It is rumored that NVIDIA’s next-generation gaming graphics card “GeForce RTX 4090” will be more than twice as fast as “RTX 3090”, and it seems that “RTX 4080” may be upgraded.
It is rumored that the NVIDIA GeForce RTX 4090 graphics card will be more than twice as fast as the RTX 3090, and the RTX 4080 will be upgraded.
The latest information was shared by Kopite 7kimi, and the specs for the NVIDIA GeForce RTX 4090 graphics card are the same as previously posted, but the overall performance of the card has improved tremendously, doubling the existing RTX 3090 giant. It states that it is over.
This performance is pure rasterization, so the overall performance in the game should be different, but it’s still a big leap compared to what the previous generation offered.
Also, NVIDIA is talking regarding cards that are expected to feature a TDP of up to 450-500W out of the box, which means they are taking advantage of the rise in TDP.
I mean RTX 4090>2x RTX 3090. ????
– kopite7kimi (@ kopite7kimi) June 6, 2022
More importantly, the NVIDIA GeForce RTX 4090 is said to feature a cut-down AD102 “Ada Lovelace” GPU core, which is a full variation, perhaps the Titan or RTX 4090 Ti ends up with much higher performance than the RTX 3090. There is a possibility.
Considering the CUDA core of 70% or more than the RTX 3090 Ti, and the clock speed that is rumored to be significantly improved over the current generation due to the migration to the TSMC 4N process node, this increase should be expected.
And while the L2 cache on existing flagship GPUs is only 6MB, the new architecture, which includes a rich L2 cache of 96MB, can provide significant performance gains.
In addition to the GeForce RTX 4090, Kopite7kimi says that the NVIDIA GeForce RTX 4080 will be upgraded.
Previously, it was rumored that the graphics card would carry a 350W TBP, but now it has been changed to 420W, which has a higher board power number than the RTX 3090. The GeForce RTX 4080 still seems to utilize the AD103 GPU, which is a slightly optimized version of the AD104 GPU.
This means a TDP 100W higher than the existing RTX 3080, so this card might end up with really fast specs.
Possible RTX 4080, PG139-SKU360, has a 420W TGP, still uses AD103.
– kopite7kimi (@ kopite7kimi) June 6, 2022
Also, the TGP of the NVIDIA GeForce RTX 4060 that I mentioned the other day is 220W, which is also much higher than the previous model.
So, looking at the entire lineup, you can expect Ada Lovelace’s power numbers to be ridiculous, but at the same time, you can expect large-scale performance improvements.
“Expected” specs for NVIDIA GeForce RTX 4090 Ti
The NVIDIA GeForce RTX 4090 Ti will have a full fat configuration with all 144 SMs enabled and a total of 18432 CUDA cores. With 96MB of L2 cache and a total of 384 ROPs, it’s a madness.
The clock has not been finalized yet, but due to the TSMC 4N process, a clock of 2.0 to 3.0 GHz is expected.
In terms of memory specs, the GeForce RTX 4090 Ti is expected to rock the capacity of the 24GB GDDR6X, which may come at faster 24Gbps speeds with a 384-bit bus interface.
This will provide up to 1.152 TB / s of bandwidth. With these improvements in specs, power consumption is also high, and this flagship model is expected to operate at a TBP of approximately 600W.
For 600W, one 16-pin Gen 5 connector is enough, but AIB doesn’t always fit within specifications, so most custom models will definitely use dual Gen 5 connectors, and a few. Even with factory overclocking, TBP is over 600W, which is the limit of a single Gen 5 power connector.
Allegations of NVIDIA GeForce RTX 4090 Ti heatsinks and cooler shrouds have also been identified, implying the use of rugged cold plates that cover both the GPU and memory dies with an overall large structure.
The leaked cooler is a Founders Edition design, and given its size, the AIB model will be significantly larger, with quad slot designs from all partners.
NVIDIA GeForce RTX 4080 “expected” specs
The NVIDIA GeForce RTX 4080 will have the same core as the existing GA102, with a higher frequency tuned cutdown configuration, or between 9000 and 1000 cores in the future with a “Ti” variation in a full fat configuration. It is expected that there will be room left.
The GPU has 64MB of L2 cache and up to 224 ROPs, which is a madness.
The clock has not yet been determined, but given the adoption of the TSMC 4N process, a clock of 2.0-3.0 GHz is expected.
