News AMD Backs Off RDNA 3 Efficiency Comparison to RTX 4090

[Admin]

AMD reportedly hid a performance per watt slide at the very last moment, relating to its RX 7900 XTX/XT coverage on November 15th. AMD's reasoning is unknown, but we suspect its related to Nvidia's RTX 4080, and excellent power efficiency.

AMD Backs Off RDNA 3 Efficiency Comparison to RTX 4090 : Read more

 

[Luneder]

Does not matter when the 4080 is competing with the 4090 for the most expensive card of the year.
The choices are go for the RX 7900 which is much cheaper with performance that will likely be similar to the 4080 but have a bit higher power use or throw power consumption out the window and go for a 4090 with a 30-40% performance increase for a similar price as the 4080.

I've been keeping track of ibuypower and cyberpower custom builds and there is only a few hundred dollar difference between a 4080 and a 4090 PC on builds that are all over $3k except for the i7 versions that are just under $3k.

 

[Math Geek]

makes sense. if the 7900 will perform similar to the 4080 (blatant guess based on rumor mill) and use similar power, then obviously there is no efficiency gap to brag about.

even the extra power used by the 4090 is still pretty good on a perf/watt scale compared to the 4080/7900 .

i'd stick with the "our card won't melt the connections and possible ruin your psu and $1600-2000 gpu" angle for my press at this point

 

[Luneder]

makes sense. if the 7900 will perform similar to the 4080 (blatant guess based on rumor mill) and use similar power, then obviously there is no efficiency gap to brag about.

even the extra power used by the 4090 is still pretty good on a perf/watt scale compared to the 4080/7900 .

i'd stick with the "our card won't melt the connections and possible ruin your psu and $1600-2000 gpu" angle for my press at this point

they've pretty much solved the connector thing, only 50 cards out of 124,000 sold had the problem (0.04% of cards) and the 50 cases were all because they didn't have the connector seated in all the way. the fix is pretty simple either they add a better clip or they shorten the sensor pins so the cable won't send power if the connector is not seated all the way, the current sensor pins are too long and giving a false positive on being seated in all the way.

 

[nimbulan]

makes sense. if the 7900 will perform similar to the 4080 (blatant guess based on rumor mill) and use similar power, then obviously there is no efficiency gap to brag about.

even the extra power used by the 4090 is still pretty good on a perf/watt scale compared to the 4080/7900 .

i'd stick with the "our card won't melt the connections and possible ruin your psu and $1600-2000 gpu" angle for my press at this point

Yeah based on the few performance numbers AMD's posted for the 7900 XT, compared to independent reviews of the 4080, it puts them basically equal in rasterization performance. The 7900 XTX looks around 15-20% faster than that in rasterization, but the 4080 will still beat it by 30% in raytracing.

My guess is that AMD was expecting nVidia's cards to hit their TGP limit under most workloads, as has been the case in the past, and prepared a slide with comparisons based on that. But with the 40 series frequently using significantly less power than the TGP limit (and to be clear I'm not expecting the same to be true for AMD,) that will likely make nVidia's cards more power efficient this gen.

 

[Alvar "Miles" Udell]

To borrow an AMD phrase, did nVidia just "jebait" AMD into showing their hand?

 

[criticaloftom]

All I know as a consumer is that with a 'hyper' expensive purchase either way; I won't be trusting Nvidia.
The press about how many inserts the plug is rated for; and the mere fact that they have any that create a system fire means they have done their dash.
Maybe in a few years consumers will forget, but for now that connector is a dead selling point.

 

[missingxtension]

Soo about the RYX 4080....

 

[russell_john]

AMD reportedly hid a performance per watt slide at the very last moment, relating to its RX 7900 XTX/XT coverage on November 15th. AMD's reasoning is unknown, but we suspect its related to Nvidia's RTX 4080, and excellent power efficiency.

