Has anyone found Rocket Lake? Benchmark request.
Has anyone found Rocket Lake? Benchmark request.
Just been looking around at some local retailers. The model I want to buy is not available and I don't want to buy the higher priced ones, so I guess I'm waiting. There's one data point I'd really like but it is too obscure for mainstream reviews. Even though many are happy to use Prime95 for max power, it seems no one uses (or understands) the built in benchmark. Can I request a Prime95 throughput benchmark? Set min/max FFT size to 128k, test HT off, rest can be default, reduce the duration to minimum of 5 seconds if desired as it doesn't really make a difference with adequate cooling. Observe running clock during the benchmark, optionally reported power. Bench output is the maximum iter/s (usually n workers, n cores at this FFT size).
I obtained a data point from another source. For workloads similar to Prime95, Rocket Lake appears to offer roughly 50% more IPC than Skylake variants. The best estimate was around double, while the worst didn’t show improvement or even a decline. I wouldn’t dismiss the possibility of others having Rocket Lake results, but this gives me enough reason to consider testing a sample myself. I also plan to compare it with Skylake-X once I’m back home.
Today I received a sample and began testing it. The Prime95 benchmark at 128k FFT (one per core) indicates a performance boost of around 50% per core, per clock, versus AVX2 Intel CPUs Skylake or newer. This aligns with what I previously observed. This improvement is notable but falls short of the over 80% seen with Skylake-X. The efficiency per watt remained consistent with earlier AVX2 chips, suggesting AVX-512 still offers good power efficiency. Higher power consumption means more work done, which balances out. On my machine, the 11700k operates AVX-512 at 4.6 GHz with a voltage of 1.35V, which feels a bit excessive. The maximum observed power usage reached 216W. I’m considering setting a power cap around 150W to enhance efficiency without hurting performance significantly. This will be a test soon. I also ran some software called LLR, which uses the same math library as Prime95 for intensive tasks. While it provides similar results, there may be variations. This isn’t a final conclusion yet, but I estimated roughly a 42% improvement per clock compared to AVX2 here. Slightly lower than the 50% reported in the benchmark, possibly due to workload characteristics. Another point to explore is how much AVX-512 improves this type of workload. The 8-core 11700k delivers about double the throughput of the 6-core 10600k I had before. In today’s configuration, Zen 2 or Zen 3 should likely stay more efficient, though tuning might help bridge the gap.