![]() ![]() Ray-tracing is a typical multithreaded test, with each ray being a potential thread in its own right ensuring that a workload can scale in complexity easily. The Core i7-7700K shows the benefits of frequency over a stock i7-6700K, however at the same frequency they perform roughly the same as expected. As a utility, similar to WinRAR, high thread counts, frequency and UPC typically win the day here. 7-zipĪs an open source compression/decompression tool, 7-zip is easy to test and features a built-in benchmark to measure performance. Again, Kaby Lake as a whole seems to do well here, thanks to 4.2 GHz turbo modes on the i5-7600K and i3-7350K. Similar to other tests, the i7-7700K takes the single thread crown, again beating an overclocked Devil’s Canyon i7-4790K, showing the out-of-the-box performance. We run the benchmark in an automated fashion three times in single-thread and multi-thread mode and take the average of the results. Cinebench R11.5ĬB11.5 has been popular for many years as a performance test, using easy to read and compare numbers that aren’t in the 1000s. This sort of approach to multithreading has a different approach to frequency, cores and IPC, hence why R11.5 and R15 do bigger separations with core workloads. If a thread finishes early, it will try and cut the work of another thread in half. Rather than the new Cinebench tests dividing the scene up into over a hundred pieces (depends on threads), CB10 purely divides the scene into exactly how many threads are present. The multithreaded test gives different results, as this test typically prefers many cores. Being high frequency the i7-7700K pulls out a lead here, but it’s worth noting that Kaby Lake as a whole scores well, perhaps indicating that other features (such as frequency speed changing) can help. The benchmark is similar to that of the newest R15 version, albeit with a simpler render target and a different strategy for multithreading.įor a few years I was under the impression that CineBench’s workload was not amenable to more IPC increases, as we hovered around 7000 pts with new microarchitectures not making much of a difference. The R10 version of Cinebench is one of our oldest benchmarks, with data going back more than a few generations. Nonetheless, our benchmark database spans to a time when that is all we had! We take a few of these tests for a pin with the latest hardware. At AnandTech, I’ve taken somewhat of a dim view to pure synthetic tests, as they fail to be relatable.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |