New Nvidia RTX 4070 variant based on larger GPU rumoured
Hot on the heels of reports that Nvidia could be working on a 16GB variant of the imminent RTX 4060 Ti (opens in new tab) comes news that the RTX 4070 could be getting a reworking, too.
This time, however, it’s not the VRAM allocation but the RTX 4070’s GPU die itself that’s rumoured to be the target of the changes. Long-time GPU soothsayer kopite7kimi (opens in new tab) (via Videocardz (opens in new tab)) says that Nvidia could release a new RTX 4070 (opens in new tab) variant based on the AD103 die.
That GPU, of course, is the chip you’ll find in the much more powerful RTX 4080 board. The RTX 4070 and RTX 4070 Ti (opens in new tab) currently use the smaller, cheaper AD104 GPU.
Currently, the RTX 4070 runs 5,888 CUDA cores. However, the AD103 chip has fully 10,240 cores. That means Nvidia could disable nearly half the cores, allowing for chips with multiple manufacturing defects to be salvaged.
Nvidia could also do a variant of the RTX 4070 based on AD103 with more cores than the current AD104 version. There is precedent for that kind of thing.
However, it wouldn’t make sense to increase the CUDA core count or performance too much. After all, Nvidia also sells the RTX 4070 Ti, another AD104 variant, this time with 7,680 CUDA cores.
Still, there was arguably a bigger gap between the RTX 4070 and 4070 Ti models than you normally see between a standard and Ti version of a given GPU tier. So, a slightly faster 4070 could make sense.
It would also make sense to go on and create RTX 4070 Ti’s out of broken AD103 chips, too, and arguably more so given the existing RTX 4070 already uses a heavily cutdown version of the AD104 die.
Whatever, it all adds to the sense that Nvidia is willing to adjust its maturing Ada Lovelace generation of GPUs. First, it cancelled the RTX 4080 12GB (opens in new tab), rebranding it into the RTX 4070 Ti.
Then we got the news that Nvidia might do a 16GB version of the RTX 4060 Ti to head off concerns that the standard card’s 8GB isn’t good enough for the latest games. Now this mooted AD103-based RTX 4070 variant.
Of course, using a larger GPU order to salvage partially broken chips is nothing new. Both AMD and Nvidia do this in order to effectively boost yields. But it will be interesting to see if Nvidia uses what is usually a commercial measure intended to be largely invisible to GPU buyers to slightly reposition its RTX 40-series GPUs.
It’s probably fair to say they’ve had something of a mixed reception so far. But pretty much all of that has been down to how Nvidia has positioned and price the various models so far. Nobody doubts that underlying technology and architecture of Ada Lovelace. That’s great.
But in really simple terms, it would be nice have better specified variants for a bit less cash. Watch this space.
