Nvidia's forthcoming GeForce RTX 4070 Super seems to have what it takes to rival the best graphics cards. A pair of RTX 4070 Super benchmark results were leaked onto the Geekbench browser, illuminating the GPU's potential performance. The benchmark results were published by Benchleaks, showing RTX 4070 Super results in Geekbench 6's CUDA and OpenCL benchmarks. The new GPU is swift, featuring benchmark scores close to Nvidia's outgoing RTX 4070 Ti. But being Geekbench 6, take the results with a pinch of salt and wait for the full review.
The RTX 4070 Super scored 195,384 points in Geekbench 6's OpenCL test and 219,237 in Geekbench 6's CUDA benchmark. Compared to its predecessor, the RTX 4070, the 4070 Super is 16% faster. Compared to the RTX 4070 Ti, the 4070 Super is just 5% slower than its Ti counterpart.
These results make a lot of sense; the alleged specifications of the RTX 4070 Super land its CUDA core count next to the RTX 4070 Ti. The RTX 4070 Ti features a CUDA core count of 7680, while the 4070 Super allegedly comes with 7168 CUDA cores, which is just 7.1% fewer than what the 4070 Ti comes with. By contrast, the RTX 4070 Super has 17.8% more cores than the RTX 4070.
| GPU | OpenCL Performance |
|---|---|
| RTX 4070 Ti | 206,646 |
| RTX 4070 Super | 195,384 |
| RTX 4070 | 168,268 |
Additionally, the RTX 4070 Super and RTX 4070 Ti allegedly come with identical L2 cache capacities (48 MB), further improving the RTX 4070 Super's performance. The RTX 4070 comes with 36MB of L2 cache. Memory capacity and bandwidth are identical on all three GPUs.
Suppose these results are indicative of real-world performance. In that case, the RTX 4070 Super will effectively be an RTX 4070 Ti clone, featuring performance so close to the 4070 Ti that you will be able to make up the difference with GPU overclocking. We saw this from the RTX 2060 Super, which performed practically identical to the RTX 2070.
| Header Cell - Column 0 | GPU | FP32 CUDA Cores | Memory Configuration | L2 Cache | TBP | MSRP |
|---|---|---|---|---|---|---|
| *GeForce RTX 4090 Ti | AD102 | 18176 (?) | 24GB 384-bit 24 GT/s GDDR6X (?) | 96 MB (?) | 600W (?) | Arm+Leg |
| GeForce RTX 4090 | AD102 | 16384 | 24GB 384-bit 21 GT/s GDDR6X | 72 MB | 450W | $1,599 |
| *GeForce RTX 4080 Super | AD103 | 10240 | 16GB 256-bit 24 GT/s GDDR6X | 64 MB | 320W | $999–$1,099 |
| GeForce RTX 4080 | AD103 | 9728 | 16GB 256-bit 22.4 GT/s GDDR6X | 64 MB | 320W | $1,199 |
| *GeForce RTX 4070 Ti Super | AD103-275/AD102-175 | 8448 | 16GB 256-bit 22.4 GT/s GDDR6X | 48 MB | 285W | $799–$849 |
| GeForce RTX 4070 Ti | AD104 | 7680 | 12GB 192-bit 21 GT/s GDDR6X | 48 MB | 285W | $799 |
| *GeForce RTX 4070 Super | AD104-350/AD103-175 | 7168 | 12GB 192-bit 21 GT/s GDDR6X | 48 MB | 225W | $599–$649 |
| GeForce RTX 4070 | AD104 | 5888 | 12GB 192-bit 21 GT/s GDDR6X | 36 MB | 200W | $599 |
| GeForce RTX 4060 Ti | AD106 | 4352 | 8GB/16GB 128-bit 18 GT/s GDDR6 | 32 MB | 160W | $399/$499 |
| GeForce RTX 4060 | AD106 | 3072 | 8GB 128-bit 17 GT/s GDDR6 | 24 MB | 115W | $299 |
At first glance, Nvidia's performance "tiering" makes no sense, with the 4070 Super so close to the 4070 Ti. But if Nvidia repeats its RTX 20 series Super formula to the tee, it will discontinue the RTX 4070 Ti and RTX 4080 in favor of their Super counterparts. This is inevitably why the RTX 4070 Super is so close to the RTX 4070 Ti in performance.
But we can't take these Geekbench benchmark results as complete truth. OpenCL is just one benchmark and won't tell us the whole story about the 4070 Super and its possible real-world performance. But we will know soon enough where the 4070 Super lays. Nvidia has hinted at new Ada Lovelace graphics cards at CES 2024.
There's a budget GeForce GPU selling in China that not even Nvidia knew it made — RTX 4010 turns out to be a modified RTX A400 workstation GPU
US to patch loopholes that allow China to buy banned AI GPUs from other countries — new regulations include national quotas on GPU exports and a global licensing system