AMD Backs Off RDNA 3 Efficiency Comparison to RTX 4090 : Read more

Architecture only gives you small gains in efficiency usually in single digit percentages. In the last 20 year nearly all the efficiency gains have come from using a smaller node size and since the 4000 and 7000 series cards are on essentially the same node they are likely to have very similar performance per watt. The edge will likely go to AMD since they don't have the additional Tensor Cores which add overhead even when they aren't being used. However when you look at performance per watt for Ray Tracing Nvidia is still likely to blow AMD away although they may whittle down the 23% efficiency edge Nvidia had last generation but I suspect not by much

 

[russell_john]

Soo about the RYX 4080....

Yes proofreading seems to be a lost art .... Too much reliance on automatic proofreading that can't catch errors like that one

 

[russell_john]

All I know as a consumer is that with a 'hyper' expensive purchase either way; I won't be trusting Nvidia.
The press about how many inserts the plug is rated for; and the mere fact that they have any that create a system fire means they have done their dash.
Maybe in a few years consumers will forget, but for now that connector is a dead selling point.

You better get used to it because that connector is going to be on motherboards eventually (In fact that was what Intel originally designed it for) since it is part of the ATX 3.0 standard. Intel's intention with ATX 3.0 is to get rid of the 24 pin motherboard connector, and the 12V Molex connector and replace it with this single 12V connector and have all the voltage conversion/regulation for the other voltages on the motherboard itself. You'll basically have two of these 16 pin connectors coming from the power supply, one for the motherboard and another for the graphics cards.

What is throwing everyone for a loop is Intel has been using this connector on server motherboards for almost two years without issue but then again those were installed by professionals

BTW you can't start a fire without a flame and the composite plastic material used for those connectors are flameproof .... Melting DOES NOT equal fire

 

[zecoeco]

RTX 4090 is untouchable in terms of performance, but AMD will still be ahead in efficiency because it's the balanced approach they've decided to follow.

 

[gruffi]

AMD already made it clear. There are no comparisons to RTX 4080 because it wasn't available back then and AMD didn't get a sample from Nvidia for comparisons. All AMD could do was some comparisons to RTX 4090. But as the author correctly recognized, Navi 31 is targeted at AD103, not AD102. AD102 is a much larger monolithic design and much more expensive than AD103 or the chiplet based Navi 31 design.

 

[Warrior24_7]

AMD reportedly hid a performance per watt slide at the very last moment, relating to its RX 7900 XTX/XT coverage on November 15th. AMD's reasoning is unknown, but we suspect its related to Nvidia's RTX 4080, and excellent power efficiency.

AMD Backs Off RDNA 3 Efficiency Comparison to RTX 4090 : Read more

The reasoning is known. They require the same power requirements as the 4090, and are LESS powerful! They want you to be stupid enough to buy the card first, and realize that later!

 

[King_V]

In fact, according to our tests, the RTX 4080 Founders Edition consumed just 221W at 1080p.

That is a bit misleading, isn't it? After all, according to the very test article that links to:

That's especially true of our 1080p results, where the average power use for the card was just 221W. With a faster CPU like a Core i9-13900K or Ryzen 9 7950X, we'd likely see higher power use from the 4080, though it's likely even an overclocked 13900K wouldn't max out its power limit at 1080p.

Basically, this is saying that the 4080 uses just 221W for gaming when it's forced to sit there waiting for the CPU to provide it with enough data.

Don't get me wrong, given that Nvidia officially rates the card at 320W, the fact that, in the Additional Power, Clock, and Temperature Testing table, in those cases where the RTX 4080 is pushed to its maximum (at 99.0%) independent of resolution, it will use as little as 277.6W (Horizon Zero Dawn, 4K Ultimate), or as much as 308.8 (Bright Memory Infinite, 4K Very High )

side note: kind of interesting that its highest power consumption AND lowest power consumption, when at full utilization, were both at 4K.

Still impressive, don't get me wrong, relative to what they stated as the official TDP. But, unless I'm missing something, to say that 221W usage is "impressive efficiency" for the card isn't really telling us much when we're looking at the CPU being the limiting factor... it's not telling us about performance-per-watt.


Alternately, to compare apples to apples, I'd say that other competing cards, whether from AMD, Intel, or even other 4000-series Nvidia models, would have to have a similar detailed table to truly be able to compare. Or, a "this is the Geomean across our tests at 1080p, 1440p, and 4K" as comparisons.

