Ryzen 9 3900X, SMT ON vs. SMT OFF, 36 Game Benchmark

Ryzen 9 3900X, SMT ON vs. SMT OFF, 36 Game Benchmark

Disclaimer: Any pricing information shown or mentioned in this video was accurate at the time of video production, and may have since changed

Disclosure: As an Amazon Associate we earn from qualifying purchases. We may also earn a commission on some sales made through other store links

FOLLOW US IN THESE PLACES FOR UPDATES

[RonaldoFan-zzrk]

this video is a great demonstration of why exactly Steve is the benchmark king. some dudes test like 2-3 games, see some performance gain with smt disabled and rush to make a half-assed video spreading misinformation and people gladly buy it. while Steve dives deep into the subject, tests a very large amount of games and forms a reasonable conclusion to it. thank you so much Steve, we all love you

[zerospampls]

Could you please do a deep dive into memory Frequency & Latenzy scaling inkl. IF OC - that would be so awesome!

[althaz]

Wow, you guys are absolute beast-moding this Ryzen testing. Another day, another 72 test results!

[busyBaldurus]

Thank you Steve for answering this request. Love your work ❤️

[riba]

Thank you a lot for doing this test! It would also be nice to see the upcoming 3950x with SMT off :-)

[wisdoom]

Oh, how much I love, LOVE, sweet sweet blue-orange charts.





That's not a sarcasm, i love the benchmarks and deep explanation, one of few reasons to sub HWU.

[itst]

Take some time off mate. How much sleep are you gonna lose for us? Thanks for all the contents you bring us

[davidholomakoff]

Somebody send this to Moore's Law Is Dead... Hes saying in his videos 10% increase disabling SMT without doing any testing himself.

[konforzone]

IT can push cores more when oc'ing smt off (4.4-4.5ghz range) with good cooler which can push fps further compare to overclocked smt on 3900x

[skyhawk]

Thanks Steve!!!! Nothing was sugar coated in this video.... straight to the point with the proof....

[Lucifuriuz]

Thanks for all the hard work Steve, we appreciate you!

[KooYu]

Yes! I was waiting for this, great timing Steve!

[soraaoixxthebluesky]

If I’m not mistaken 9900K with HT disable, you can further push it’s clock limits. But as shown in this video, some titles favour more thread and some don’t and even lose performance at some point.

[Bogdan]

For those who are new to the PC terms: SMT=Simultaneous multi threading

[Najvalsa]

Could you please do a test with one of the CCD's disabled? The slower one if there really is a slower and faster one.
Also, the 3600 vs a 9700K or 9900K would be an interesting test. How much more do you get in gaming by spending an extra ~$200 - 350 (when factoring in a ~$50 aftermarket cooler)?

But first, have a rest, mate. Thanks for the hard work.

[bradmorri]

Absolutely agree with your assessment that the dual interconnects spread graphics memory access load over two interconnects as opposed to a single interconnect.


If you Do the maths on each 32Byte/cycle interconnect. At 1800Mhz, each interconnect has a total theoretical bandwidth of 56.7GB/s that must be shared by both the CPU getting memory data and the GPU and its need for about 15GB/s of data to render frames to the monitor. The dual dies chips can potentially spread that load over two interconnects. Less threads, without SMT on, also potentially leave more bandwidth for the GPU in some games.


ON the other hand, Intel Ringbus with a x42 multiplier shares the ring with cpu and GPU as well but has 144GB/s bandwidth to play with.


An interesting experiment might be to rebench the two top performers and the two bottom performers and record the utilisation over time of each logical CPU over the benchmark run for both smt on and off.
Load the CPU utilization data into Excel and create a 3D graph that covers the utilization of each core over time. It makes it really easy to visualize how the application allocates threads to different cores on the CPU.


ON older Ryzen some games would peg a core at 100% if you let windows manage things but smooth things out if you managed affinity. Maybe this is happening here and smt off fixes things? Maybe one game is forcing everything on one die and the other game is spreading the load to both dies? I have no idea exactly what you will find but i suspect that the good and the bad will show measurable differences in the way the threads are being distributed on the two die in the package.

[johnmellinger]

Awesome work Steve! I am picking up my 3900X on Wed. However I am still waiting for the G-Skill memory I want to launch. The 3600 CL14 Neo Ram has peaked my interest!

[prague]

Excellent and timely analysis. Thank you for being so thorough. Question answered.

[vladislavkaras]

Thanks for such great benchmarks!
Would be interesting to see the same results with modern CPUs and with low core count CPUs as well!

[mongotrip]

great video! but i have to say, 9900k without HT would have been also very interesting