Ryzen Owners Zen Garden

[dont whant to set it"']

@tabascosauz I'm playing with it these days in the bios with all-core oc and down-core in this case.

Yes, if I hit stop , then start benchmark again the score goes back to above 3000 points.

 

[R-T-B]

So... I just pulled the trigger on a 5950X. My early holiday update.

Quick question to you all Ryzen peoples:

Is swapping out Ryzen CPUs on Windows 11 still producing anomalies with the cache latencies? If so, is there a way to avoid that? Any articles on this?

 

[Mussels]

So... I just pulled the trigger on a 5950X. My early holiday update.

Quick question to you all Ryzen peoples:

Is swapping out Ryzen CPUs on Windows 11 still producing anomalies with the cache latencies? If so, is there a way to avoid that? Any articles on this?

I had a guess that you could do an in-place upgrade install by using the setup.exe from an ISO, but it's untested

 

[freeagent]

Win 11 totally T-Bags my L3 performance in Aida. A clean install is a little better than an upgrade from 10 with my system. Everything else is good though, but most CPU benches are lower compared to 10 with my system.

Hoping the next leap will be the leap home.

 

[tabascosauz]

So... I just pulled the trigger on a 5950X. My early holiday update.

Quick question to you all Ryzen peoples:

Is swapping out Ryzen CPUs on Windows 11 still producing anomalies with the cache latencies? If so, is there a way to avoid that? Any articles on this?

Win 11 totally T-Bags my L3 performance in Aida. A clean install is a little better than an upgrade from 10 with my system. Everything else is good though, but most CPU benches are lower compared to 10 with my system.

Hoping the next leap will be the leap home.

Ryzen 5000 L3 performance has never been completely fixed in Windows 11, it doesn't have anything to do with swapping CPUs. I very much respect HWUB, but they were barking up the wrong tree with that tweet, unless swapping CPUs makes it noticeably even worse than it already is.

22000.282 was supposed to be the fix, didn't do a thing. 22000.318 (KB5007215) again was supposed to be a fix, didn't do a thing.

All you have to do is run the L3 test repeatedly (double click the "L3 Cache" text). On clean Windows 10, every single run should be within margin of error. On Windows 11, all 4 numbers will just do whatever they want. It's quite the night and day difference between 10 and even the latest 11. Here's a few consecutive runs (I did 15 runs, but here's some highlights):

 

[R-T-B]

Ryzen 5000 L3 performance has never been completely fixed in Windows 11, it doesn't have anything to do with swapping CPUs. I very much respect HWUB, but they were barking up the wrong tree with that tweet.

22000.282 was supposed to be the fix, didn't do a thing. 22000.318 (KB5007215) again was supposed to be a fix, didn't do a thing.

All you have to do is run the L3 test repeatedly (double click the "L3 Cache" text). On clean Windows 10, every single run should be within margin of error. On Windows 11, all 4 numbers will just do whatever they want. It's quite the night and day difference between 10 and even the latest 11. Here's a few consecutive runs (I did 15 runs, but here's some highlights):

View attachment 225080View attachment 225081View attachment 225082View attachment 225083

Yeah. I guess the real question is does it show outside of synthetics like before?

 

[tabascosauz]

Yeah. I guess the real question is does it show outside of synthetics like before?

ST benchmarks are slightly lower, MT benchmarks bout the same here. Haven't noticed anything adverse in games, if anything all my games seem to run better in 11.

CPU clocks in game are way down in Win 11 though. Allegedly Shadow of the Tomb Raider benchmark is noticeably affected in Win 11, allegedly because of the L3.

 

[R-T-B]

ST benchmarks are slightly lower, MT benchmarks bout the same here. Haven't noticed anything adverse in games, if anything all my games seem to run better in 11.

CPU clocks in game are way down in Win 11 though. Allegedly Shadow of the Tomb Raider benchmark is noticeably affected in Win 11, allegedly because of the L3.

Since I do an allcore OC on my Ryzen for my highly threaded compiles, I sort of doubt it will impact me then. But I can certainly see how it's not completely fixed then.

 

[Mussels]

Since I do an allcore OC on my Ryzen for my highly threaded compiles, I sort of doubt it will impact me then. But I can certainly see how it's not completely fixed then.

all core OC will likely bypass all these issues

 

[ViperXTR]

 

[tabascosauz]

ninja'd 3 seconds before I was going to hit post

Not going to comment on him using Cinebench to infer all-core stability (lol), but both sides have a point. Ryzen 5000 definitely looks much more resistant to degradation as a whole (unannounced N7P?). At the same time, early Ryzen 3000 owners trying the same, especially owners of bad chips, probably ends in disaster. I remember one or two Renoir APUs dropping freq after just 30 min or so of cursed Clocktuner automatically "testing" with 1.5V, I know I definitely lost some freq on my 4650G after a week of Cinebench at 1.45V.

