Does Spectre/Meltdown mitigation show up in E@H?
22 Jun 2018 1:05:27 UTC
Topic 215515
(moderation:
All the machines on my LAN that participate in BOINC, are amd64 cores (from AMD). Most of these machines have Polaris GPUs, which is something that E@H can use with Mesa (OpenCL-1.1).
I don't upgrade all my kernels at the same time, but I think I can see changes in slope for various kernel upgrades which revolve around Spectre/Meltdown issues.
Most recently, I brought up a Ryzen 1600 (with RX-560) to go along side a Ryzen 1600X (with RX-560). The Ryzen 1600 is only running the stock AMD cooler. But I would think that something about benchmarking machines out of your data could be if nothing else, interesting. The machines in question have different motherboards, and one has twice the RAM of the other. So, there are too many variables to confuse issues.
It is entirely possible that any of this possible analysis is moot by GDPR.
Now that the solstice is here, enjoy what most people call summer.
Hmm, no comments? I've
)
Hmm, no comments?
I've got 5 machines here running E@H (I believe). They don't all get kernel updates at the same time. I think I seen that when I updated the fastest machine to the 4.16 kernel, that the slope of the total of all my E@H computing as a function of time levelled off slightly (slope decreased). That fast machine has a Ryzen 1600x running on stock settings on an Asus Prime x370 Pro with 32GB of RAM using a NVMe SSD for the root partition.
When the Spectra and Meltdown kernel fixes first started coming out (4.9 kernel?), the systems I had stuttered to the point they were annoying to do desktop work on. So, to only see a performance change as compared to annoyances is still quite good. But it made me wonder if some place like E@H might actually have data such as this is how many FLOPs this job did on such a such kernel/CPU combination.
Is that any less opaque? :-)