NVIDIA GeForce RTX 4000 GPUs receive revised specs in new leak: more cores, less video memory, 160-bit memory bus and more

As we get closer to the launch of the RTX 40 series, every day there is a new leak that contradicts the previous one. Just a couple of days ago, I broke the news about a leak about the power limitations of the upcoming Ada Lovelace GPUs. Today we have a kind of continuation.

The rumor comes from the same leaker, @koptite7kimi, who took to Twitter again to share updated specs for the RTX 40 series graphics cards. Since day one, the exact specs for these GPUs have been constantly changing. Let’s see what Kopite has to say this time. 

In his tweet pinned above, you can see him going over the specs of the initial line of RTX 40-series GPUs. This includes the RTX 4090, RTX 4080, and RTX 4070. Starting with the top SKU, we only see one update to its specs.

RTX 4090 updated look

The RTX 4090 is rumored to have 16128 CUDA cores. The card will use the flagship AD102 GPU, but not the full die that is reserved for the RTX 4090 Ti (and possibly the new TITAN). The new, updated specs suggest the RTX 4090 will run 16384 CUDA cores instead.

This increases the number of stream processors from 126 to 128. As already mentioned, the full AD102 matrix is ​​much more productive – 144 SM. Regardless, the rest of the RTX 4090 remains unchanged. It still reportedly ships with 24GB of 21Gb/s GDDR6X memory running on a 384-bit bus. TDP is also the same – 450 watts. 

NVIDIA GeForce RTX 4090 Ti Breadboard | Notebook check

RTX 4080 updated look

Moving on to the RTX 4080, this is the next flagship in line for Ada Lovelace. The updated Kopite specs actually give us a first look at the internal configurations of the RTX 4080 and RTX 4070. Previously, we only really knew which GPU they would use.

So the RTX 4080 is based on the NVIDIA AD103 GPU and the card is likely not utilizing the entire die due to the prospect of unlocking that potential in a future RTX 3080 Ti. The RTX 4080 will reportedly have 10240 CUDA cores. 

In terms of memory, we’re looking at 16GB of GDDR6 spread across a 256-bit memory bus. The memory runs at 18Gbps. This would make it slower than the RTX 3080 which has 19Gbps ​​memory due to the use of faster modules.

Unlike the RTX 4090 (and most of the current 30-series), the RTX 4080 does not appear to be using GDDR6X memory. It will most likely not be left out as more leaks emerge in the coming weeks. 

For those who don’t know, GDDR6X is NVIDIA’s special variant of the GDDR6 standard, co-developed with Micron and exclusively for its GPUs. It runs hotter, but presumably faster, and in NVIDIA’s eyes, the tradeoff is worth it.

Micron GDDR6X compared to previous generations of video memory | Micron via OC3D

In any case, even if the RTX 4080 comes with slower memory than its predecessor, it will only be so on paper. Because Ada Lovelace GPUs have a larger L2 cache , which makes them much more efficient, which in turn can make the GPU as fast or even faster than the GDDR6X variant in terms of memory. 

RTX 4070 updated look

Finally, we have the RTX 4070, the card that will arguably sell best at launch of all the 40-series offerings. Here, the only spec update concerns the amount of memory. The card is now expected to feature 10GB of GDDR6 memory instead of the previously announced 12GB. 

This 2 GB reduction will narrow the memory bus from 192-bit initially to 160-bit. In addition, the number of CUDA cores is reported to be 7168, which is 512 fewer than the full-fledged AD104 GPU. Again, the RTX 4070 Ti is where we’re likely to see a fully unlocked AD104.

New generation

That concludes our updated look at the next generation of NVIDIA GPUs. At the end of Save’s tweet, he mentions (or rather warns) not to expect a drop in the suggested retail price. We are already paying more than suggested retail prices (with the exception of the recent cryptocurrency crash), so further price increases could be a real disappointment.

According to DigiTimes , TSMC has increased the prices of its plates by about 10-20%. This, coupled with the ongoing hyperinflation seen around the world, will inevitably push costs up and customers are expected to swallow the consequences. 

Overall, the RTX 40 series looks promising. With higher power requirements and more boastful specs than ever, this will be a pretty drastic upgrade. AMD is also poised to introduce its RDNA 3 GPUs to the world very soon, a lineup that is arguably even more exciting than NVIDIA given the hill AMD currently stands on.

RDNA 2 came so close to toppling the Green Team because of its superior graphics card, but RDNA 3 is expected to be the nail in the head. Not only that, Intel is also refining its Arc A-Series GPUs, which could pose another unanticipated threat to NVIDIA’s dominance. 

Intel Arc Rendering for Mobile and Desktop | Intel

The battle for consumer pockets between AMD, Intel and NVIDIA promises to be an interesting one, and there may not be a clear winner. Whether all of these companies can triumph and find their footing in the rapidly changing GPU market with next-generation products remains to be seen.

We don’t have to wait long to see it all as RDNA 3 GPUs are expected to launch sometime in November and Intel Arc GPUs are already starting to hit the shelves. The NVIDIA RTX 40 series is rumored to launch in Q2/Q4 2022, with each of the above three GPUs launching one month apart. 

Leave a Reply

Your email address will not be published.