AMD RDNA3 Offers Over 50% Perf/Watt Uplift Akin to RDNA2 vs. RDNA; RDNA4 Announced

[Mussels]

Some Amperecards are truly impressing on efficiency, most notibly 3060ti and 3070 which are far superior to 2060S and 2070 (25-30% more perf pr watt). The rest however were only marginally more efficient (5-10%) than previous gen. GDDR6X is the reason for this on 3070ti and up, 3050 and 3060 seems to scale poorly efficiencywise when BW/cores are reduced.

The node itself cannot be blamed since it is almost twice as dense as previous node on Turing (TSMC 12nm vs Samsung 8nm):

Tech Node name (MTr/mm²)

TSMC 5nm EUV 171.3
TSMC 7nm+ EUV 115.8
Intel 10nm 100.8
TSMC 7nm Mobile 96.5
Samsung 7nm EUV 95.3
TSMC 7nm HPC 66.7
Samsung 8nm 61.2
TSMC 10nm 60.3
Samsung 10nm 51.8
Intel 14nm 43.5
GF 12nm 36.7
TSMC 12nm 33.8
Samsung/GF 14nm 32.5
TSMC 16nm 28.2

Going for Lovelace the nodeshrink can be even greater than Turing to Ampere depending on what Nvidia goes for (TSMC 6 or 5nm?). Mranwhile AMD has had a huge advantage over Turing (TSMC 7nm hpc) on RDNA1 and slight advantage over Ampere (still TSMC 7nm hpc).

It's the power consumption of the VRAM that hurts the top tier ampere cards - that GDDR6X is power hungreh

 

[AusWolf]

Yes, cards above 3070 has the issue of very power hungry GDDR6X, not the arciteture by itself, but the result is minor gains vs Turing in efficiency. 2060 super vs 3060ti is also a good comparison where 2060S uses 185W vs 200W on most 3060ti`s (except 2x8-pin variantsl ike MSI trio), performance is 32-45% higher depending on resolution (TPU chart 2060S FE vs 3060ti FE). Ampere scales very poorly powerwise at small cores like 3050 and 3060. Turing scaled better the larger the die grew (2080ti most efficient), with Ampere we will never know since GDDR6X destroys good powerscaling.

It's the power consumption of the VRAM that hurts the top tier ampere cards - that GDDR6X is power hungreh

Fair enough. I didn't know that "X" made such a big difference. I guess we could say then, that Ampere could bring more efficiency on the table if lower-end cards weren't so severely cut down, and higher-tier ones didn't use GRRD6X. The 3060 is on par with Turing, the 3060 Ti is a bit above it, but all the rest of the line-up is forgettable from an efficiency point of view.

 

[Cowboystrekk]

Fair enough. I didn't know that "X" made such a big difference. I guess we could say then, that Ampere could bring more efficiency on the table if lower-end cards weren't so severely cut down, and higher-tier ones didn't use GRRD6X. The 3060 is on par with Turing, the 3060 Ti is a bit above it, but all the rest of the line-up is forgettable from an efficiency point of view.

Except for 3050, 3070ti and 3090ti all Amperecards are more efficient (5-10%) compared to previous gen, or cards with same consumption, but only 3060ti and 3070 are far more efficient than it's predesessors (2060S vs 3060ti and 2070 vs 3070).

 

[ModEl4]

Who knows, regarding RDNA3 efficiency maybe AMD has a secret weapon up its sleeve, it may just add an RX in model number and magically gain 25% more performance/W.
jk:
https://videocardz.com/newz/amd-slaps-sapphires-wrist-new-radeon-card-is-called-rx-6700-not-6700
&
https://videocardz.com/newz/amd-rad...3dmark-timespy-up-to-10-faster-than-rx-6650xt

 

[deksman2]

I'm starting to believe, that the perf/watt is a dead end in the graphics and CPU industries. It no longer satisfies me when companies say that and obviously the growing power consumption for these has a lot to do with it. I'm looking forward for the new tech but if the power consumption is through the roof, I will literally skip buying and investing in graphics cards and CPUs for that matter.


where do you have 2x performance increase over RDNA2? AMD said 50% increase.

Actually, AMD said, 50% increase in performance per watt.

That means, that RDNA 3 would be able to perform 50% faster at same power consumption... or would give same performance as RDNA 2 at 50% lower power consumption.

