Alleged images of NVIDIA GeForce RTX 4090 Ti graphics card with full-fat AD102 GPU have been captured Tyrannosaurus 55.
NVIDIA’s full-fat GA102 GPU powered GeForce RTX 4090 Ti graphics card pictured, large triple fan cooler for 600W+ Powerhouse
The card pictured is said to have an NVIDIA full-fat AD102 GPU based on the Ada Lovelace chip architecture. A full-fat chip may have a TBP of 600W (+), so it will require some robust cooling solution such as the one shown here.
According to the leaker, this appears to be an early engineering sample to test the full limits of the huge AD102 GPU that will hit the consumer market within a few months. The final variant of the Founders Edition design is expected to use a very different cooler, but this seems to confirm that the AD102 will be the hard chip for AIB cooling.
Most likely an engineering example, the reference version shouldn’t be designed to be that simple.
— Greymon55 (@greymon55) September 5, 2022
This isn’t the first time we’ve heard of a triple-fan cooler for the reference AD102 board. Not long ago, Kopite7kimi pointed out that the triple fan and 3-slot cooler were made purely for the AD102 GPU, and this leak seems to confirm that.
At the very least, they designed a three-fan cooler for the AD102’s reference board. https://t.co/vQb2KnpE4y
— kopite7kimi (@kopite7kimi) June 9, 2022
As for the card itself, both in size and width, it appears very chunky. The triple fan solution is based on an axial vane design with a large heatsink under the shroud. The shield itself is pretty simple, with a matte black texture, and it looks like the card has a 3-slot form factor. NVIDIA has always relied on a 16-pin connector, so if this is indeed a 600W variant, a single Gen5 connector should be enough to power it, unless the ES design has two of these.
NVIDIA GeForce RTX 4090 Ti “expected” specs
The NVIDIA GeForce RTX 4090 Ti will be a full-fat configuration with all 144 SMs enabled for a total of 18432 CUDA cores. The GPU will come with 96 MB of L2 cache and a total of 384 ROPs, which is insane. Clock speeds are yet to be determined, but considering the TSMC 4N process is being used, we expect clocks in the 2.0-3.0 GHz range.
As for memory specs, the GeForce RTX 4090 Ti is expected to have 24 GB of GDDR6X capacity, which will likely be available at a faster 24 Gbps over a 384-bit bus interface. This will provide up to 1.152 TB/s of bandwidth. Now, all these boosted specs will also lead to higher power consumption, with the flagship expected to run at a TBP of around 600W. Now for 600W a single 16 pin gen 5 connector should be enough, but most custom variants will definitely end up using dual gen 5 connectors as the AIB doesn’t necessarily meet specs and even the slightest factory overclock will push Above TBP 600W is the limit for a single Gen 5 power connector.
We’ve also seen so-called NVIDIA GeForce RTX 4090 Ti radiator and cooler shrouds, which hint at the use of a more powerful cold plate that can cover both the GPU and memory chips and a larger overall structure. The previously leaked cooler seems more reminiscent of the final FE or Founders Edition design than the new leak which is more of an internal test design and from the looks of it the AIB model will end up being bigger and we Maybe even end up with a four-slot design for all partners.
NVIDIA GeForce RTX 4090 Ti and RTX 4090 “preliminary” specs:
|Graphics card name||Nvidia GeForce RTX 4090 Ti||Nvidia GeForce RTX 4090||Nvidia GeForce RTX 3090 Ti||Nvidia GeForce RTX 3090|
|GPU name||Ada Lovelace AD102-350?||Ada Lovelace AD102-300?||Ampere GA102-350||Ampere GA102-300|
|process node||TSMC 4N||TSMC 4N||Samsung 8nm||Samsung 8nm|
|transistor||pending||pending||28 billion||28 billion|
|TMU/ROP||pending / 384||pending / 384||336 / 112||328/112|
|Tensor/RT Core||pending / pending||pending / pending||336 / 84||328/82|
|base clock||pending||pending||1560 MHz||1400 MHz|
|boost clock||~2800 MHz||~2600 MHz||1860 MHz||1700 MHz|
|FP32 computing||~103 TFLOPs||~90 TFLOPs||40 TFLOPs||36 TFLOPs|
|RT TFLOPs||pending||pending||74 TFLOPs||69 TFLOPs|
|Tensor-TOP||pending||pending||320 tops||285 likes|
|Memory Capacity||24GB GDDR6X||24GB GDDR6X||24GB GDDR6X||24GB GDDR6X|
|memory bus||384 bit||384 bit||384 bit||384 bit|
|bandwidth||1152 GB/sec||1008 GB/sec||1008 GB/sec||936Gbps|
|Three Gorges Project||600W||450W||450W||350W|
|start (availability)||2023?||October 2022?||March 29, 2022||September 24, 2020|