New rumors surrounding the NVIDIA GeForce RTX 40 series graphics card lineup which includes the GeForce RTX 4090, RTX 4080, RTX 4070 & a 900W flagship have been posted by Kopite7kimi. The rumors further detail the configurations we are going to see in the upcoming lineup along with a few prototype cards that may or may not make it to retail shelves.
NVIDIA GeForce RTX 40 Graphics Card Rumors: Top AD102 GPU With 900W TDP & 48 GB Memory, RTX 4090 24 GB WIth AD102, RTX 4080 16 GB With AD103, RTX 4070 12 GB With AD104
According to the leaker, currently, NVIDIA has only started testing the AD102 GPUs which will go on to power the likes of the GeForce RTX 4090 Ti and GeForce RTX 4090. The NVIDIA GeForce RTX *090 graphics cards are already positioned as BFGPUs and have replaced the Titan-class offerings. There still are workstation offerings that offer some useful features such as ECC & graphics virtualization support but besides that, the GeForce BFGPU and RTX Workstation cards are pretty much the same mostly.
Now NVIDIA based on the leak, we have further specifications of three graphics cards & a prototype that is not known if it's just a test unit or a graphics card that would eventually make its way to the market later. The known variants include the GeForce RTX 4090, GeForce RTX 4080 & GeForce RTX 4070.
RTX 4080 will use AD103 chips, build with 16G GDDR6X, have a similar TGP to GA102.
RTX 4070 will use AD104 chips, build with 12G GDDR6, 300W.
Both of them haven't started testing yet, but soon do them.
— kopite7kimi (@kopite7kimi) April 27, 2022
In fact, there is another full-fat AD102 SKU with 900W TGP, 48G 24Gbps GDDR6X, 2*16pin and higher frequency.
But no one knows whether it will become an actual product. Because the test board of AD102 has more than two 16pin connectors, so everything is possible.
— kopite7kimi (@kopite7kimi) April 27, 2022
NVIDIA Ada Lovelace 'GeForce RTX 40' GPU Configurations
|GPU Name||GPCs / TPCs||SMs Per TPC / Total||CUDA Cores||L2 Cache||Memory Bus||Power Limit Desktop (Expected)||Power Limit Mobile (Expected)|
|AD102||12 / 6||2 / 144||18432||96 MB||384-bit||800W||N/A|
|AD103||7 / 6||2 / 84||10752||64 MB||256-bit||450W||175W|
|AD104||5 / 6||2 / 60||7680||48 MB||192-bit||400W||175W|
|AD106||3 / 6||2 / 36||4608||32 MB||128-bit||260W||140W|
|AD107||3 / 4||2 / 24||3072||32 MB||128-bit||TBD||TBD|
NVIDIA GeForce RTX 4090 Ti / Titan-Class BFGPU:
You may remember a while ago when we reported that NVIDIA's Ada Lovelace GPUs will feature power consumption close to the 1KW range. That was much higher than the now-known 600W TGP that the GeForce RTX 4090 graphics card is expected to rock. It looks like this specific SKU is being tested by NVIDIA and also features the full-fat AD102 GPU but with a 900W TGP, that's a 300W TGP increase over the RTX 4090. The graphics card is said to be powered by dual 16-pin connectors and that would require a similar power setup as the GeForce RTX 3090 Ti KINGPIN which we saw a few days back.
Now one reason why there's such a huge increase in TGP is that the card is expected to rock 48 GB of GDDR6X memory, that's twice the memory capacity of the existing RTX 3090 & also the RTX 4090 which launches later this year. NVIDIA already uses 2 GB memory dies on one side of the RTX 3090 Ti to reach the 24 GB capacity so a card of this scale would have a total of 12 2GB memory chips on both sides of the PCB. This means that we might once again end up with really high GDDR6X temps on the graphics card which is most likely the case given the frequencies these chips will operate at.
It is further pointed out that the GeForce RTX 40 flagship graphics card will have the memory clock at 24 Gbps versus the 21 Gbps of the RTX 4090. Such a memory clock across a 384-bit bus will be providing up to 1.15 TB/s of bandwidth. And this excess clock for the memory and GPU plus twice the memory dies are going to consume a lot of power so 900W while sounding absurd, does look very possible for such a monstrous power-chugging GPU configuration.
