L1 and L2 cache specifications have been released for Nvidia’s up-to-date RTX 5090 and RTX 5080 cards. HardwareLuxx Reports that the RTX 5090 and 5080 have the same L1 cache capacity on SM as the 4090 and 4080, and the 5090 has 36% more L2 cache compared to its predecessor.
L1 cache capacity on MS it reportedly remains the same on the GB202 as it did on the AD102, offering 128kB of capacity per SM. As a result, the RTX 5090 offers a total of 21.7 MB of L1 cache, giving the Blackwell GPU 5.4 MB more L1 cache compared to the RTX 4090, thanks to an improved SM number of 170 compared to 128 in the RTX 4090 (21 760 CUDA cores compared to 16,384).
| Graphics Processor: | L1 cache: | L2 cache: |
| RTX5090 | 21.7MB | 98.3MB |
| RTX4090 | 16.3 MB | 72MB |
| RTX5080 | 10.7MB | 65 MB |
| RTX 4080/Super | 9.7MB | 64 MB |
The same applies to GB203, the die powering the RTX 5080. However, the difference in the SM count between the RTX 5080 and 4080 is much smaller, which gives a total of almost identical L1 cache capacity for both GPUs. The RTX 5080 comes with 10.7 MB of L1 cache and the RTX 4080 has 9.7 MB of L1 cache, a difference of just 1 MB.
We can expect the remaining Blackwell dies to follow the same trend. In fact, Blackwell’s 128KB L1 cache size on SM is comparable not only to Ada Lovelace, but also to Ampere. Ampere represents the last time Nvidia increased the L1 cache capacity on SM, doubling it compared to Turing.
The RTX 5090 gets 36% more L2 cache capacity compared to the RTX 4090, equipped with almost 100 MB. The RTX 5080 receives virtually no updates and only offers 1MB more L2 cache than the RTX 4080 and 4080 Super.
Blackwell’s petite cache size improvements stand in stark contrast to Ada Lovelace’s massive cache size enhance over Ampere, particularly the L2 cache. For example, the RTX 4090 had as much as 12 times the cache compared to the RTX 3090 series (72 MB vs. 6 MB).
To compensate for Blackwell’s minor cache improvements, the entire RTX 50 series has been upgraded to faster GDDR7 memory modules, running at 28Gbps, except for the 5080, which has received special treatment with even faster 32Gbps modules.
Some 50 Series models, in addition to GDDR7 memory expansion, also offer the option to enhance the bus width to further enhance memory performance. The RTX 5090 gets a 512-bit memory bus and the RTX 5070 Ti gets a 256-bit memory bus. Both are improvements over their RTX 4090 and 4070 Ti/Super predecessors. The RTX 5080 and RTX 5070 maintain the same bus width as their predecessors.