That still leaves room for AMD to increase performance further... so, they COULD conceivably increase the performance to 100% via clocks and/or core counts, or other areas... and the total power consumption COULD increase to 500W, but I don't necessarily think it will - it could be less due to the chiplet nature of higher end RDNA 3 gpu's and it will also depend on whether AMD does voltage optimizations etc.

 

[ratirt]

Actually, AMD said, 50% increase in performance per watt.

That means, that RDNA 3 would be able to perform 50% faster at same power consumption... or would give same performance as RDNA 2 at 50% lower power consumption.

That still leaves room for AMD to increase performance further... so, they COULD conceivably increase the performance to 100% via clocks and/or core counts, or other areas... and the total power consumption COULD increase to 500W, but I don't necessarily think it will - it could be less due to the chiplet nature of higher end RDNA 3 gpu's and it will also depend on whether AMD does voltage optimizations etc.

Sure but they did not specify at what wattage. You can look at this that way. How many people say that Alder Lake is so efficient at 35W? Sure it is but as a desktop processor it must have some certain performance level and 35W is not gonna cut it because it will still be much slower than current desktop CPUs in the market. If you increase the wattage to 100watts the performance is higher and still efficient but not as much if you consider 35watts power cap. Yet, these can consume more than 200W to squeeze every ounce of performance.
Same goes for GPUs. We don't know exactly at what power level these measurements are based on. What if these were calculated at a power level of 150Watts-200Watts for instance which gives this 50% performance per watt. Crank up the power and it will run faster but the increase in performance may not be proportional to power increase. That is why I'm skeptic about the performance per watt metric. We don't know what the baseline is here for the power consumption. 6900XT consumes 300Watts (mine does) I wonder if the 7900XT will be 50% faster than mine 6900XT and still consume 300Watts of power.
Same goes for NVidia Ada architecture. So many rumors about these cards and how performance per watt is so much better versus previous gen and yet they still consume almost two times more according to rumors.
I only hope we won't get stiffed again with those processors and graphics cards with price, actual performance and power consumption.

 

[Valantar]

Sure but they did not specify at what wattage. You can look at this that way. How many people say that Alder Lake is so efficient at 35W? Sure it is but as a desktop processor it must have some certain performance level and 35W is not gonna cut it because it will still be much slower than current desktop CPUs in the market. If you increase the wattage to 100watts the performance is higher and still efficient but not as much if you consider 35watts power cap. Yet, these can consume more than 200W to squeeze every ounce of performance.

They didn't specify the wattage, however it must be applicable to a non-outlier actual retail SKU (so, not the horrendously inefficient 6500 XT) and a non-outlier retail 7000-seres SKU. If not, they run the risk of shareholder lawsuits or SEC fines for misleading shareholders. They did hedge their bets somewhat though: the slide says "projected perfromance/watt uplift", i.e. there's room for final hardware to not pan out in the way that is projected (but again, in order to avoid the aforementioned penalties there must be a pretty good reason for that discrepancy). FWIW, they also say ">50%", which does signal a degree of confidence. I still wouldn't take it to mean that every SKU will deliver a 50% performance uplift at the same wattage.

Actually, AMD said, 50% increase in performance per watt.

That means, that RDNA 3 would be able to perform 50% faster at same power consumption... or would give same performance as RDNA 2 at 50% lower power consumption.

That's not how percentages work. A 50% increase in performance/watt means 150% total performance/W vs. the 100% baseline of the previous generation. At the same performance, that would mean 33% lower power consumption, not 50% - that would require 200% perf/W to be true. A 50% increase in perf/W allows for iso performance at 33% less power or 50% more performance at iso power - or some mix of the two.

Some example math:
GPU A delivers 1000 performance in Benchmark X, at 200W power consumption = 5 score/W
GPU B delivers 1500 performance in Benchmark X at 200W power consumption = 7,5 score/W - a 50% increase.
GPU C matches the efficiency of GPU B, but is tuned to match the performance of GPU A at 1000: 1000 score/ 7.5 score/W = 133.33333 W, or a 33.3333% power reduction from GPU A.

Of course, voltage/frequency scaling isn't linear (nor does this take shader count, clock speed or VRAM bus width into account), so these calculations are massively oversimplified.

 

[Vayra86]

It looks like one chiplet will be with the shaders and the uncore, while the other chiplets will have the Infinity Cache.
Navi 31: 1 main chiplet called GCD and 6 supplementary chiplets with Infinity Cache.
Navi 32: 1 main chiplet called GCD and 4 supplementary chiplets with Infinity Cache.

