Although there is nothing logically official from NVIDIA, what is leaked on Geekbench is as curious as it is hopeful for the market. And not for the simple fact that they are new models that improve the present, but because NVIDIA opens a door to end the problem of stock and the overpricing of GPUs.
GA102-250 chip shown, a range GPU above the RTX 3080 Ti?
A wild GA102 config appears ?
> 1.71GHz core clock
> 20GB VRAM GDDR6 19Gbpshttps: //t.co/PBXdOz2ACa
I think this is GA102-250 that was supposed to be in the RTX 3080 Ti pic.twitter.com/NkdDSg24Pu
– _rogame (@_rogame) June 29, 2021
It was released earlier this month and its hours may be numbered, surreal some might think, and they wouldn’t go wrong. But the reality is that the chip that was originally going to arrive for the RTX 3080 Ti was not such, but a version with a different nomenclature called GA102-225-A1.
The expected chip was therefore discarded for some reason and NVIDIA provided the one named above, but what is our surprise that yesterday the older of the two reappeared on Geekbench, where really interesting data was shown. This chip is the GA102-250 that we saw months ago and as such it is not new, but it has not been discarded according to what we are going to see next. The most obvious question is why didn’t NVIDIA include it in its latest high-end GPU? And of course, what do you intend with this new chip?
New range of GPUs RTX 30 SUPER?
The surprising thing about the leaked is that it is specified that said NVIDIA graphics card contains 82 SM at a frequency of 1.71 GHz and nothing less than 20 GB of VRAM. To put us in context, the RTX 3090 gets those same 82 SM at 1695 MHz with 24GB of VRAM, while the RTX 3080 Ti stays at 80 SM at 1.66 GHz and 12GB of GDDR6X.
Therefore, and looking at the GB of VRAM memory, we would be talking about a chip with a 320-bit bus in memory that if the 19.5 Gbps of its older sister are not modified would give a bandwidth of 780 GB / s.
That is, it would have more muscle than the RTX 3080 Ti, but less bandwidth in memory. This move is curious, since NVIDIA has always commented that the current bottleneck in GPUs was precisely the bandwidth with VRAM. Or are you planning to release this chip with 384 bits even though it has disabled 2 IMC controllers?
Rumors are on the table, even claiming that NVIDIA would be working on a new series of GPUs as happened with the RTX 20, that is, a partial or complete renewal of ranges to launch RTX 30 graphics cards. SUPER. In this way, it would leave out of stock (there is not equally) of the current models except the last three presented and would launch new GPUs with blocking for mining.
We will have to wait to find out more about this, the first tiles have already been placed on the board.