General Discussions

[black_zion]

It's a pretty thorough test by TomsHardware, but the upcoming AGESA update will determine how much truth there is behind it.

https://www.tomshardware.com/news/amd-ryzen-3000-boost-clock-controversy-intel-attack,40231.html

[ajlueke]

The strange thing about this article, and all these analyses in general, is that they seem to home in on Shamino saying the boost changes were made for reliability, and then supporting that with "the Stilt's" comments on reduced temperature thresholds.

Yes, reduced temp thresholds could lengthen chip life, and also reduce boost clocks to stay in that envelop, the only problem is, that is true for all core boosts.

The issue is, users aren't seeing the advertised boost clocks in lightly threaded workloads, a situation where you are no where near the temp thresholds referenced by the Stilt.  So even if AMD did make adjustments to keep Zen 2 within a lower temperature, why would that affect the single core boosts?  It may have lowered our all core boosts, but that isn't the issue users are complaining about. 

[black_zion]

Clarification should be provided tomorrow by AMD, but I think it all stems with the decision to use mismatched cores causing conflicts with the boost states and wreaking havoc behind the scenes. I also have to think that SMT is playing a role, along with the OS scheduler. What I'd like to see is a comparison of boost speeds under each of the following circumstances under Cinebench R20 and R15 single thread:

• SMT on, Affinity set to all
• SMT off, Affinity set to all
• SMT on, Affinity set to fastest core only
• SMT off, Affinity set to fastest core only

Take SMT and the scheduler out of the picture and see if it makes any difference.

Also I still wouldn't be surprised if the processors are actually hitting full boost speeds, but are jumping between the full boost and some lower boost state so quickly it's not able to be reported, since AMD touted Windows 10's new scheduler allowing states to jump much faster than before.

[ajlueke]

There is definitely something going on, as per my previous testing, my 3900X hits 4.5 GHz and applies about 1.48V when precision boost is allowed to do the boosting.  If I manually set the fastest core to 4.6 GHz and give it the same voltage, it hits that clockspeed.  I know that isn't clock stretching, as I see a performance uptick in CB20 of over 3%.

So my chip at least, can hit the advertised clocks with the voltage provided, it just doesn't when the boost algorithm is in play.  It seems the algorithm is trying to use vastly more voltage than it needs to.  The 4.5 GHz at 1.48V, I was able to replicate the clockspeed and performance using only 1.3V manually.

[hardcoregames_]

I am going to post an article on my observations, if I had more processors I could have done this earlier but I am not that rich

[ajlueke]

Update over on the gaming blog.

https://community.amd.com/community/gaming/blog/2019/09/10/ryzen-community-update-bios-updates-for-b...

[qwixt]

"AGESA 1003ABBA beta BIOS"

I guess someone was a "Dancing Queen" fan.

[black_zion]

What's funny: https://www.tomshardware.com/news/intel-apollo-lake-refresh-degradation-cpu-failure,40362.html

[ajlueke]

Well, I guess Intel didn't make baseless accusations.  They clearly have experience with this phenomenon.

[black_zion]

Don't forget 2 years ago: https://www.tomshardware.com/news/intel-cpu-failure-atom-processor,33538.html

[hardcoregames_]

Those problems were blow way out of proportion

[hardcoregames_]

I inquired with Intel and they said that if anyone's CPU is affected and croaks they will fix it. Desktop models can have the CPU replaced but mobile machines tend to heave a soldered processor.

My laptops all have Intel processors and they all still work and they all can run Windows 10 fine. Most as I spent a lot of pesos on RAM to upgrade them.