Let’s now look at the winners and losers in the two tested resolutions and when comparing DLSS, FSR and XeSS and look at the percentage differences. All FPS were normalized, cumulated, and then averaged together. I have to remind you again that the Radeon RX 6950XT Gaming X Trio I used is the only card that is not a reference card, as AMD could not provide it. MSI’s custom model is about 5% faster than the reference, but with significantly higher power consumption.
Summary of the gaming performance in FPS and with the P1 Low
QHD with 2560 x 1440 pixels
In QHD, the GeForce RTX 4090 is plenty of 50 percentage points faster than the GeForce RTX 3090, the comparison is interesting in that both cards have been trimmed very similarly. Therefore, the use of the GeForce RTX 3090 Ti as a 100% mark is also prohibited because it is the full upgrade that was only added very late. The (factory overclocked) Radeon RX 6950XT is about as fast as the RTX 3090, while the rest, as already known, has to rank further behind.
The lead in the min FPS (P1 Low) is also very clear, but a bit smaller, which was demonstrably due to the already starting CPU limitation (!) of the Ryzen 9 7950X in combination with the extremely fast GeForce RTX 4090.
Ultra HD with 3080 x 2160 pixels
In ultra HD, the GeForce RTX 4090 is then already almost 64 percentage points ahead of the GeForce RTX 3090 and almost 70 percentage points ahead of the overclocked RX 6950XT. You can’t publicly call that humiliation, but you can at least think it. Even the old full GeForce RTX 3090 Ti is caught cold with over 54 percentage points.
Since the CPU only limits the GeForce RTX 4090 very sporadically, the card’s lead in the P1 also returns to normal. With a lead of almost 60 percentage points over the RTX 3090 and almost 66 percentage points over the Radeon RX 6950XT, everything has actually already been said. There’s no need to go on about it verbally for ages, it’s just the way it is.
Ultra HD with 3080 x 2160 pixels and DLSS, FSR as well as XeSS
If you use pixel accelerators like DLSS, FSR or XeSS, the frame rates of the cards partly increase again to very playable regions, but the gap is once again downright astounding. Almost 61 percentage points advantage of the GeForce RTX 4090 over the RTX 3090 is a real slap in the face for Ampere. Here, too, the RX 6950XT gets a real snub, because the almost 71 percentage points gap is almost unbelievable. Apart from that, the picture quality in FSR should actually get a few points deducted. Intel’s XeSS looks a bit better, but also brings a bit less performance. But on this very point, I refer you to the follow-up tomorrow and the day after tomorrow, where we will offer countless video comparisons and supplementary benchmarks. You can write a lot about it, but you have to see it.
The P1, i.e. the min FPS, is again within the CPU limit for the GeForce RTX 4090 and I cannot think of a CPU that could supply this card with data without drawbacks. This is also (unfortunately) a statement.
Efficiency evaluation
Let’s now move on to the efficiency evaluation, as the amount of watts needed to achieve a certain gaming performance in terms of an FPS, i.e. the watts per FPS metric. This is the opposite of what the manufacturers use, but I think it is much easier to use the larger numbers before the decimal point. And because I don’t want to force you to use a calculator, I’m now including the percentage points as a calculation service. Let’s start in QHD. The efficiency of the GeForce RTX 4090 increases by a staggering 70 percentage points compared to the GeForce RTX 3090! The efficiency advantage over the (factory overclocked) Radeon RX 6950XT is even 89 (!) percentage points. Overtaking without catching up? AMD will almost have to work some magic here, but as we all know, hope dies last.
And what happens now in Ultra HD? The new GeForce RTX 4090 is still 60 percentage points ahead of the GeForce RTX 3090, because power comes from fuel and in ultra HD the card is less limited by the CPU. Free ride, then. However, the efficiency even increases to a depressing 100 percentage points compared to the Radeon RX 6950XT, which is somewhat weaker in UHD! This is almost an energetic execution.
Various processes can also be used to make Ultra HD more playable, but how does this affect efficiency? the GeForce RTX 4090 takes 72 percentage points from the RTX 3090 thanks to the new Tensor cores, and I would have preferred to sweep the 115 percentage point lead over the Radeon RX 6950XT under the carpet with a red head. We can only pray that AMD has come up with an adequate solution and that we will know more soon.