Obviously other factors such as temp, the fact that it was really only ~1600hr of balls-to-the-wall testing, nothing said about LLC and droop, etc. But good to know that Ryzen 5000 can take it, since can't really measure FIT voltage anymore with how P95 is throttled now.

 

[freeagent]

I have to agree.. 1.25v should be ok for all loads, 1.3-1.35 for most loads, and 1.4-1.45 for fringe type stuff, short term..

 

[ViperXTR]

mine shoots up momentarily to this on single core (2 of my best cores according to ryzen master) , but generally 1.2v on all core stress

 

[tabascosauz]

mine shoots up momentarily to this on single core (2 of my best cores according to ryzen master) , but generally 1.2v on all core stress View attachment 225179

Single thread boost is irrelevant as long as it stays below 1.5V sustained. It's all well within stock operating parameters.

VID is not relevant to what's actually happening. SVI2 TFN Vcore shows accurately in an actual single core test, but isn't smart enough for more transient boosts or idle. You may also be able to see a socket sense Vcore under the board sensors section, off the SuperI/O on the Mortar but it might not be as accurate. Otherwise, Ryzen Master Vcore should be accurate in all conditions.

 

[Mussels]

Oh good, someone finally got reliable testing!


My 3700x seemed to degrade fast, so it went to PBO settings really rapidly.

My 2700x is using pieces of SATA cable to replace three of its pins so... it's not really a good case study.

 

[dyonoctis]

So my 3700x with PBO turned off boosted to 4.7Ghz. Every core somehow managed to go above 4.6GHZ...

 

[Deleted member 215115]

So my 3700x with PBO turned off boosted to 4.7Ghz. Every core somehow managed to go above 4.6GHZ...

That's not real. Don't use HW Monitor. Use HWInfo64.

 

[Mussels]

So my 3700x with PBO turned off boosted to 4.7Ghz. Every core somehow managed to go above 4.6GHZ...
View attachment 226272

HWmonitor gives bullshit numbers, use HWinfo

 

[Space Lynx]

HWmonitor gives bullshit numbers, use HWinfo

Yes, this 100%.

 

[tabascosauz]

Okay, after spending a number of weeks with 11, there seems to be some fundamental changes in how 11 handles CPU-bound games, that no one seems to have talked about. At least on Ryzen.

Win 10:

• Always clocks higher.
• Starts off with one or both preferred cores, or Core 0.
• If more threads needed, this expands to include the rest of CCD1. Usually stops here.
• Core 7 always used for background tasks unless all 12 cores in action (extremely rare).

Win 11:

• Always clocks lower.
• Starts off with Core 7 mostly.
• If more threads needed, it'll bring the two preferred cores online for 2 or 3 cores active (+SMT).
• Then if it needs more threads, it'll just steadily bring more cores online until the load is spread evenly across all 12 cores. But at low usage/volts/wattage/temp per core or thread.

The new propensity for all-core seems to help some games that weren't coded to take advantage of >1 or >2 cores in Win 10. MW2019/BFV are unchanged and run a bit cooler, CSGO/Arma 3/BL2/MCC/ROR2/War Thunder/Genshin are okay, but they all run on Core 7.

That is, until we get to Insurgency: Sandstorm. Performance is very poor only on 11. Game often rapidly jumping between 70-165Hz a dozen times a second, and not even in response to anything. Game runs almost exclusively on Core 7. No other variables aside from Win 10 vs. 11. After exhausting the game and NVCP settings it's probably unfixable, so I went back to 10 for the final time.

Yes, core affinity can be set through Task Manager, but sometimes it doesn't do anything, sometimes it just makes the game run like shit, or sometimes is simply not possible and gives "Access Denied" message (Insurgency).

Microsoft - look at this hierarchy and tell me with a straight face that always putting 70% of the load on Core 7 is CPPC2 'working as intended':

 

[freeagent]

Okay, after spending a number of weeks with 11, there seems to be some fundamental changes in how 11 handles CPU-bound games, that no one seems to have talked about. At least on Ryzen.

Win 10:

• Always clocks higher.
• Starts off with one or both preferred cores, or Core 0.
• If more threads needed, this expands to include the rest of CCD1. Usually stops here.
• Core 7 always used for background tasks unless all 12 cores in action (extremely rare).

Win 11:

• Always clocks lower.
• Starts off with Core 7 mostly.
• If more threads needed, it'll bring the two preferred cores online for 2 or 3 cores active (+SMT).
• Then if it needs more threads, it'll just steadily bring more cores online until the load is spread evenly across all 12 cores. But at low usage/volts/wattage/temp per core or thread.

