Application Performance

PCMark 10 has been included just to check general performance. There are no oddities here as the 2990WX is comparable to the Core i9-7960X which is perfectly acceptable for a super core heavy CPU. The 2950X does do much better though it comes in just behind the 1920X.

Excel is a real-world application we’ve been using to benchmark CPUs for year and years now, so why not try the Monte Carlo simulation with the 2nd gen Threadripper CPUs. As expected the 2950X is slightly faster than the 1950X, shaving 2% off the completion time. The 2990XW however is slower than the 2950X, basically matching the previous generation 16-core part, so a disappointing result there.

AMD’s Ryzen architecture has always excelled for decompression work and we see that the 2990WX is no exception, absolutely smashing the 7-Zip 32MB dictionary decompression test with a throughput of 187,000 MIPS. That said it was just 31% faster than the 2950X despite packing 100% more cores. Still AMD’s SMT implementation works wonders in this test and as a result the 2990WX was 116% faster than the Core i9-7980XE. But as impressive as the decompression performance is, compression performance leaves a lot to be desired.

For compression work the Core i9-7960X was 9% faster than the 2950X which isn’t bad for AMD, and it was nice to see the 2950X providing an 8% boost over the 1950X. The 2990WX though, well it fell into a heap. Despite 100% core utilization throughout the test performance was less than that of the 1920X and just 23% higher than the 8-core 2700X. This memory sensitive test gives us our first good look at how poorly things can go for the 2990WX, it’s pretty ugly and short of shutting cores off there appears to be no fix. I’ll look at this issue more closely in a moment, for now let’s continue with the application benchmarks.

VeraCrypt provides a built-in benchmark that allows users to test out how the various encryption algorithms perform on a given system. These tests aren’t run from local storage but rather in system memory and this proved to be a problem for the 2990WX. Now please note this benchmark uses 100% of the 2990WX, all 64-threads are fully loaded, so the odd performance isn’t a result of under utilization, though even then if that was the case it should at least match the 2950X.

The problem is what we have here is another memory intensive benchmark, though I should note it’s not that memory intensive. The 50MB test only loads just 50MB’s of data into the system memory so this has little impact on the 2990WX, even so it was still slower than the 2950X and even the 1950X.

The 1GB buffer test loads 1GB worth of data into the system memory and this proves to be a massive issue for the 2990WX, as it’s 32-cores become choked to the point where they’re significantly slower than the 2950X, offering half the performance. That’s pretty shocking to see.