The higher clock than usual is that NVIDIA is doing a two node jump, considering that the Ampere GPU with Samsung’s 8nm node was actually a 10nm process node with some optimizations. Derived from.
NVIDIA is skipping 7nm and going straight to the 5nm node, trying to make it an optimized version, not even a vanilla variant.
With TSMC’s 16nm node Pascal, NVIDIA has achieved a significant frequency leap, and we can expect a similar leap this time as well.
In terms of memory specifications, the GeForce RTX 4080 is expected to feature a 16GB GDDR6X and operate at a speed of 21Gbps with a 256-bit bus interface.
This will provide up to 672GB / s of bandwidth. With this enhancement in specifications, power consumption will also increase, and this flagship model is expected to operate at a TBP of approximately 420W.
For the 420W, one 16-pin Gen 5 connector should suffice for both the reference and custom models.
Regarding the release, the lineup of graphics cards is scheduled for the third quarter to the fourth quarter of 2022, and a more detailed schedule can be confirmed here.
NVIDIA GeForce RTX 40 Series Graphic Card Lineup (Rumor):
Explanation:
Rumors that turn around
Other sources predict that the RTX4080 will be a cutdown for the AD102, but wccftech seems to have information that the RTX4080 will be a cutdown model for the AD103.
In this specification, AD103 is expected to have 10752 CUDA cores, which means that RTX4080 has a slightly smaller number of CUDA cores.
With this, I think that the performance difference with Titan A and RTX4090 / Ti that adopt AD102 will be too large, but nVidia does not intend to sell so much AD102?
Even if the RTX4090 is between 200,000 and 300,000 yen, I don’t think it will sell that much.
For the general public, it is a convincing theory if you want to sell the RTX 4080 equipped with the AD103 with a small die size.
If you don’t do that, you may not be able to drastically reduce costs.
However, RTX4080 is also written as over 420W, so from the conventional wisdom, it is certain that it will be a GPU that eats power foolishly.
You can think of it as the same as RTX3090Ti.
Another uncertain part is the game performance of the Ada Lovelace series.
Compared to Turing, Ampere had considerably lower game performance for FP32 computing performance, but as rumored in the streets, if the game engine can not use up the increased number of CUDA cores, the game performance will be It will not increase unless it corresponds to the increased number of CUDA cores, but if there is a factor in another place, the game performance will increase explosively.
This time, the large-capacity L2 cache and the architecture itself have been revamped, so if there is a problem with either of them and it becomes a bottleneck, the game performance will explode.
I think it’s too strange to say that the game performance is regarding the same as the RX6900XT, which has an FP32 calculation performance of regarding 60-70%.
By the way, it is not possible to know how much performance it will actually be until the real thing comes out.
Those who are used for rendering and encoding seem to have demonstrated their abilities in line with the FP32 computing performance, so I don’t think there will be much change.
nVidia RTX 3000 series GPU
RTX3090
ASUSTek
¥267,800 (2022/06/07 18:51:49 As of Amazon research-detailed)
RTX3080Ti
ZOTAC
¥209,800 (2022/06/08 03:45:04 Amazon research-detailed)
RTX3080 10GB LHR
ZOTAC
¥127,778 (2022/06/08 03:49:28 As of Amazon research-detailed)
RTX3070Ti
ASUSTek
¥110,000 (2022/06/08 02:16:23 As of Amazon research-detailed)
Geforce RTX3070 8GB GDDR6 LHR
ZOTAC
¥125,244 (2022/06/08 03:45:05 Amazon research-detailed)
ZOTAC
¥75,422 (2022/06/08 03:45:25 As of Amazon research-detailed)
RTX3060Ti 8GB GDDR6
ZOTAC
¥99,860 (2022/06/08 04:10:16 As of Amazon research-detailed)
RTX3060 12GB GDDR6
ZOTAC
¥64,150 (2022/06/08 10:26:26 As of Amazon research-detailed)
RTX2000 series
RTX2060 12GB
RTX2060 is reprinted with 12GB memory
ZOTAC
¥55,980 (2022/06/08 04:07:03 As of Amazon research-detailed)
nVidia Geforce RTX 1600 series
GTX1650SUPER
ASUS
¥41,480 (2022/06/08 04:07:04 Amazon research-detailed)
GTX1650 GDDR6
ZOTAC
¥29,273 (2022/06/08 03:45:25 As of Amazon research-detailed)