Summary and conclusion
Even though today’s article is just the beginning and we will deliver interesting results and evaluations with further follow-ups (thanks here to Fritz Hunter, who has been doing nothing else for days!), the first picture is surprisingly positive. Because NVIDIA offers a new generation for a long time, which offers between 50 and 60 percentage points more gaming performance with similar or even lower consumption compared to the predecessor! This is not only an evolution, but a real revolution, monolithic chip or not.
But it’s also worth looking at the overall package and leaving the price aside for now. Besides the almost striking performance increase and the outstanding efficiency (in the context of the gaming performance provided), Ada namely offers much more than just an increased raster performance in the usual pixel orgies! The entire feature set of extremely increased ray tracing performance, DLSS 3.0 and Reflex is accompanied by other hardware solutions like the dual video encoder (NvEnc), which can even take on parallel tasks. Simultaneous streaming and recording are only one facet, because the overall increased computing power of the GeForce RTX 4090 including the Tensor cores will also be very much appreciated in productive use.
Blender is just one example of many, as many of the standard software’s CUDA-based plugins benefit immensely from the new performance. Here, too, there are many interesting topics that just call for a retest. There will be more details about DLSS 3.0, a high-performance continuation of DLSS based on the new architecture, tomorrow, because it also shows in which direction it could still go and that you can certainly still achieve decent results with smaller cards and significantly less energy in the end.
As a reviewer, you are of course obliged to test and judge emotionlessly and objectively. But in view of such an explosion in performance and efficiency (which outsiders wouldn’t expect), it’s fair to show something like enthusiasm. With the nearly 2000 Euro MSRP for the so-called “MSRP cards”, which every board partner has to deliver, we are in a price league that is rather unaffordable for most buyers, but it definitely makes fun for more, when smaller cards will follow.
What will matter most is how the cards will scale down. The already announced GeForce 4080 16GB with only 60% of the shaders would have a theoretical gap of around 40 percentage points, assuming the same to slightly higher boost clock. This is plausible insofar as the power consumption also drops by around 40 percentage points. You would then still be about 10 percentage points ahead of a GeForce RTX 3090 Ti and about 20 percentage points ahead of the GeForce RTX 3090 and about 30 percentage points ahead of the overclocked RX 6950 XT in ultra HD. However, if you use the 50% increase of Perf/W reported by AMD, the race is open again. But in November we will finally know more.
Until then, we can ponder why NVIDIA leaves such a huge gap between the brute GeForce RTX 4090 24GB and the 60% solution in the form of the GeForce RTX 4080 16GB. Because there would really be more than enough room for a GeForce RTX 4080 Ti. But today and here, it’s all about the GeForce RTX 4090, which I introduced to you in the form of the NVIDIA GeForce RTX 4090 FE 24GB. And this card is really a very efficient pixel breaker. There is no need to write anything more about it today.
As a preview of what’s to come in the following days, I can already ante up a very comprehensive look at DLSS 3.0, the latencies and Reflex with a lot of video comparisons, as well as the conversion instructions to a water block and of course UV tests to find the sweet spot of the card. Overclocking is also possible, but we will wait and see if there are new BIOSes or drivers. That’s a completely different story again. For now, let’s hope for a wide and good availability of the new card. And just between you and me, there are certainly some bargains to be had with the current Ampere cards, if you don’t want to wait for the new, smaller cards.
- 1 - Introduction, technical data and technology
- 2 - Test system in igor'sLAB MIFCOM-PC
- 3 - Teardown: PCB, components and cooler
- 4 - Gaming Performance WQHD (2560 x 1440 Pixels)
- 5 - Gaming Performance UHD (3840 x 2160 Pixels)
- 6 - Gaming Performance UHD + DLSS/FSR/XeSS (3840 x 2160 Pixels)
- 7 - DLSS 3.0 and the longest bars
- 8 - NVIDIA Reflex and Latency
- 9 - Workstation graphics and rendering
- 10 - Power consumption and load sharing
- 11 - Load peaks, capping and power supply recommendation
- 12 - Temperatures, clock rate, OC, fans and noise
- 13 - Summary and Conclusion
545 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Veteran
Urgestein
Mitglied
Veteran
Urgestein
Veteran
Urgestein
Urgestein
Mitglied
Mitglied
Veteran
Alle Kommentare lesen unter igor´sLAB Community →