 

[JarredWaltonGPU]

That is a bit misleading, isn't it? After all, according to the very test article that links to:



Basically, this is saying that the 4080 uses just 221W for gaming when it's forced to sit there waiting for the CPU to provide it with enough data.

Don't get me wrong, given that Nvidia officially rates the card at 320W, the fact that, in the Additional Power, Clock, and Temperature Testing table, in those cases where the RTX 4080 is pushed to its maximum (at 99.0%) independent of resolution, it will use as little as 277.6W (Horizon Zero Dawn, 4K Ultimate), or as much as 308.8 (Bright Memory Infinite, 4K Very High )

side note: kind of interesting that its highest power consumption AND lowest power consumption, when at full utilization, were both at 4K.

Still impressive, don't get me wrong, relative to what they stated as the official TDP. But, unless I'm missing something, to say that 221W usage is "impressive efficiency" for the card isn't really telling us much when we're looking at the CPU being the limiting factor... it's not telling us about performance-per-watt.


Alternately, to compare apples to apples, I'd say that other competing cards, whether from AMD, Intel, or even other 4000-series Nvidia models, would have to have a similar detailed table to truly be able to compare. Or, a "this is the Geomean across our tests at 1080p, 1440p, and 4K" as comparisons.

The point wasn't that the 4090 uses 221W, but that it also doesn't generally use the listed 450W either. If you look a the Gigabyte RTX 4090 review's power page, or the MSI RTX 4090 review's power page, both have full data on power use under most of the tests. (RDR2 doesn't because that game crashes if you try to use FrameView with it.) Here's a better view of the GB results from Excel, which doesn't munge the tables like our CMS:



Across all 13 games, it averaged just under 400W at 4K and under 350W at 1440p. Only in DXR testing did it get pretty close to the 450W figure, where it averaged 432W. Except it's a safe bet that the 4090 will be a lot faster than the 7900 XTX in demanding DXR tests, which would give Nvidia the win on performance per watt. If you only look at non-DXR games, at 1440p the 4090 averaged 311W (geometric mean of 303.7W) and at 4K it averaged 376W (geomean of 373.8W). So that's 75W lower than the rated maximum TBP and any assumption AMD might have made about power going into the testing probably proved incorrect once it had actual data.

From my reviews and testing, using the seven non-DXR games and geometric means of average FPS and average power:

RTX 4080 perf-per-watt (PPW) at 1440p: 0.526 FPS/W
RTX 4080 perf-per-watt (PPW) at 4K: 0.287 FPS/W

RTX 4090 perf-per-watt (PPW) at 1440p: 0.494 FPS/W
RTX 4090 perf-per-watt (PPW) at 4K: 0.319 FPS/W

So based on those results, the 4090 was more efficient than the 4080 at 4K ultra settings! Sadly, I don't have full matching data for other GPUs (YET!), but I'll definitely be capturing that on future testing. I'm going to be retesting everything on a new 13900K testbed in the coming weeks/months as well, using FrameView and PCAT v2 to measure real power data, and that should prove quite interesting.

 

[JarredWaltonGPU]

Actually, here's another set of data, using Spider-Man: Miles Morales testing. AMD values are "approximate" because AMD basically lies on its software power reporting. (Looking at real vs. software values, AMD's power use is anywhere from ~18 to ~35 percent higher than the software reports, as it only gives the GPU power use rather than the full board. This is why we need to use a tool like PCAT or Powenetics. But Powenetics is rather useless for capturing this on every game, so I'm hoping to use PCAT soon.)

