The second major rumor dropped by veteran-leaker, Kopite7kimi, today is that NVIDIA's next-gen GeForce RTX flagship based on the "Ada-Next" GPU architecture will be utilizing a massive 512-bit bus interface.
NVIDIA GeForce RTX 4090's Gaming GPU Successor "Ada-Next" To Feature A Massive 512-Bit Bus
While NVIDIA is rumored to be axing its flagship GeForce RTX 4090 Ti GPU, it looks like the green team may already be working on a next-gen successor with the information highlighting a huge upgrade to the memory configuration. According to Kopite7kimi, the Ada-Next flagship gaming GPU, which is expected to arrive sometime in 2025, will be featuring a 512-bit bus interface & I'll explain why that's such a big deal.
Combined with multiple sources, I confirm the gaming flagship of Ada-next will have a 512-bit memory interface.
— kopite7kimi (@kopite7kimi) July 27, 2023
NVIDIA's GeForce gaming graphics card lineup including the brand new ones hasn't featured a memory configuration above 384-bit since the GeForce GTX 285 which was released all the way back in 2009. This is significant because having a 512-bit memory configuration essentially lets NVIDIA experiment with some massive memory capacities. Currently, the 384-bit bus graphics cards such as the RTX 4090 and RTX 6000 Ada feature up to 24 GB and 48 GB capacities. With a 512-bit bus interface, you can get 32 GB capacities and expand up to 64 GB capacities on a single card. That's a 33% increase in memory capacities.
Also, that's not even considering the newer 24 Gb GDDR7 modules which are being designed by DRAM makers including Micron and Samsung. If NVIDIA was to use these modules, you can even get up to 48/96 GB VRAM capacities which would be insane amounts and a big boost for content creators and AI workloads but I think gaming won't essentially take much advantage of such high capacities so let's stick to 32 GB for now. If the stock 32 Gbps speeds of GDDR7 memory are used, then you get up to 2 TB/s of bandwidth and 2.3 TB/s with the boosted 36 Gbps dies. Even with the 24 Gbps GDDR6X dies, a 512-bit memory configuration will lead to 1.5 TB/s of bandwidth or a 50% boost over the RTX 4090.
NVIDIA has made some massive improvements to its memory and cache design over recent years along with optimizations to its memory compression algorithms which allow a reduction in reliance on the available bandwidth of the GPU. The biggest cache upgrade was seen in Ada cards which led to NVIDIA significantly reducing the memory bus interfaces on its mid-range lineup such as the RTX 4060 series which sits at 128-bit and RTX 4070 which peaks out at 192-bit. This helps reduce cost & improves power efficiency but the reduced memory bandwidth can lead to some bottlenecks in a variety of workloads such as games that utilize high-quality textures NVIDIA also has some AI techniques coming up that will help resolve texture caching and loading through the utilization of its tensor core technology.
It isn't confirmed if NVIDIA's next-gen GPUs will utilize a GDDR7 memory standard or stick with GDDR6X for another generation but it would be unwise not to go with the newer technology since it hits mass production in late 2024. Once again, the next-gen cards aren't due until 2025 as per NVIDIA's own roadmap.
NVIDIA's Flagship Gaming GPU "Memory" Specs Comparison:
| Graphics Card | Memory Type | VRAM | Memory Bus | Memory Bandwidth |
|---|---|---|---|---|
| RTX 5090? | TBD | TBD | 512-bit? | TBD |
| RTX 4090 | GDDR6X | 24 GB | 384-bit | 1008 GB/s |
| RTX 3090 Ti | GDDR6X | 24 GB | 384-bit | 1008 GB/s |
| RTX 2080 Ti | GDDR6 | 11 GB | 352-bit | 616 GB/s |
| GTX 1080 Ti | GDDR5X | 11 GB | 352-bit | 484 GB/s |
| GTX 980 Ti | GDDR5 | 6 GB | 384-bit | 336 GB/s |
| GTX 780 Ti | GDDR5 | 3 GB | 384-bit | 336 GB/s |
| GTX 680 | GDDR5 | 2 GB | 256-bit | 192 GB/s |
| GTX 580 | GDDR5 | 1.5 GB | 384-bit | 192 GB/s |
| GTX 480 | GDDR5 | 1.5 GB | 384-bit | 177 GB/s |
| GTX 285 | GDDR3 | 1.0 GB | 512-bit | 159 GB/s |
from Wccftech https://ift.tt/TW9ylQt
0 Comments