A user on Hardwareluxx’s Forums has confirmed the existence of NVIDIA’s Ampere GA102-250 GPU SKU which was previously expected to be featured on the GeForce RTX 3080 Ti graphics card. It was spotted when the forum member took apart his GeForce RTX 3090, only to reveal that the GPU had initially been labeled as the GA102-250 SKU instead of the GA102-300 which is the official SKU used on the RTX 3090.
NVIDIA’s Rumored GeForce RTX 3080 Ti GPU, The Ampere GA102-250, Spotted on A GeForce RTX 3090 Graphics Card
The NVIDIA GA102-250-KD-A1 GPU only existed in the rumor mill but now we can confirm that this is indeed a real chip. Initially planned for the GeForce RTX 3080 Ti graphics card, it looks like NVIDIA decided that they could use a further cut-down SKU to power the upcoming flagship while this SKU can be utilized for the GeForce RTX 3090 graphics card. You can clearly see that the GA102-250-KD-A1 GPU label has been cut-off and instead, it’s been repurposed as the GA102-300-A1 SKU. What’s interesting is that the GA102-250-KD-A1 was expected to feature 78 SM units enabled for a total of 9984 CUDA cores whereas the RTX 3090 with its GA102-300-A1 GPU packs 82 SMs or 10496 cores.
It could be possible that NVIDIA decided to repurpose the GA102-250 SKU into GA102-300 given the shortages of its high-end graphics cards. We cannot say for sure if the 250 variant offers the same performance/power limits as the 300 variant but given that it has a lower naming scheme, one could expect it to be a lower bin than the standard GA102-300 SKU. The GPU was also made during the 49th week of 2020 which is between the end of November and early December. Again, we do not know the extent to which NVIDIA is repurposing its GA102-250 chips to GA102-300 but since they exist and this is out now, we can expect more users to find this out by themselves.
The NVIDIA GeForce RTX 3080 Ti on the other hand has seen a spec-change and will now be using the Ampere GA102-225-A1 GPU SKU. Following are the expected specifications of the upcoming RTX 30 series flagship.
NVIDIA GeForce RTX 3080 Ti 12 GB ‘Rumored’ Graphics Card Specifications
NVIDIA’s GeForce RTX 3080 Ti 12 GB FE (Founders Edition) graphics card is expected to feature the PG132-SKU18 PCB design and the GA102-225-KD-A1 graphics core. The GA102-225 GPU has also changed since the last time we saw them and will now feature 10240 CUDA cores within a total of 80 SM units. Since the main core configuration is entirely brand new, the previously leaked clocks & TMU/ROP counts should now be dismissed.
As for memory, the card will feature 12 GB of GDDR6X memory. Unlike the 19.5 Gbps speeds of the RTX 3090, the GeForce RTX 3080 Ti is said to retain the same memory speeds as the RTX 3080 at 19 Gbps. Since we are getting 12 GB memory, the NVIDIA GeForce RTX 3080 Ti will be using a 384-bit bus interface which equals a total bandwidth of 912 GB/s. So while we see an 8 GB VRAM reduction over the previous spec, the larger bus interface should drive higher memory bandwidth.
The TGP for the card is set to be the same as the RTX 3080 at 320 Watts. That’s definitely needed to feed the extra cores so NVIDIA might have to optimize the clocks a bit here. The NVIDIA GeForce RTX 3080 Ti is expected to launch in mid-April for an MSRP of around $999 US. The card is rumored for launch in May.
NVIDIA GeForce RTX 30 Series ‘Ampere’ Graphics Card Specifications:
Graphics Card Name | NVIDIA GeForce RTX 3050 | NVIDIA GeForce RTX 3050 Ti | NVIDIA GeForce RTX 3060 | NVIDIA GeForce RTX 3060 Ti | NVIDIA GeForce RTX 3070 | NVIDIA GeForce RTX 3070 Ti? | NVIDIA GeForce RTX 3080 | NVIDIA GeForce RTX 3080 Ti? | NVIDIA GeForce RTX 3090 |
---|---|---|---|---|---|---|---|---|---|
GPU Name | Ampere GA107 | Ampere GA106? | Ampere GA106-300 | Ampere GA104-200 | Ampere GA104-300 | Ampere GA104-400 | Ampere GA102-200 | Ampere GA102-225? | Ampere GA102-300 |
Process Node | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm |
Die Size | TBA | TBA | TBA | 395.2mm2 | 395.2mm2 | 395.2mm2 | 628.4mm2 | 628.4mm2 | 628.4mm2 |
Transistors | TBA | TBA | TBA | 17.4 Billion | 17.4 Billion | 17.4 Billion | 28 Billion | 28 Billion | 28 Billion |
CUDA Cores | 2304? | 3584? | 3584 | 4864 | 5888 | 6144? | 8704 | 10240? | 10496 |
TMUs / ROPs | TBA | TBA | 112 / 64 | 152 / 80 | 184 / 96 | 192/ 104? | 272 / 96 | 320 / 112? | 328 / 112 |
Tensor / RT Cores | TBA | TBA | 112 / 28 | 152 / 38 | 184 / 46 | 192/ 48? | 272 / 68 | 320 / 76? | 328 / 82 |
Base Clock | TBA | TBA | 1320 MHz | 1410 MHz | 1500 MHz | TBA | 1440 MHz | TBA | 1400 MHz |
Boost Clock | TBA | TBA | 1780 MHz | 1665 MHz | 1730 MHz | TBA | 1710 MHz | TBA | 1700 MHz |
FP32 Compute | TBA | TBA | 12.7 TFLOPs | 16.2 TFLOPs | 20 TFLOPs | TBA | 30 TFLOPs | TBA | 36 TFLOPs |
RT TFLOPs | TBA | TBA | 25.4 TFLOPs | 32.4 TFLOPs | 40 TFLOPs | TBA | 58 TFLOPs | TBA | 69 TFLOPs |
Tensor-TOPs | TBA | TBA | 101 TOPs | 129.6 TOPs | 163 TOPs | TBA | 238 TOPs | TBA | 285 TOPs |
Memory Capacity | 4 GB GDDR6? | 6 GB GDDR6? | 12 GB GDDR6 | 8 GB GDDR6 | 8 GB GDDR6 | 8 GB GDDR6X | 10 GB GDDR6X | 12 GB GDDR6X | 24 GB GDDR6X |
Memory Bus | 128-bit | 192-bit? | 192-bit | 256-bit | 256-bit | 256-bit | 320-bit | 384-bit | 384-bit |
Memory Speed | TBA | TBA | 15 Gbps | 14 Gbps | 14 Gbps | TBA | 19 Gbps | 19 Gbps | 19.5 Gbps |
Bandwidth | TBA | TBA | 360 Gbps | 448 Gbps | 448 Gbps | TBA | 760 Gbps | 912 Gbps | 936 Gbps |
TGP | 90W? | TBA | 170W | 200W | 220W | 250W? | 320W | 320W | 350W |
Price (MSRP / FE) | $149? | $199? | $329 | $399 US | $499 US | $599 US? | $699 US | $999 US? | $1499 US |
Launch (Availability) | 2021? | 2021? | February 2021 | December 2020 | 29th October | May 2021? | 17th September | April 2021? | 24th September |