Performance and Thermals
For performance testing, I installed the Viper Steel kit on our X299 test bench and ran through our test suite. Memory when it comes to performance testing is mostly a numbers game. The clock speed and timings make most of the difference with the capacity and number of sticks play roles as well, especially when it comes to jumping up to quad channel kits on X299 and Threadripper. Our specific kit is clocked at 3866 MHz while running the XMP settings with 18-22-22-40 for timings. For comparison the Dominator kit I tested previously was at a slower clock speed (3000MHz) but had much better timings (15-17-17-35) and the Ballistix Elite kit tested was a quad channel kit, this should give you a better idea of how timings, clock speeds, and single/dual/quad channels change performance. So because of the clock speed, this was the second fastest kit tested when it came to pure read speeds and to a lesser extent the write speeds, but copy speeds were a little lower.
The latency graph, on the other hand, is determined by both the timings, especially the CAS Latency number and the clock speeds. You can see the Dominators did well with their great timings but were still a little behind the Viper Steel kit.
Now memory performance can affect CPU performance, this is especially true with Ryzen which the infinity fabric ties the memory controller in with the timing of the CPU. To check how the kits do with this I ran Cinebench R15 where I did both the single core and multi-core tests and you can see that the timing of the Dominator kit well dominated here but the Viper Steel kit did really well as well. High bandwidth with the quad channel kits didn’t really help here with the exception of the single core test.
My last test is just an overall benchmark the Passmark Performance Test 9 Memory Mark test. This is just a portion of the normal Performance Test 9 benchmark but I like it because it is a combination of a lot of tests. They look at read and write speeds like we did as well as uncached tests. It also takes into account the total ram available and does a latency test. The threaded memory test looks at overall bandwidth and how the memory handles when operations get stacked up. Then there is a database operations test. Passmark then combines all of those results and puts them into an overall score. I’ve found that this does tend to favor memory capacity as you can see the 32GB kits score a little higher than the 16GB kits. But even looking at the 16GB kits I was surprised to find the Dominator kit as low as it was. The Viper Steel kit did well here when just compared with the other 16GB kits.
The last thing I took a look at was overall thermals. To do this I ran AIDA64’s stress test with the Stress system memory option. I ran the test for a half hour then took a look at the kit with our thermal camera. Unless you are overclocking, most memory don’t have too much of a problem staying cool, but I wanted to see if the heatspreader design was doing a good job of pulling the heat away from the memory and to look for any hotspots. The end result? Well as you can see that even after being stressed for a while things didn’t really heat up much. The sides which contact the RAM directly were a little warmer but only by a few degrees so that heat is transferring up into the thicker top section and the hotspots were only 3 degrees over the rest of the memory. You can see in the background how the VRMs on our motherboard are MUCH hotter, like I said memory cooling isn’t a big concern at stock voltages.