NVIDIA GeForce RTX 4090 / AD102 GPU But Optimized
The NVIDIA GeForce RTX 4090 will also rock the same AD102 GPU but in a cut-down configuration based on the rumor. The AD102 GPU will make use of 144 SMs, a 71% increase over the existing GA102 GPU and house a massive 18,432 CUDA core count. In addition to the SM counts, the Ada Lovelace GPUs will also feature increased L2 cache sizes. Starting with the AD102 GPU, the flagship would be outfitted with up to 96 MB of L2 cache, an insane 16x increase over the 6 MB L2 cache featured on GA102.
The graphics card will additionally rock 24 GB of GDDR6X memory clocking in at 21 Gbps across a 384-bit bus interface and delivering over 1.0 TB/s bandwidth. From the TGP figures we have seen, the card will have a TBP of 600W at reference and that's going to go up with the non-reference designs.
As we saw with the GeForce RTX 3090 Ti, there are certain models that go beyond the 450W TBP (reference) and up to 516W (custom). The same is going to be the case with Ada Lovelace-based cards. And while this will require lots of cooling, it looks like AIBs are well and ready. We are likely going to see the use of more hybrid-styled GPU coolers rather than traditional air-coolers once we go above the 600W TBP limit. This will mark a 33% increase in TBP over the RTX 3090 Ti.
NVIDIA GeForce RTX 4080 / Top AD103 GPU For Gamers
The other two configurations that have been rumored include the GeForce RTX 4080 and GeForce RTX 4070. For the RTX 4080, NVIDIA is expected to utilize its AD103 GPU. The AD103 GPU will replace the GA103 GPU which was recently introduced on mobile and feature the same SM count as the GA102 GPU at 84. The AD103 GPU will feature 64 MB of L2 cache. The AD103 GPU is said to feature a TGP similar to the GA102 GPU so around 350-450W and features 16 GB GDDR6X memory across a 256-bit bus interface. The memory clocks have not been stated but they are likely going to be around 20 Gbps.
NVIDIA GeForce RTX 4070 / AD104 at 300W
Finally, there are the NVIDIA GeForce RTX 4070 class GPUs which will utilize the AD104 GPUs. The AD104 GPUs will feature 60 SM units for a total of 7680 CUDA cores & a total of 48 MB of L2 cache. The RTX 4070 series in specific will feature 12 GB of GDDR6 memory running across a 192-bit bus interface and feature a TGP of 300W. Considering that these are GDDR6 chips, we are likely going to get 16-18 Gbps speeds which should provide a bandwidth of around 400 GB/s.
The NVIDIA Ada Lovelace GPU family is expected to bring a generational jump similar to Maxwell to Pascal. It is expected to launch in the second half of 2022 but expects supply and pricing to be similar to current cards despite NVIDIA spending billions of dollars to acquire those good good TSMC 5nm (4N) wafers.
NVIDIA GeForce RTX 40 Series Graphics Card Lineup (Rumored):
|Graphics Card||GPU||PCB Variant||SM Units / Cores||Memory / Bus||Memory Clock / Bandwidth||TBP||Power Connectors||Launch|
|NVIDIA Titan A / GeForce RTX 40?||AD102-450?||PG137-SKU0||142/ 18176?||48 GB / 384-bit||24 Gbps / 1.15 TB/s||~800W||2x 16-pin||TBD|
|NVIDIA GeForce RTX 4090 Ti||AD102-350?||TBD||144 / 18432?||24 GB / 384-bit||24 Gbps / 1.15 TB/s||~600W||1x 16-pin||TBD|
|NVIDIA GeForce RTX 4090||AD102-300?||PG137/139 SKU330||128 / 16384?||24 GB / 384-bit||21 Gbps / 1.00 TB/s||~450W||1x 16-pin||Q4 2022|
|NVIDIA GeForce RTX 4080||AD103-300?||PG136/139-SKU360||76 / 9728?||16 GB / 256-bit||21 Gbps / 672 GB/s||~420W||1x 16-pin||Q4 2022|
|NVIDIA GeForce RTX 4070||AD104-400?||PG141-SKU331||60 / 7680||12 GB / 192-bit||21 Gbps / 504 GB/s||~300W||1x 16-pin||Q4 2022|
|NVIDIA GeForce RTX 4060||AD106-***?||TBD||>36 / 4608?||8 GB / 128-bit||TBD||~200W||1 x 16-pin||Q1 2023|
News Source: Videocardz