.
A large part of choosing your next GPU will not only have the metrics of performance and price to consider but will also have power use involved in the equation to a degree we have never seen before. If one company can give 90% of the preformance of the other whilst using 75% of the power then I am going for that card.
I really like my 6800 only using 200W and would be a big shock to have to use 600W to get the 2x performance upgrade that I look for in GPU upgrades.
*sigh*
AMD's RDNA 2 is using a much more power efficient process than Nvidia's Ampere GPUs? Yes or No?
That if you take the difference in process out of the equation, it's very hard to know which architecture is actually more power efficient right now.
That you don't know which company is going to have the most power efficient architecture for their next round of GPUs?
Even if the 600TGP turns out to be correct, that still doesn't point to what the actual power efficiency is going to be.
All we got now is rumours. That's it.
You telling me to open my eyes is extremely funny. If opening my eyes means believing rumour sites and clickbait, then no thanks. Everything they say should be taken with a massive pinch of salt, unless you are a fool of course, then just go ahead and believe it all.
And, don't call me son. You don't have a clue what the power efficiency of the next round of cards are, stop acting like you do.
Oh bless you son. I would suggest you start with Dr Ian Cutress then move to level 1 techs to learn exactly WHY you are wrong.
Answer the questions.
AMD's RDNA 2 is using a much more power efficient process than Nvidia's Ampere GPUs?
That if you take the difference in process out of the equation, it's very hard to know which architecture is actually more power efficient right now?
That you don't know which company is going to have the most power efficient architecture for their next round of GPUs?
Guys.
Zen 1, the 1800X had the same MT performance as the 6900K, AMD's CPU was on a brand new 14nm node from Global Foundries, Intel's CPU was on their mature 14nm.
The 1800X used half the power of the 6900X.
Explain that.
Zen 1, the 1800X had the same MT performance as the 6900K, AMD's CPU was on a brand new 14nm node from Global Foundries, Intel's CPU was on their mature 14nm.
The 1800X used half the power of the 6900X.
Explain that.
So should we ignore Rasterization? if so why?