AMD Ryzen 5 3600XT / Ryzen 7 3800XT / Ryzen 9 3900XT Linux Performance In 130+ Benchmarks

[phoronix]

Phoronix: AMD Ryzen 5 3600XT / Ryzen 7 3800XT / Ryzen 9 3900XT Linux Performance In 130+ Benchmarks

After the AMD Ryzen 3000XT series was announced last month, these new higher-clocked Zen 2 desktop processors are shipping today. Here are 130+ benchmarks on each of the Ryzen 5 3600XT, Ryzen 7 3800XT, and Ryzen 9 3900XT parts compared to various Intel and AMD CPUs. Tests under Ubuntu Linux and also complemented by performance-per-Watt / power and performance-per-dollar data points.

http://www.phoronix.com/vr.php?view=29370

[mlau]

Wow, the 3950x is perf-per-watt champion, didn't expect that.

[M@yeulC]

Thanks Michael for this long article, very insightful results. I am surprised at how much better the 3950x performs, but that makes sense, given it's 25% more expensive.

AMD has a strong offering, it's a shame you didn't plot the X-non-T variant on the performance/cost graph, though understandable if it is no longer sold.

I quite like the data visualization you used, but I feel it could be made a bit better. Those are just suggestions, but I'm curious to know what you think about them:

1. I like the thematic break-down. But using just mean is very misleading. Could you use something like your frametime graph that shows the distribution for each CPU instead, with benchmark results instead ?
   • That way, it would show 95th percentile, etc, preventing an outlier from changing the mean too much. Plotting mean and median on the same graph could be useful.
2. For the "seashell" graph, you use one CPU as the reference for each benchmark? However, it looks like you always pick the weakest one. How about always picking the same one? And maybe not the one weakest-on-average, but the median one, to give a better scale (thus adding inner circles at 75%, 50% performance)
3. For average, you just normalize results with a given CPU performance, right? I don't think there is an easy answer, but if I make a benchmark that triples the score for double the performance, that could skew results, unless you were to pick *two* reference points (and yet, it wouldn't help with more complex relationships, so separating them by domain is quite wise, plus this is likely how consumers think as well.

[Slartifartblast]

A somewhat underwhelming "upgrade" of the X series, wake me up when the Ryzen 4000 benchmarks land.

Techspot's conclusion hit's the nail on the head:

It’s our opinion that AMD is taking a page out of the Intel playbook and copying what they did with the Core i9-9900KS. We mean, they’re cashing in on improved yields, binning the better silicon, and selling it at a marked up price, while also removing the box cooler to maximize profit margins.

[pal666]

it must be painful to be intel these days

[pal666]

[I]It’s our opinion that AMD is taking a page out of the Intel playbook and copying what they did with the Core i9-9900KS. We mean, they’re cashing in on improved yields, binning the better silicon, and selling it at a marked up price

it only works if they sell both old and new variants. otherwise what would they do with binned worse silicon, throw it away?

[George99]

Michael, page 3 cryptography tests: I think something went wrong while testing the Ryzen 3900X this particular result doesn't make any sense.

[starshipeleven]

It’s our opinion that AMD is taking a page out of the Intel playbook and copying what they did with the Core i9-9900KS. We mean, they’re cashing in on improved yields, binning the better silicon, and selling it at a marked up price, while also removing the box cooler to maximize profit margins.

And this is bad because? This is the enthusiast level, where people pay 100$ more just because it has RGB leds.

Also who uses the stock cooler on a CPU that costs 400+ euro anyway. None complained when they didn't include a cooler for threadripper.

[starshipeleven]

it only works if they sell both old and new variants. otherwise what would they do with binned worse silicon, throw it away?

chiplets yo.

The chiplet with worse silicon goes in a Ryzen 5 or 3 or whatever. That's the one of the main benefits of not making a single die.

[loganj]

"Over the entire span of benchmarks, the Ryzen 9 3900XT saw a peak power draw of 144 Watts with an average of just 75 Watts compared to the Core i9 10900K having an average power draw of 133 Watts and a peak of 379 Watts."

nice amd