For these PPW figures, I applied a 20% higher TBP value on AMD. It's going to be slightly high on some cards, and quite a bit low on others, but it's at least a rough starting point. And this is a forum thread precisely because I don't have the hard AMD TBP figures, so don't get bent out of shape. We know the AMD data is inaccurate, take with grains of salt, etc. Anyway, I've sorted by PPW at 1080p for these tables using ultra and ultra with maxed out DXR settings:

Spider-Man: Miles Morales	FPS/Watt	AVG FPS	GPU Clock (MHz)	GPU Temp (C)	GPU Power (Watts)	GPU Utilization
RTX 4080 1080p Very High	0.994	140.7	2820	42	141.6	69.4
RTX 4090 1080p Very High	0.989	141	2775	44.1	142.6	61.3
RX 6600 XT 1080p Very High	0.785	83.4	2668.5	51	106.2	97.8
RX 6700 XT 1080p Very High	0.708	99.4	2570.9	69.4	140.3	97.2
RX 6800 XT 1080p Very High	0.597	115.3	2436	81.8	193.1	97.3
RTX 3060 1080p Very High	0.594	79.4	1912	62.3	133.6	94.6
RX 6500 XT 1080p Very High	0.582	31.5	2711.3	50.9	54.1	98.9
RTX 3070 1080p Very High	0.564	99.3	1973.2	65.4	176.1	92.3
RX 6950 XT 1080p Very High	0.531	122.9	2633.8	62	231.6	96.4
GTX 1650 Super 1080p Very High	0.500	43.2	1920	59.4	86.4	97.1
GTX 1660 Super 1080p Very High	0.487	49.2	1925.5	64.9	101	96.7
RTX 3080 1080p Very High	0.478	124.8	1950	73.8	261.1	91.8
RTX 3090 1080p Very High	0.471	132.3	1965	62.8	280.7	90.1
RTX 2060 1080p Very High	0.453	68.9	1889.7	67.5	152.1	95.5

Spider-Man: Miles Morales	FPS/Watt	AVG FPS	GPU Clock (MHz)	GPU Temp (C)	GPU Power (Watts)	GPU Utilization
RTX 4080 DXR 1080p Very High DXR	0.640	98.3	2805	46.2	153.7	83.3
RTX 4090 DXR 1080p Very High DXR	0.575	98.6	2775	49	171.4	72.2
RTX 3060 DXR 1080p Very High DXR	0.377	49.5	1912	61.7	131.4	96.7
RX 6700 XT DXR 1080p Very High DXR	0.367	57.6	2569.5	74.7	156.8	98.7
RTX 3070 DXR 1080p Very High DXR	0.353	62	1965	66	175.4	96.2
RX 6600 XT DXR 1080p Very High DXR	0.327	29.3	2682.4	48.7	89.6	98.9
RX 6800 XT DXR 1080p Very High DXR	0.320	70.4	2437.5	80.1	219.7	97.9
RTX 3080 DXR 1080p Very High DXR	0.293	79.9	1950	74.8	272.4	95.3
RTX 3090 DXR 1080p Very High DXR	0.287	84.6	1950	67.6	294.9	94.8
RX 6950 XT DXR 1080p Very High DXR	0.283	76.9	2614.8	66.1	272.2	97.3
RTX 2060 DXR 1080p Very High DXR	0.255	35.1	1890	66.8	137.4	97.4
RX 6500 XT DXR 1080p Very High DXR	0.121	4.4	2714.5	48.7	36.3	99

Some food for thought at least. It's one game, Nvidia's RTX 4090 and 4080 are at the top of both charts. I can also tell you that they're at the top of the 1440p and 4K results as well — by quite a wide margin in some cases. Actually, I'll just post the 4K figures, with the same caveats as above about AMD's power values.

Spider-Man: Miles Morales 4K	FPS/Watt	AVG FPS	GPU Clock (MHz)	GPU Temp (C)	GPU Power (Watts)	GPU Utilization
RTX 4090	0.402	119.5	2760.0	59.4	297.2	95.0
RTX 4080	0.399	100.9	2805.0	54.7	253.0	96.2
RX 6700 XT	0.266	48.2	2540.5	77.0	181.5	98.7
RX 6600 XT	0.257	27.6	2657.2	51.2	107.6	98.9
RX 6800 XT	0.249	63.1	2391.8	77.9	253.6	98.5
RX 6950 XT	0.245	69.4	2450.8	65.0	283.0	98.6
RTX 3070	0.231	48.6	1917.8	70.3	210.0	96.6
RTX 3090	0.230	76.9	1839.8	71.6	334.1	95.8
RTX 3080	0.224	69.2	1870.0	78.7	308.4	96.4
RTX 3060	0.221	36.0	1886.8	69.2	162.9	97.3
GTX 1660 Super	0.177	18.8	1922.3	64.9	106.3	98.7
RTX 2060	0.159	23.3	1879.0	68.5	146.9	97.9

