The latest Ampere leak from VideoCardz might have just ruined Nvidia's announcement. The publication has obtained new information that allegedly lends credence to the rumored specifications for the GeForce RTX 3090 and GeForce RTX 3080 that will soon vie for a spot on our list of Best Graphics cards.
Starting from the top, Ampere silicon will seemingly be manufactured on the 7nm process node. The leaked specification sheets don't specify the foundry, but TSMC's 7nm FinFET manufacturing process is the first thing to come to mind. Ampere-based graphics, such as the GeForce RTX 3090 and RTX 3080 will get to exploit the PCIe 4.0 interface. Both models are rumored to use the same GA102 die. Nvidia SLI is only supported on the GeForce RTX 3090 through the chipmaker's NVLink interface.
Physically, Gainward's custom models fall right in line with what we've seen from the Founders Edition and the Zotac Gaming GeForce RTX 3090 Trinity HoLo. The GeForce RTX 3090 and GeForce RTX 3080 are going to be gigantic graphics cards. The Gainward models adhere to a 2.7-slot design and relies on a beefy triple-fan cooling solution.
Nvidia GeForce RTX 3090, RTX 3080 Specifications
| Graphics Card | CUDA Cores | Memory Clock | Memory Capacity | Memory Bus | Memory Bandwidth | TGP |
|---|---|---|---|---|---|---|
| GeForce RTX 3090* | 5,248 | 19 Gbps | 24GB GDDR6X | 384-bit | 935.8 GBps | 350W |
| Titan RTX | 4,608 | 14 Gbps | 24GB GDDR6 | 384-bit | 672 GBps | 280W |
| GeForce RTX 3080* | 4,352 | 19 Gbps | 10GB / 20GB GDDR6X | 320-bit | 760 GBps | 320W |
| GeForce RTX 2080 Ti | 4,352 | 14 Gbps | 11GB GDDR6 | 352-bit | 616 GBps | 250W |
*Specifications are unconfirmed.
The GeForce RTX 3090 appears to come equipped with 5,248 CUDA cores and 24GB of GDDR6X memory after all. As for clock speeds, the Gainward GeForce RTX 3090 Phoenix Golden Sample has a 1,725 MHz boost clock.
The GDDR6X memory on the GeForce RTX 3090 runs at 19.5 Gbps across a 384-bit memory interface. If you do the math, the maximum theoretical memory bandwidth adds up to 936 GBps, just a bit shy of 1 TBps. The GeForce RTX 3090's TGP (total graphics power) is rated for 350W. Gainward's model confirms early rumors about the utilization of twin 8-pin PCIe power connectors. In regards to video outputs, the GeForce RTX 3090 Phoenix Golden Sample in particular provides three DisplayPort 1.4a outputs and one HDMI 2.1 port.
As for the GeForce RTX 3080, Gainward's specification table has the graphics card with 4,352 CUDA cores and 10GB of GDDR6X memory. The company's custom model flexes a 1,740 MHz boost clock. The memory checks in at 19 Gbps with a 320-bit memory bus, which amounts to a memory bandwidth of 760 GBps.
The TGP for the GeForce RTX 3080 seems to be 320W, just 30W lower than the GeForce RTX 3090. This should still merit the presence of two 8-pin PCIe power connectors. Video outputs on the GeForce RTX 3080 Phoenix Golden Sample are identical to those of its bigger brother.
The GeForce RTX 3090's 24GB VRAM screams that it's the replacement for the Titan RTX. This is puzzling because it isn't like Nvidia to release a Titan-esque graphics card this early in a product launch. The Titan models typically come a few months after. It's not like Nvidia is shooting itself in the foot, either. The GA102 silicon presumably houses up to 7,552 CUDA cores, and only 5,248 CUDA cores are enabled in the GeForce RTX 3090. If necessary, Nvidia has more than enough headroom to roll an Ampere-powered Titan graphics card later on or even a GeForce RTX 3080 Ti.
The GeForce RTX 3080, on the other hand, is a curious case. The graphics card reportedly has the same number of CUDA cores as the GeForce RTX 2080 Ti but loses 1GB of VRAM. In the Nvidia's defense, the GeForce RTX 3080 did receive a memory upgrade to push the memory bandwidth up by 23.4%. The graphics card is on a new architecture though so it remains to be seen how the situation pans out.
Nvidia's big announcement is this Tuesday. It'll be interesting to see how the chipmaker sells Ampere to the public.