I wonder what will the die size of these chiplets be?

View attachment 250478
3DCenter.org on Twitter: "AMD Navi 33/32/31 (updated) chip data, based on rumors & assumptions As @kopite7kimi pointed out, old info from last Oct is outdated updated: - 20% less WGP - no more double GCD for N31/N32 - 6 MCD for N31 = 384 MB IF$ - 4 MCD for N32 = 256 MB IF$ https://t.co/rj2G2gi9CU https://t.co/yDqeTTdSAT" / Twitter

Roflmao what's the exact difference between 'possibly' and 'allegedly' is my first question watching the above. Old info is outdated Myeah. Do you even half believe this nonsense? There are a lot of numbers. They're all guesstimates and they all suck at proper English, as do the majority of its following. Social media bottom level text.

 

[Mussels]

50% faster performance per watt doesnt mean 50% faster
It means 50% faster if it uses the same wattage, or the same performance with 66% the power usage.

Faster per watt is very easy to achieve in the low end parts, since many of those have bottlenecks they can improve upon (semi-related example, an APU with RDNA3 would blast previous APU's away because of DDR5's bandwidth)

 

[SAINT ENZO]

I'm starting to believe, that the perf/watt is a dead end in the graphics and CPU industries. It no longer satisfies me when companies say that and obviously the growing power consumption for these has a lot to do with it. I'm looking forward for the new tech but if the power consumption is through the roof, I will literally skip buying and investing in graphics cards and CPUs for that matter.


where do you have 2x performance increase over RDNA2? AMD said 50% increase.

50% increase theoretically at 100% more watt over last gen

 

[Valantar]

50% increase theoretically at 100% more watt over last gen

What? Where are you getting that from? So you're saying the top end RDNA3 card will be 600W+? Also, do you understand what performance/W means?

 

[Mussels]

What? Where are you getting that from? So you're saying the top end RDNA3 card will be 600W+? Also, do you understand what performance/W means?

this thread taught me that percentages and fractions are in fact, beyond a lot of people

 

[AusWolf]

this thread taught me that percentages and fractions are in fact, beyond a lot of people

Or maybe they think in terms of performance + W instead of performance / W. Maybe they know something we don't.

 

[Valantar]

this thread taught me that percentages and fractions are in fact, beyond a lot of people

Well, TBH, fractions are hell and percentages are intrinsically confusing due to their implicit (and often misunderstood) relativity. Still, one would assume a simple function like performance per watt would be pretty clear in its direct linking of performance AND power draw. But I guess you can never have it all

 

[AusWolf]

Well, TBH, fractions are hell and percentages are intrinsically confusing due to their implicit (and often misunderstood) relativity. Still, one would assume a simple function like performance per watt would be pretty clear in its direct linking of performance AND power draw. But I guess you can never have it all

One thing is for sure: if you want to impress, you have to have a 9 in your numbers and an X in your letters. Or is that only true for model naming?

 

[Valantar]

One thing is for sure: if you want to impress, you have to have a 9 in your numbers and an X in your letters. Or is that only true for model naming?

I think that's pretty much universal. Those are like typographic RGB - everything with a 9 and/or X are faster/better/stronger.


Wait, are go-faster stripes a car-oriented precursor to RGB?

 

[80-watt Hamster]

Wait, are go-faster stripes a car-oriented precursor to RGB?

Nah, I'd say that'd be "ground effects".

 

[Valantar]

Nah, I'd say that'd be "ground effects".

Those are far too recent. I'm talking old school.
Still, the lineage of "things that make your stuff better but really don't" is clearly quite a long one

 

[ravenhold]

When will GDDR7 be available? This RDNA3/RTX4000 needs it.

 

[Mussels]

When will GDDR7 be available? This RDNA3/RTX4000 needs it.

It might actually get it, it's been around for long enough
Samsung teases next-gen GDDR7 memory: insane 32Gbps (GDDR6X = 21Gbps) | TweakTown

 

[ravenhold]

It might actually get it, it's been around for long enough
Samsung teases next-gen GDDR7 memory: insane 32Gbps (GDDR6X = 21Gbps) | TweakTown

What about Intel Arc?

 

[Mussels]

What about Intel Arc?

I still suspect we'll see it in OEM prebuilts only for some time, and they're gunna be unicorns for a year or two