Although Nvidia hasn’t said much about its upcoming RTX 50-series graphics cards during Computex, they’re coming, and we just got a new rumor about their potential specs. According to a reputable leaker, the RTX 5080 might be less impressive than initially suggested. If this is proven true, is this the direct result of Nvidia not needing to try as hard, given that AMD may be sitting out the race for the best graphics card in this generation?
The intel comes from kopite7kimi — a frequent leaker in the GPU space, who has now said that not only will it take a while for most of the lineup to arrive but also that the GB203 GPU will be significantly cut down compared to previous rumors. Many suspect that the GB203 will be the GPU inside the RTX 5080, so this is a rather big deal.
In fact, this might stretch to more than just the GB203. According to kopite7kimi, Blackwell GPUs will all have the same graphics processing cluster (GPC) count as their Lovelace counterparts. The only exception is said to be the GB207, which is likely to be a budget card along the likes of the RTX 4060.
Get your weekly teardown of the tech behind PC gaming
Due to the order of product development, we will not see finished products other than GB202 and GB203 in the short term, we will not discuss GB205, GB206, and GB207 at this time.
I feel sorry to everyone, GB203 is not 6*8, but 7*6, similar to AD103. 😅— kopite7kimi (@kopite7kimi) June 2, 2024
Let’s circle back to the GB203. Kopite7kimi claims that the GPU configuration, or rather the GPCs, will have changed for the worse, going down from a six-by-eight setup to seven-by-six — which is similar to AD103, the GPU found within the RTX 4080. Although the leaker doesn’t give a clear indication as to how many CUDA cores or streaming multiprocessors (SMs) we can now expect, signs point to just 84 SMs instead of 96 for the GB203. That’s a major decrease, and one that would put the rumored RTX 5080 just four SMs ahead of the RTX 4080 Super. It’s not a big change at all, and a far cry from the RTX 4090 with its 128 SMs.
If the second-best GPU in the Blackwell lineup won’t even be able to outperform the RTX 4090, what is even the point? Well, it’s not as black and white as that. For starters, chances are that the RTX 5080 would be cheaper than the RTX 4090, which still holds steady at around $1,600 and above. But if the difference between the RTX 4080 and the RTX 5080 ends up being too subtle, then it may find itself in no man’s land — not powerful enough and not cheap enough to entice users to upgrade. Of course, all of this is just speculation at this point.
There’s an alternative that kopite7kimi didn’t discuss. If Nvidia wanted to make the RTX 5080 significantly better than what the GB203 chip can offer, it could always build it on a cut-down version of the flagship GB202 GPU. That way, the RTX 5090 would get most of the chip’s power, and the RTX 5080 would still be a significant upgrade over its predecessors.
Yes.
— kopite7kimi (@kopite7kimi) June 2, 2024
Kopite7kimi also said that only the GB202 and the GB203 (or the RTX 5090 and the RTX 5080) will be out anytime soon. The more midrange and entry-level chips are said to still be in development and won’t be showing up anytime soon. This could give us an interesting GPU market by the end of the year.
If Nvidia really settles for less with the RTX 5080, it might be due to a lack of competition from AMD — although nothing’s been confirmed yet on that front. Rumor has it that the top card in AMD’s RDNA 4 lineup may fall just short of what the mysterious GB203 is said to offer. As such, we’d end the year with Nvidia dominating the high-end sector, AMD being the next-best thing, and Intel filling in at the entry level. All of this is speculation, so without putting too much stock in it, it sure looks like we’ve got an interesting second half of the year ahead.
Editors’ Recommendations
