Question AMD 9900X compilation speed

[kalmquist]

I was wondering whether I should spend an extra $90 to get an AMD 9900X (12 core) instead of a 9700X (8 core), mainly to be able to compile software faster. But the Anandtech benchmarks for Linux kernel and PHP compilation have the 9900X slower than the 9700X. Here is are the number for all four AMD Zen 5 CPU's:

CPU kernel PHP
--------------------
9950X 85.41 60.44
9900X 101.64 63.06
9700X 74.29 47.44
9600X 88.95 53.63

My assumption is that something went wrong when the 9900X and 9950X were benchmarked. The two compilation benchmarks are the only anomalies; the 9900X is faster than the 9700X on everything else. But I’ll ask: does anyone think that the 9900X actually performs badly when compiling software?

 

[gdansk]

It is hard to explain that result but here's some more benchmarks with closer to expected results

AMD Ryzen 9 9950X & Ryzen 9 9900X Deliver Excellent Linux Performance Review - Phoronix

www.phoronix.com

 

[Markfw]

It is hard to explain that result but here's some more benchmarks with closer to expected results

AMD Ryzen 9 9950X & Ryzen 9 9900X Deliver Excellent Linux Performance Review - Phoronix

www.phoronix.com

First: I trust phoronix more than just about everyone else. Second, if that first test was using windows, maybe it didn't have the windoz fix in it.

 

[kalmquist]

Thanks. I should check the Phoronix site more often, especially now that this site is no longer getting new content. Phoronix shows that for kernel compilation, the 9900X is 52% faster that the 9700X but takes 15% more power. Unfortunately, Phoronix doesn't do the math, and embeds the numbers you need to do the math in images so you can't cut and paste them, which is one reason I haven't been going to the site.

 

[Tech Junky]

9900X

Take the numbers for the 7900X and deduct 10-15% from the time. That's about all of the performance boost the 9 series offers.

 

[igor_kavinski]

AMD is supposed to be readying some inter-CCD latency mitigation patch that may improve performance on 9900X/9950X but my impression so far is to prefer the 9700X. It's the one with the most consistent performance gains of the Zen 5 family so far.

 

[MS_AT]

AMD is supposed to be readying some inter-CCD latency mitigation patch that may improve performance on 9900X/9950X but my impression so far is to prefer the 9700X. It's the one with the most consistent performance gains of the Zen 5 family so far.

Hard for me to speak about PHP, but for CPP/C interCCD latency is meaningless as you spawn an independent compiler process per file and since multi-threaded linkers are seldomly used it's in the end single threaded bound activity. What is more important is how build system is able to cope with scheduling work and what code base you are compiling. With efficient build system and when using MT linker you should be able to see nice scaling from 9700x to 9950x.

 

[igor_kavinski]

With efficient build system and when using MT linker you should be able to see nice scaling from 9700x to 9950x.

In general that should be fine for someone new to 9950X but if they are going from 7950X to 9950X, it's a bit hard to recommend that right now due to the performance anomalies with 9950X.

 

[MS_AT]

In general that should be fine for someone new to 9950X but if they are going from 7950X to 9950X, it's a bit hard to recommend that right now due to the performance anomalies with 9950X.

I would call them test setup anomalies, really. Then I don't see original poster mentioning anything about doing the transition from 7950x to 9950x, only asking about anomalies in AT benchmark database. To be honest in ideal case you find out a way to benchmark your workload on the CPU in question or you pick one reviewer and try to fit their benchmark scores to the scores you get by running the same workload they do etc.