5900x PPT, TDC, EDC

[maka0]

I have found my negative offset for each core 

Now I want to find the optimal PPT, TDC and EDC with similar performance to stock but hopefully lower voltage and temp

 

I have found keep PPT close to stock 142, gave me best results 

TDC reducing this gave me worst performance so I left it at stock 

EDC however, the lower I went the more benefit I got until your hit a certain number 

 

Stock values 

PPT - 142 

TDC - 90 

EDC - 140

CBR23 score: 20406

AVG voltage: 1.23

 

Stock values + curve optimiser

PPT - 142 

TDC - 90 

EDC - 140

CBR23 score: 21091

AVG voltage: 1.21

 

Optimised values 

PPT - 135

TDC - 90

EDC - 100

CBR23 score: 21154

AVG voltage: 1.19

 

My question, is lowering PPT, TDC and EDC harmful in anyway to the CPU ? Will it make it decreasse quicker over time? 

[freeagent]

I use around 200 130 145 and that gets me closer to 24K with voltage around 1.45-1.5+ depending on what’s going on.

 

But no, you won’t hurt it like that..

[maka0]

1 hour ago, freeagent said:

I use around 200 130 145 and that gets me closer to 24K with voltage around 1.45-1.5+ depending on what’s going on.

 

But no, you won’t hurt it like that..

Will running these PPT, TDC, EDC lower than stock cause stability issues? 

 

I've already stability checked all my curve optimiser on default PPT, TDC, EDC... 

 

Should I re check my curve for 

PPT - 135

TDC - 90

EDC - 100

?

 

 

[freeagent]

No but you are gimping it.

[YoungBlade]

1 hour ago, freeagent said:

No but you are gimping it.

The CPU is still performing above stock. That's not gimping it. It's maximizing efficiency to produce equivalent or better results while using less power.

[freeagent]

I was playing with mine last night and discovered I could drop edc waaay down too while maintaining my tippy top clocks

[maka0]

11 hours ago, freeagent said:

I was playing with mine last night and discovered I could drop edc waaay down too while maintaining my tippy top clocks

The real question now is, will it keep stability with a lower edc with no curve optimiser?

[freeagent]

7 minutes ago, maka0 said:

The real question now is, will it keep stability with a lower edc with no curve optimiser?

Sure, every CPU is different, so how much performance you are allowing it to have is dictated by your settings. It wont be unstable or anything like that. You don't have the tap wide open, so the hose will only spit so much out.

[Guest]

4 hours ago, freeagent said:

Sure, every CPU is different, so how much performance you are allowing it to have is dictated by your settings. It wont be unstable or anything like that. You don't have the tap wide open, so the hose will only spit so much out.

Really extending the EDC and PPT limits don't do a great deal because of temps. 

The "High Temperature Integer Threshold" is reset at 70c for AMD processor family 19h, the same as family 17h. 

This information can be read on page 353 of the "Processor Programming Reference" PPR for short. 

That means with a stock set up, 70c and up the cpu fans are ramped to 100% duty cycle. (This should tell us something?)

Anyhow, Enjoy  

[maka0]

1 hour ago, ShrimpBrime said:

Really extending the EDC and PPT limits don't do a great deal because of temps. 

The "High Temperature Integer Threshold" is reset at 70c for AMD processor family 19h, the same as family 17h. 

This information can be read on page 353 of the "Processor Programming Reference" PPR for short. 

That means with a stock set up, 70c and up the cpu fans are ramped to 100% duty cycle. (This should tell us something?)

Anyhow, Enjoy  

Okay with all this in mind
What is your recommendation for Ryzen 5000 in regards to PBO - PPT TDC EDC, curve optimiser vs stock settings?

[YoungBlade]

2 hours ago, maka0 said:

Okay with all this in mind
What is your recommendation for Ryzen 5000 in regards to PBO - PPT TDC EDC, curve optimiser vs stock settings?

I just use the lowest possible stable per-core values for the curve optimizer and keep the rest of the settings stock. Also, I have my fans ramp to 100% at 63C in the hopes of keeping it below 70C. Those get me just above 22,000 in Cinebench. Stock, I was getting just over 21,000.

 

If your cooling system is good enough, raising PPT, EDC, and TDC will let the CPU boost higher in all-core loads, but my computer is air cooled, so running at 180W+ instantly gets me over 70C and the result is only a minor improvement in R23. Perhaps this winter I'll open up the windows for an afternoon and see what happens with a much lower ambient temp, but otherwise, using those settings alone, that's all she's got.

[freeagent]

I run 200 ppt 130 tdc 180 edc and my fans ramp at 60. For something like OCCT that gets me 4700-4750 for an hour. R23 comes in the upper 23K mark, closer to 24K. I don’t bother with per core values because windows doesn’t use the cpu that way. At least not win 10. I set all core -30 +200.

[YoungBlade]

1 hour ago, freeagent said:

I run 200 ppt 130 tdc 180 edc and my fans ramp at 60. For something like OCCT that gets me 4700-4750 for an hour. R23 comes in the upper 23K mark, closer to 24K. I don’t bother with per core values because windows doesn’t use the cpu that way. At least not win 10. I set all core -30 +200.

Using those settings on my computer, OCCT Large, Variable, AVX2 held a sustained 4.7GHz all-core for 15 minutes and only reached a temperature of 58C. Not bad. My Cinebench R23 rose to 22,883 for a single run, which might seem underwhelming, but it's because my temps went up to a blazing 85C almost instantly. Power usage was at nearly 197W. Clocks tanked from the throttling. It was noticeable in the Cinebench run itself, because it went from "wow, that's fast" to "oh, nevermind" speeds.

 

Personally, I don't think an extra 800 points in R23 is worth 32% more power consumption and 20C higher temps, but that's just me. If I had a cooler more powerful than the Scythe Fuma 2, or at least put beefier fans on it, I'm sure I could reach close to the 24K mark, but the thermal constraints on the CPU are too great to overcome with my config. So I'll stick with stock power limits.

[maka0]

I have enabled PBO with the following 
PPT 140
TDC 95

EDC 113

 

Curver Optimiser set to negative 15

 

And I have seen a increase in peformance compared to stock PPT, TDC, EDC + CO

 

Question, I like to run my PC when Power Saver Mode when doing idle tasks, reddit/youtube/downloading/etc and change the Balanced when gaming/editting 
Will running the above PPT, TDC, EDC + CO cause any stability issues in Power Saver Mode?

 

Will I only see stability issue if I start to raise PPT, TDC, EDC beyond stock ?

[Guest]

6 hours ago, maka0 said:

Okay with all this in mind
What is your recommendation for Ryzen 5000 in regards to PBO - PPT TDC EDC, curve optimiser vs stock settings?

Based off that, whatever you do aiming for best efficiency with lowest temps possible. That's the idea of curve optimizer. To run a slightly lower v-core to get that extra 100 - 200mhz boost from it by reducing heat.

 

Or you find cpu cooling that gives you the thermal headroom, then you opt in for Increased PBO limit, +200mhz possibly more.

 

 

[maka0]

Just now, ShrimpBrime said:

Based off that, whatever you do aiming for best efficiency with lowest temps possible. That's the idea of curve optimizer. To run a slightly lower v-core to get that extra 100 - 200mhz boost from it by reducing heat.

 

Or you find cpu cooling that gives you the thermal headroom, then you opt in for Increased PBO limit, +200mhz possibly more.

 

 

I have ordered a Arctic 280mm AIO

So I will wait and see!

[Shoopman]

Sorry to necro, but what conclusions were you able to draw @maka0?

 

I want to try something similar to what you did, setting the TDC/EDC/PPT lower than normal to see if I can save power while also maintaining stock performance (or even a bit better).

 

Mostly concerned with single core performance because of gaming, but not many games use just one core anymore so I'm really looking at "several core" performance meaning I want the cores in use to go as high as possible, but not necessarily all of them.

[YoungBlade]

7 hours ago, Shoopman said:

Sorry to necro, but what conclusions were you able to draw @maka0?

 

I want to try something similar to what you did, setting the TDC/EDC/PPT lower than normal to see if I can save power while also maintaining stock performance (or even a bit better).

 

Mostly concerned with single core performance because of gaming, but not many games use just one core anymore so I'm really looking at "several core" performance meaning I want the cores in use to go as high as possible, but not necessarily all of them.

TDC, EDC, and PPT shouldn't impact your performance in lightly threaded workloads. Those values are maximums for the whole CPU. You have to drop them crazy low to have it impact performance in games that only use a few cores. I know this from personal experience: I needed to drop them to the values of the 35W TDP APUs before I started to lose some single core performance.

 

Have you already used curve optimizer to give a negative voltage offset? That noticeably impacted my boosting behavior in games, letting the CPU boost to 4.95GHz much more frequently.

[Shoopman]

1 hour ago, YoungBlade said:

TDC, EDC, and PPT shouldn't impact your performance in lightly threaded workloads. Those values are maximums for the whole CPU. You have to drop them crazy low to have it impact performance in games that only use a few cores. I know this from personal experience: I needed to drop them to the values of the 35W TDP APUs before I started to lose some single core performance.

 

Have you already used curve optimizer to give a negative voltage offset? That noticeably impacted my boosting behavior in games, letting the CPU boost to 4.95GHz much more frequently.

Yes, I have curve optimizer set to -30 on all cores. I haven't tested lightly threaded loads yet but I boost to 4.95GHz during Cinebench R23 single core. I'll spend some time today testing with various games and configurations.

 

 

[YoungBlade]

31 minutes ago, Shoopman said:

Yes, I have curve optimizer set to -30 on all cores. I haven't tested lightly threaded loads yet but I boost to 4.95GHz during Cinebench R23 single core. I'll spend some time today testing with various games and configurations.

If you want to test stability, I'd recommend using OCCT. Run the large, variable AVX2 stress test for 15 minutes on each core individually (advanced options, uncheck all but one core, also set it to check both physical and virtual, disable the other options). If errors show up, your curve offset is not truly stable.

[freeagent]

2 hours ago, YoungBlade said:

TDC, EDC, and PPT shouldn't impact your performance in lightly threaded workloads.

I didn’t think so either until I was trying to hit and maintain top boost clock on single core benchmarks. Those values do make a difference.. it’s not just for all core stuff.

 

But running at 5GHZ should be pretty easy, I think I can do it with near stock values, though tweaked slightly.

[YoungBlade]

1 minute ago, freeagent said:

I didn’t think so either until I was trying to hit and maintain top boost clock on single core benchmarks. Those values do make a difference.. it’s not just for all core stuff.

 

But running at 5GHZ should be pretty easy, I think I can do it with near stock values, though tweaked slightly.

I'm not talking about overclocking. If you hit 5GHz or more, then you aren't running stock anymore. For stock clocks, even down to the 65W TDP numbers (88 PPT, 60 TDC, 90 EDC), my Cinebench R23 single-core hardly moved and my boost clock was still hitting up to 4.95GHz and regularly sustaining over 4.8GHz - which is the rated max boost for the 5900X. Once I dropped it to the 35W TDP numbers (48 PPT, 30 TDC, 60 EDC) then, yes, I started to lose that boost performance in a meaningful way. I think my sustained single-core boost was only 4.6GHz at that point.

[freeagent]

As soon as you touch curve optimizer you are overclocking.

 

You only need to bounce off 4600 to see your R23 score, so you may have been hitting that high, but what were the effective clocks? 

[YoungBlade]

1 minute ago, freeagent said:

As soon as you touch curve optimizer you are overclocking.

 

You only need to bounce off 4600 to see your R23 score, so you may have been hitting that high, but what were the effective clocks? 

From the AMD definition, yes, but that is a bit weird considering I'm not raising the clockspeed. I would define overclocking as clocking something over its rated clockspeed, but the point I was making is that unless you do that, which is going to push more power into a single core, you have to drop PPT insanely low to have any impact on the single-core performance.

 

What do you mean by "effective clocks?" I'm actually running the 35W TDP test again. Most of the time, it's running at 4.725GHz in R23 single-core - it bounces between 4.6GHz and 4.825GHz in Cinebench R23. The final score was 1546, which is noticeably lower than I usually get, but not insanely lower than stock. This demonstrates again that those values hardly matter for lightly threaded tasks.

 

It does, however, make a huge difference to multi-core:

[freeagent]

5 minutes ago, YoungBlade said:

What do you mean by "effective clocks?

Ahh. its probably not a big deal anyways. 

 

The one thing I can ascertain is that none of these CPU's made by AMD are alike, nor are they consistent.