Geforce RTX 4090: rumor mill corrects data; RDNA3 “disappointing”?
There are new rumors about the technical data of the Geforce RTX 4090. It is based on the AD102 and there are fresh values that are slightly reduced and are closer to what Navi 31 supposedly has to offer.
The data for the Geforce RTX 4090, based on the AD102, has been known for some time. Corrected values are now doing the rounds, which are slightly below the previously assumed shader numbers. Accordingly, it goes down from 144 to 126 streaming units – makes 16,128 FP32 shaders instead of the previously mentioned 18,432 FP32 shaders. Of course, this would also give all the statements about performance a new meaning. In any case, the Geforce RTX 4090 would be closer to Navi 31, where 15,360 FP32 shaders are in circulation. That would be so close that the mentioned 2.7 GHz for the Geforce and 3 GHz for the Radeon could establish a certain parity – assuming the raw performance is the same.
Geforce RTX 4000: Rumors focus on Q3/2022
Many conjunctives – the rumor about the Geforce RTX 4090 comes from the otherwise well-informed “kopite7kimi” and 126 instead of 144 clusters is a value that one would not have expected after months of speculation with 144 clusters. At least 12.5 percent difference. It is now assumed that the Geforce RTX 4090 Ti gets the 144 clusters and is pushed up. In the course of this, the large gap to the Geforce RTX 4080, which is said to have 82 clusters, would also become somewhat smaller. And Nvidia would still have room for a Titan with 48 GiB memory and 900 watt TDP – both of which have also been heard before.
|AD102||144||12||72||18,432||96MiB||384 bits||24 GiB||≤ 600W|
|AD102-300||126||–||–||16.128||96MiB||384 bits||24 GiB||≤ 450W|
|AD103||84||7||48||10,752||64MiB||256 bits||16 GiB||~ 350W|
|AD104||60||5||30||7,680||48MiB||192 bits||12 GiB||≤ 300W|
|AD106||36||3||18||4,608||32MiB||128 bits||8 GiB||–|
|AD107||24||3||12||3,072||32MiB||128 bits||8 GiB||–|
The picture in the Geforce RTX 4000 portfolio would at least be a little more homogeneous. Despite the lower equipment, the card should still deliver around twice the performance of an RTX 3090. It also remains the case that he names GDDR6X with full 21 Gb/s for the top models, which means around 1 TB/s transfer capacity at 384 bits. There are also two controversial statements about this: Firstly, the Geforce RTX 4090 release notice in mid-July, which would actually be the beginning of the third quarter and not, as previously assumed, the end of the third quarter in September. From this point of view, the Geforces would really be there earlier than previously thought.
“Disappointment” at RDNA 3
What will be even more controversial is the statement about the Radeon. It says succinctly: “I am disappointed with RDNA3. That’s all.” Of course, this can be interpreted in many different ways, but it sounds very general at first. Is it lacking in raster performance? Is it lacking in ray tracing performance? Do the Radeon need more power? Speculation is allowed and fans of the camps will do so very emotionally. The Navi 31 may also be the reason why 126 streaming units are now set for an AD102-300. Nvidia might think that’s enough and have adjusted their portfolio accordingly. Maybe it’s also the later launch date of RDNA 3 – according to the latest information, it will start at the end of Q3 or beginning of Q4 2022 and probably only with Navi 31, while Navi 32 and 33 should appear next year. So AMD would only have one Radeon RX 7900/XT for the Christmas business. The swamp of speculation is getting deep here – but since “kopite7kimi” is communicative, we will surely find out.
Source: Twitter (@kopite7kimi)