The new propensity for all-core seems to help some games that weren't coded to take advantage of >1 or >2 cores in Win 10. MW2019/BFV are unchanged and run a bit cooler, CSGO/Arma 3/BL2/MCC/ROR2/War Thunder/Genshin are okay, but they all run on Core 7.

That is, until we get to Insurgency: Sandstorm. Performance is very poor only on 11. Game often rapidly jumping between 70-165Hz a dozen times a second, and not even in response to anything. Game runs almost exclusively on Core 7. No other variables aside from Win 10 vs. 11. After exhausting the game and NVCP settings it's probably unfixable, so I went back to 10 for the final time.

Yes, core affinity can be set through Task Manager, but sometimes it doesn't do anything, sometimes it just makes the game run like shit, or sometimes is simply not possible and gives "Access Denied" message (Insurgency).

Microsoft - look at this hierarchy and tell me with a straight face that always putting 70% of the load on Core 7 is CPPC2 'working as intended':

View attachment 226791

That is pretty much what I saw too.. I thought I was losing it a bit.. Doubting known good settings.. starting fresh.. I will go back to 11 at some point. So pretty must touch..

 

[Mussels]

Okay, after spending a number of weeks with 11, there seems to be some fundamental changes in how 11 handles CPU-bound games, that no one seems to have talked about. At least on Ryzen.

Win 10:

• Always clocks higher.
• Starts off with one or both preferred cores, or Core 0.
• If more threads needed, this expands to include the rest of CCD1. Usually stops here.
• Core 7 always used for background tasks unless all 12 cores in action (extremely rare).

Win 11:

• Always clocks lower.
• Starts off with Core 7 mostly.
• If more threads needed, it'll bring the two preferred cores online for 2 or 3 cores active (+SMT).
• Then if it needs more threads, it'll just steadily bring more cores online until the load is spread evenly across all 12 cores. But at low usage/volts/wattage/temp per core or thread.

The new propensity for all-core seems to help some games that weren't coded to take advantage of >1 or >2 cores in Win 10. MW2019/BFV are unchanged and run a bit cooler, CSGO/Arma 3/BL2/MCC/ROR2/War Thunder/Genshin are okay, but they all run on Core 7.

That is, until we get to Insurgency: Sandstorm. Performance is very poor only on 11. Game often rapidly jumping between 70-165Hz a dozen times a second, and not even in response to anything. Game runs almost exclusively on Core 7. No other variables aside from Win 10 vs. 11. After exhausting the game and NVCP settings it's probably unfixable, so I went back to 10 for the final time.

Yes, core affinity can be set through Task Manager, but sometimes it doesn't do anything, sometimes it just makes the game run like shit, or sometimes is simply not possible and gives "Access Denied" message (Insurgency).

Microsoft - look at this hierarchy and tell me with a straight face that always putting 70% of the load on Core 7 is CPPC2 'working as intended':

View attachment 226791

my brother brought up issues with sandstorm as well, but as i dont play i had no answers for him

I would have suggested trying process lasso and seeing it it solved the issue (and then all the various chipset drivers, blah blah)

 

[YautjaLord]

Might've underestimated how long it will take me to build the system, but overall as i said it'll be up & running by end of December, the video might just be out in January 2022. Will have 2 more components (RTX 3080 Master & 980 Pro 1TB) by December 12, for now. Why 1TB & not 2? Moneyz cash 1st of all, but also not too many games in my Steam library to fill up 1TB, let alone 2.

Pics of 2 waterblocks + spare parts, all EKWB:

P.S. Yes, that's me in reflection of Vector Xtreme.

P.P.S. Shot with Galaxy A71.

 

[Mussels]

Might've underestimated how long it will take me to build the system, but overall as i said it'll be up & running by end of December, the video might just be out in January 2022. Will have 2 more components (RTX 3080 Xtreme & 980 Pro 1TB) by December 12, for now. Why 1TB & not 2? Moneyz cash 1st of all, but also not too many games in my Steam library to fill up 1TB, let alone 2.

Pics of 2 waterblocks + spare parts, all EKWB:

View attachment 227011View attachment 227012View attachment 227013

P.S. Yes, that's me in reflection of Vector Xtreme.

P.P.S. Shot with Galaxy A71.

Those are the same parts i have, i think we're family now

Wait holy crap your PC is so close to matching one of mine, it's like every spec is one number off
Are you me, from another dimension? (Or do we just have good taste ;P )

 

[tabascosauz]

my brother brought up issues with sandstorm as well, but as i dont play i had no answers for him

I would have suggested trying process lasso and seeing it it solved the issue (and then all the various chipset drivers, blah blah)

Unfortunately it looks like Process Lasso just functions in the same way as Task Manager. So programs that work still work, programs that crash still crash, and programs can't set affinity still can't set affinity.