In order to keep the speculation factor as low as possible on the one hand and not to violate one of the NDAs at the same time, I have to elaborate a bit more this time. The real problem, however, was the extremely different puzzle pieces from both manufacturers in various announcements, which doesn’t exactly simplify a preview of raster performance for both cards. Different settings or games, missing details and much more are not exactly a serious invitation to publish such speculations without a bland aftertaste. Especially since I am already under NDA with the RTX 4080 and not with the RX 7900XTX. And so the entire article is still highly speculative, of course, because things could turn out completely differently at the last moment. But today’s calculation should still not be completely wrong. Let’s just be surprised later then.
We estimate a Radeon RX 7900XTX in AMD’s “Best Case”
AMD unfortunately only listed very few, hand-picked game titles in its slides at the announcement event on November 3, 2022 and also declared most of them as “Up to”. Since the ray tracing performance of the GeForce RTX 4080 16 GB, as AMD’s declared opponent, is difficult to extrapolate with a calculator, I’ll limit myself to the pure raster performance of the new card today and will only dare to make some kind of forecast about DXR at the end. By the way, the “best case” in ultra HD was Cyberpunk 2077 with a factor of 1.7 compared to the Radeon RX 6950XT, which I assume was supposed to be a reference card.
Therefore, due to the lack of a reference card, I first measured my factory overclocked MSI RX 6950XT Gaming X Trio in its out of the bpx state with a significantly higher power limit and clock. After that I wrote the reference BIOS data into the PowerPlayTables using MorePowerTool and used it together with Boardpartner card to bring this factory overclocked card back to the reference level. After that, I created the second benchmarks, which now serve as a basis for the Radeon RX 6950XT, as AMD might have used it. With the FPS values of the very different settings, it is also relatively easy to put the increased power consumption of the CPU (measured) into relation and extrapolate it for the new card. A plausibility test with the measured values for the GeForce RTX 4090 FE then confirmed this quite accurately.
AMD states a factor of 1.7 for Cyberpunk 2077, Ultra-HD and Ultra-Preset, by which the RX 7900XTX should be faster in the selected setting. I now used this multiplier to modify all recorded single frames of the RX 6950XT accordingly and to downscale the frame time appropriately. However, this is very speculative for the percentiles, because AMD unfortunately doesn’t give any information about this. But I assume that not so much should have changed here. At most, it could turn out a bit better. Always provided that AMD’s specified value of 1.7 is correct.
Overclocking the reference card? It’s going to be close!
AMD sets the Radeon RX 7900XTX to a TBP of 355 watts. Since both external 8-pin sockets allow a total of 300 watts, a maximum of 66 watts are allotted to the PCIe slot, of which 55 watts have already been estimated. This would leave a margin of only 11 watts for a possible and standard-issue overclocking with an increase of the power limit, which is rather symbolic because it turns out very meager. Anything beyond that would already violate the applicable PCI SIG standards. Of course, you can assume that the TBP was originally much lower and that AMD only drilled out the TBP later and took the customer’s possible overclocking leeway into the factory clock.
The 11 watts up to the standard limit then serve more as a buffer for the measurements, because the heated voltage transformers usually work more inefficiently. At least AMD paid attention this time and did not repeat the mistake of the Radeon RX 480. I had already shown the symbol graphic of a board partner card with three 8-pin sockets in a similar way (as well as added a real AIB overlay today, which proves how correct my estimate was). The already published render graphics of the board partner cards (e.g. Asus) fully confirm my graphics, also regarding the card’s equipment and the (theoretically) maximum possible power limit of now 516 watts. In practice, this will probably remain more of a lecherous thought experiment, because it will hardly be possible to implement it realistically and still justifiably in performance.
I can include a small side note: the AIB cards will use the same boards for the Radeon RX 7900XTX and RX 7900XT. Depending on the AIB, the smaller card may have to do without one or the other voltage transformer, but for some AIBs it may not. This makes the availability of suitable water blocks much easier for the manufacturers of the accessory coolers. And what exactly is the composition of the 355 watts? This can also be roughly extrapolated if you take the usual (and thus known components) as a basis and also include the voltage converters analogous to the card’s power dissipation. These values are also purely hypothetical and not set in stone, but they do provide a certain overview:
GPU (incl. memory controller) | 240 watt |
24 GB GDDR6 (12 modules á 2 GB) | 30 watt |
Voltage transformer losses GPU (DRMOS, coils) | 30 watt |
Voltage transformer losses memory (DRMOS, coils) | 10 watt |
Board losses | 17 watt |
Other (extra-low voltages, fan, MCU, LED, etc.) | 28 watt |
Total Board Power (TBP) | 355 watt |
The GeForce RTX 4080 FE 16 GB would need a different consideration…
Here, unfortunately, the situation is completely different. I already wrote: NDA. However, from the benchmarks already published by NVIDIA itself, we can deduce that the raster performance in Cyberpunk 2077 should be about 15% to 20% above the GeForce RTX 3090 Ti FE and still about 10% above the very highly overclocked board partner card in the form of the MSI GeForce RTX 3090 Ti SUPRIM X tested here. This is the only clue that can already be derived from the data found in the network. Therefore, the highly overclocked RTX 3090 Ti, which is presumably about 10% slower, is representative for the GeForce RTX 4080 FE, which will be launched on November 16, 2022. You’ll just have to figure out the rest for yourself. Or wait, even if it’s hard. Stupidly, my estimate was just too accurate to not pass as a suspected NDA violation.
Test setup
By popular demand, I’m going for a closed system in the form of a PC specially assembled for igor’sLAB by MIFCOM (exemplary product link, not an affiliate) this time. The case used is really good, has enough internal volume and also an excellent airflow. The used AiO compact water cooling in the form of Be Quiet’s Silen Loop II visually matches the GeForce RTX 4090 FE and the RGB accents of the RAM and motherboard can either be set to plain white or turned off completely in case of doubt. When things get too colorful again.
I deliberately did without all the bling-bling in my configuration, but use a current AMD CPU and a solid motherboard for the potent underpinnings. Here are a few pictures of my test system from the lab, including the NVIDIA RTX 4090 Founders Edition 24GB. The PC was assembled nice and neat by MIFCOM (I’m always like that with the cables) and can be purchased in this configuration (see below) exactly the same or modified at the dealer. It was simply important to me here to use a commercially available solution and not my extremely chilled lab hardware in the climate oasis.
There is enough space between the side panel (glass) and the graphics card and the storage space of the XL case was not even close to being used. Thus, even all other tested cards still have enough room to breathe. So that should fit. I also only need two fans in the front.

I have also summarized the individual components of the test system in a table:
101 Antworten
Kommentar
Lade neue Kommentare
Veteran
Veteran
Urgestein
1
Urgestein
Mitglied
Urgestein
Urgestein
Urgestein
Mitglied
1
Urgestein
Urgestein
1
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →