Tweet
Originally posted by stativ
View Post
-
Quite a crowd could play Solitaire on a dual Epyc system, remote desktop'ing in.. just an odd thought. -
everyone knows the issues with IO/latency with the Naples version of the Epyc processor. More interest in figuring out the chipset/io interface for the Rome platform.Comment
-
Congratulations on now being so significant that AMD has sent you the hardware for testing. Soon we should expect similar methods from Intel, Microsoft & other hardware manufacturers. The supplied hardware should have its individual items custom optimized, by the time you receive it, to show the best possible results. So your reports should be far better than we normal people use.
The other variation in the reported test results come from other differences: Linux kernel used, partition types (EXT4, ZFS, XFS & BTRFS). It'd be interesting if each of these two variables are strongly influencing the different results.
Good to see "SE" (Standard Error) on your later test results. Plus an overall summary of the many tests. Each test is either generalist (many components), or specialist (just a few components). So it is hard to know the meaning of the test results, except by the experts on bench tests.
Eventually we might see comparative results which include the RISC CPU's? More RISC CPU's are being used world-wide than CISC versions from Intel and AMD. This is in terms of total retail money sales, and numbers of units (smartphones, handhelds & IOT). However, for the moment, the big money spenders seem to not care about the relative bench test results of RISC-based computers. This should be changed in the next year or two.Comment
-
I'm very interested in Clear's IO performance... Just out of curiosity, was it a fresh installation before the testing? Will fstrim -av make any difference to the FIO scores?Comment
-
I'm a free software supporter, so I definitely want Windows Server to have a poor showing here.
But I think if Microsoft was smart, the best thing they could do to make Windows more performance competitive is replace ntfs with something faster. That's no easy thing, but clearly it's worthwhile. It seems like filesystem IO is their bottleneck on a lot of these benchmarks. (Edit: most obviously with the git benchmarks, but probably in some of the others too. Using git on Windows is a form of cruel and unusual punishment if your repository and commit history are anything other than trivially small.)Comment
-
MS owns Github. Alternative for MS would be making the github slower for Linux :PComment
Comment