Spider-Man: Miles Morales 4K	FPS/Watt	AVG FPS	GPU Clock (MHz)	GPU Temp (C)	GPU Power (Watts)	GPU Utilization
RTX 4090 DXR	0.250	72.3	2760.0	58.7	289.4	94.9
RTX 4080 DXR	0.238	56.8	2790.0	55.7	238.6	96.8
RTX 3090 DXR	0.130	42.4	1870.2	71.0	327.4	97.3
RTX 3060 DXR	0.126	18.5	1896.9	66.1	146.7	98.3
RX 6800 XT DXR	0.126	32.1	2364.9	78.3	254.7	99.0
RX 6950 XT DXR	0.122	34.6	2417.4	66.8	283.4	98.9
RX 6700 XT DXR	0.116	19.0	2540.2	76.5	164.2	98.8
RTX 3070 DXR	0.115	20.6	1962.0	67.5	179.9	98.3
RTX 3080 DXR	0.112	32.6	1915.2	76.3	291.3	97.4

Based on just these figures, the PPW of the 4090 is 98% higher than the closest AMD card at 4K DXR, and 51% higher than the closest AMD card at 4K without DXR. If you take the RX 6950 XT values and apply a 50% improvement in PPW, the 4090 would still be 9% 'better' at 4K and 37% better at 4K DXR. I strongly suspect when AMD did the actual testing, they discovered exactly this and so dropped the slide and forgot to remove the footnote. We'll find out how it actually fares in real-world testing in December! 🙃

 

[Wassco]

AMD reportedly hid a performance per watt slide at the very last moment, relating to its RX 7900 XTX/XT coverage on November 15th. AMD's reasoning is unknown, but we suspect its related to Nvidia's RTX 4080, and excellent power efficiency.

AMD Backs Off RDNA 3 Efficiency Comparison to RTX 4090 : Read more

Who's buying a 40 series to play at 1080?

That's fkn moronic logic.

 

[gruffi]

AMD values are "approximate" because AMD basically lies on its software power reporting.

BS. They don't.

 

[Elusive Ruse]

Although I'm impressed by the power efficiency of the 4000 series, a lot of it is due to the horrible P/W of the 3090 and Ti. I wouldn't put much into gaming benchmarks for the 4090 efficiency though as it's just overkill for gaming purposes.

 

[LastStanding]

they've pretty much solved the connector thing, only 50 cards out of 124,000 sold had the problem (0.04% of cards) and the 50 cases were all because they didn't have the connector seated in all the way. the fix is pretty simple either they add a better clip or they shorten the sensor pins so the cable won't send power if the connector is not seated all the way, the current sensor pins are too long and giving a false positive on being seated in all the way.

Out that 100k+ cards, how many of those cards are in data centers, at Hollywood studios, in scalpers' hands, sitting on Ebay, etc. so, that subjective opinion is still very unknown and remember, NVIDIA has shown no FACTUAL data yet to conclude its finding AFTER GN's support video but only delivered a well-orchestrated PR stunt.

 

[Math Geek]

yah the whole "we investigated ourselves and found we did nothing wrong and we are awesome" PR nonsense is crazy.

but i guess so long as folks continue buying their defective cards, all is good.

 

[Elusive Ruse]

yah the whole "we investigated ourselves and found we did nothing wrong and we are awesome" PR nonsense is crazy.

but i guess so long as folks continue buying their defective cards, all is good.

I wouldn't be surprised if they just relied on GN's investigation without putting any effort themselves.

 

[bit_user]

the only workload we found that would max out the RTX 4080's power budget was Furmark combined with a GPU overclock.

LOL! Furmark: still gettin' it done!

 

[Math Geek]

to me the scary thing is that these connections melted down at much lower power than what we thought at first.

sure 600w is easy to imagine going all meltdown. but at the lower day to day gaming power draw, this is more alarming to me. leaves me wondering if we'll see this with the 4080 cards soon as well.