Nvidia’s RTX 4070 GPU might not be as powerful as we’ve recently heard in the rumor mill, with Team Green apparently still not finalizing some key specs – and we’re talking about the CUDA Core count here, as well as the amount of video memory.
The latest update from the prolific Twitter-based leaker Kopite7kimi gives us two different possible specs for the RTX 4070 graphics card (apply your own spice and don’t be stingy either).
Review the two possible RTX 4070 specs: PG141-SKU340/341, 7680FP32, 12G 21Gbps GDDR6X, total card power 285W, TSE <11000. PG141-SKU336/337, 7168FP32, 10G 21Gbps GDDR6X, total card power 250W, TSE <10000. I found out that I had already made a typo before, wrongly I wrote 341 as 331. 😅August 29, 2022
As you can see in the tweet, one of them is a more powerful configuration of the RTX 4070 that offers 7,680 CUDA Cores and 12GB of GDDR6X VRAM (at 21Gbps), with a power consumption of 285W.
The other option that Nvidia can run with reduces the number of CUDA cores to 7,168 and VRAM to 10GB, with power consumption set to a less demanding 250W.
The difference between these two possible RTX 4070 GPUs could be that the beefier old card would be around 10% faster, at least following Kopite7kimi’s predictions for a 3DMark (Time Spy Extreme benchmark) result.
Analysis: Is Nvidia involved in more than one careful balancing act?
Rumor watchers will no doubt remember that the smallest of the aforementioned specs is an older setup that went live some time ago. Kopite7kimi has since changed his tune on the RTX 4070 and stated that it would be like the more powerful configuration mentioned above, with 12GB of VRAM and 7,680 cores; the only difference between the previous leak and this new one is that the power usage has apparently been adjusted to 285W (instead of 300W).
So this is good news if Nvidia does indeed run with the more robust possibility in the tweet above, but a disappointing step up from previous speculation if the lower spec is the chosen route. There have long been concerns about the RTX 4070 only having 10GB of VRAM, with many feeling this is too short on the card – so there was a lot of relief when rumor talk turned to a 12GB loadout. But now we have a cloud of doubt about it.
Is Nvidia still wondering if 12GB should be saved for the RTX 4070 Ti? And indeed, looking at the lower spec option that the Kopite7kimi provides is considering the overall alignment of the Lovelace GPU range and making sure the RTX 4070 fits and makes sense compared to the RTX 4080. RTX 4070’s most powerful rumor will take away the 4080’s sales (and its higher profit margin), and presumably Nvidia will want to tread carefully there.
All of this assumes that Kopite7kimi is correct, of course, and perhaps the leaker could be wrong. But assuming not, and that Nvidia is still effectively putting the pieces of the Lovelace puzzle in place as to where the various RTX 4000 models stand in relation to each other in terms of specs, that seems to suggest that the RTX release 4070 is not close. .
Another possibility that has come out of the vine is that Nvidia may release the RTX 4090 in October, and instead of the RTX 4080 and 4070 in a month (or two), they could be delayed until 2023. (To make it easier to shift GPU stock RTX 3000 in excess, which would otherwise not be sold if high-end cards are available or imminent).
If Team Green hasn’t yet decided the exact specs of the RTX 4070 as suggested, that seems to point to the perhaps more distant release date. Of course, card partners will need a lot of time and notice about the official configuration of these products to start designing their 3rd party graphics cards and consider cooling solutions and so on.
There are two ways, then, that this latest rumor is a little disappointing – it suggests that the less powerful theorized spec could still be used for the RTX 4070, and that we might not see this Lovelace graphics card for some time yet.