L1 and L2 cache specs for Nvidia’s new RTX 5090 and RTX 5080 have cropped up. HardwareLuxx reports that the RTX 5090 and 5080 have the identical L1 cache capability per SM because the 4090 and 4080, and the 5090 has 36% extra L2 cache in comparison with its predecessor.
L1 cache capability per SM reportedly stays the identical on GB202 as is on AD102, that includes 128 kB of capability per SM. In consequence, the RTX 5090 options 21.7 MB of L1 cache capability in whole, giving the Blackwell GPU 5.4MB extra L1 cache over the RTX 4090, due to its improved SM rely of 170 in comparison with 128 on the RTX 4090 (21,760 CUDA cores vs 16,384).
| GPU: | L1 Cache: | L2 Cache: |
| RTX 5090 | 21.7MB | 98.3MB |
| RTX 4090 | 16.3MB | 72MB |
| RTX 5080 | 10.7MB | 65MB |
| RTX 4080/Tremendous | 9.7MB | 64MB |
The identical can be the case on GB203, which is the die that powers the RTX 5080. Nevertheless, the SM rely variance between the RTX 5080 and 4080 is considerably smaller, giving each GPUs virtually an identical L1 cache capability in whole. The RTX 5080 comes with 10.7 MB of L1 cache, and the RTX 4080 comes with 9.7MB of L1 cache, a mere 1MB distinction.
We are able to anticipate the remaining Blackwell dies to share the identical pattern. Actually, Blackwell’s 128 KB L1 cache measurement per SM not solely compares precisely with Ada Lovelace, however Ampere as effectively. Ampere represents the final time Nvidia upgraded L1 cache capability per SM, doubling it in comparison with Turing.
The RTX 5090 receives a 36% enchancment in L2 cache capability over the RTX 4090, that includes virtually 100MB of the stuff. The RTX 5080 will get nearly no improve, that includes simply 1MB extra L2 cache than the RTX 4080 and 4080 Super.
Blackwell’s minor cache measurement enhancements starkly distinction Ada Lovelace’s large cache capability improve over Ampere, particularly the L2 cache. The RTX 4090, for instance, featured a whopping 12x extra cache in comparison with the RTX 3090 collection (72MB vs 6MB).
To compensate for the minor cache enhancements on Blackwell, the RTX 50 collection as a complete will get an improve to speedier GDDR7 reminiscence modules, working at 28Gbps, apart from the 5080, which will get particular therapy that includes even sooner 32Gbps modules.
A few of the 50 collection fashions additionally include bus-width upgrades on high of the GDDR7 improve to spice up reminiscence efficiency additional. The RTX 5090 will get a 512-bit reminiscence bus, and the RTX 5070 Ti will get a 256-bit reminiscence bus. Each are upgrades in comparison with their RTX 4090 and 4070 Ti/Super predecessors. The RTX 5080 and RTX 5070 keep on with the identical bus widths as their predecessors.
