Ryzen: Strictly technical

[Gyronamics]

Ever considered that it's Gigabyte (who has the worst bioses in the industry) who's playing with the stock settings?

Techpowerup who made their tests on MSI X470 M7 also got > 1800pts in Cinebench R15, meaning that PBO was definitely enabled.

https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/9.html
Also these CPUs don't have any 105W limit.
The minimum limit is the 141.75W, even when you disable PBO manually.

Not something I considered due to few sources (two I found) which mention PSO settings and the power draw of the CPU.

The lack of information makes it appear that PSO being on or off is the reason for exceeding TDP.

So you would suggest that Gigabyte is enforcing the 105W TDP with their BIOS and other boards should see 141.75W with PSO disabled.

Ignore that, tomshardware was using a MSI X470 Gaming M7 AC board with PSO disabled to get the 105W reading.

Gigabyte was being used by no one, I only mentioned the Gigabyte board because it was showing a warning to enable PSO.

So Gigabyte doesn't have PSO as default and MSI is not exceeding TDP with PSO disabled.

Both of them are not using the settings you mention so stock or PSO disabled apparently has different meaning depending on the brand you are using.

 

[majord]

If S3 is a standby state, yes, it's broken on my MB. I run with SMT disabled and -- no sleep for me!

yes that's it, thank you. that's a shame.

 

[dmnk7]

Ever considered that it's Gigabyte (who has the worst bioses in the industry) who's playing with the stock settings?

Techpowerup who made their tests on MSI X470 M7 also got > 1800pts in Cinebench R15, meaning that PBO was definitely enabled.

https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/9.html
Also these CPUs don't have any 105W limit.
The minimum limit is the 141.75W, even when you disable PBO manually.

This is what AMD is saying

Precision Boost Overdrive enables operation of the processor outside of specifications and in excess of factory settings, we anticipate use of the feature will invalidate the AMD product warranty and may also void warranties offered by the system manufacturer or retailer. Subject to change.

So it doesn't look like stock.

 

[Abwx]

There s no review that display those 140W, Hardware.fr whose numbers cant be much discussed, got 129W@Prime95 at the ATX and with a MB that increase voltage 50mV above stock.

All things being equal this amount to 103W at the CPU level when accounting for this excess voltage, the CPU could eventualy take as much as 115W in X264 encoding involving AVX2, at wich point it can consume more than in Prime 95@256K FFT...

FTR in Prime 95 the CPU did run at 4GHz at stock settings.

https://www.hardware.fr/articles/974-7/overclocking-pratique.html

https://www.hardware.fr/medias/photos_news/00/55/IMG0055234.png

https://www.hardware.fr/medias/photos_news/00/55/IMG0055238.png

 

[IRobot23]

There s no review that display those 140W, Hardware.fr whose numbers cant be much discussed, got 129W@Prime95 at the ATX and with a MB that increase voltage 50mV above stock.

All things being equal this amount to 103W at the CPU level when accounting for this excess voltage, the CPU could eventualy take as much as 115W in X264 encoding involving AVX2, at wich point it can consume more than in Prime 95@256K FFT...

FTR in Prime 95 the CPU did run at 4GHz at stock settings.

https://www.hardware.fr/articles/974-7/overclocking-pratique.html

https://www.hardware.fr/medias/photos_news/00/55/IMG0055234.png

https://www.hardware.fr/medias/photos_news/00/55/IMG0055238.png

Yes, @The Stilt got above the average.
https://www.tomshardware.com/reviews/amd-ryzen-7-2700x-review,5571-12.html

 

[The Stilt]

Cannot seem to find what method the different reviews use to measure the power consumption.

As I said before, I'm using the telemetry from the VRM controller itself (DCR measurements, ±2% accuracy according to the controller specs).
If the other reviews are using the power reported by the CPU itself for the cores and the package (i.e. calculated power), then obviously the values are not comparable.
Next I'll see what's the difference between the reported and the measured power, if there is one.

 

[mat9v]

Cannot seem to find what method the different reviews use to measure the power consumption.

As I said before, I'm using the telemetry from the VRM controller itself (DCR measurements, ±2% accuracy according to the controller specs).
If the other reviews are using the power reported by the CPU itself for the cores and the package (i.e. calculated power), then obviously the values are not comparable.
Next I'll see what's the difference between the reported and the measured power, if there is one.

This is how TomsHardware is measuring their power use: https://www.tomshardware.com/review...ent-cpu-gpu-components-powenetics,5481-3.html
This is measured between power supply and 12V connectors on the motherboard, so if they show low power use... on the other hand, the power use they report is suspiciously low, especially compared to typical readings "at the wall".

 

[The Stilt]

This is how TomsHardware is measuring their power use: https://www.tomshardware.com/review...ent-cpu-gpu-components-powenetics,5481-3.html
This is measured between power supply and 12V connectors on the motherboard, so if they show low power use... on the other hand, the power use they report is suspiciously low, especially compared to typical readings "at the wall".

The term ("package power consumption") they use in the 2700X review would indicate that they used the power figures reported by the CPU.
Also as you said, 104.7W as VRM input power would mean ~89W consumption for the CPU itself. Which obviously is not going to happen in Prime95.

I'm looking into it.

 

[The Stilt]

Based on the tests I made, the "CPU Package Power (SMU)" value reported by the most recent version of HWInfo is very accurate.

I swapped the CPU to a specimen which has an average SIDD characteristics of those samples I currently have, so that silicon related anomalies can be excluded.

Precision Boost Override = Disabled
Precision Boost Override Scalar = Disabled

Spoiler

Limits programmed manually to 105W (PPT), 95A (TDC) and 140A (EDC).

Spoiler

Performance at manual 105W limit in CB15 nT:

Spoiler

The only way to see 104.7W ish "Package Power" quoted by Tom's Hardware is to actually manually limit the PPT to 105W, instead of the default 141.75W (which is used when PBO is disabled).
Tom's claim Cinebench R15 nT score of 1859, meaning that obviously they neither did manually set the PPT to 105W nor have PBO disabled.

Silicon variance might normally explain part of the difference, however not in the case of Pinnacle Ridge.
Unless the PPT was limited to 105W (which they claim to be seeing), the CPU would increase it's clocks (and utilize the headroom) until any of the limiters (PPT, TDC, EDC or temps) are reached.
The power consumption wouldn't get stuck to 105W. Obviously the CPU will not increase the clocks infinitely, however the 2700X boosts up to 4.15GHz on all cores (when not restricted by the limiters) and I highly doubt that they hit the limit in Prime95.

Based on that, I said their power measurements were botched.

 

[mat9v]

Based on the tests I made, the "CPU Package Power (SMU)" value reported by the most recent version of HWInfo is very accurate.

I swapped the CPU to a specimen which has an average SIDD characteristics of those samples I currently have, so that silicon related anomalies can be excluded.

Precision Boost Override = Disabled
Precision Boost Override Scalar = Disabled

Limits programmed manually to 105W (PPT), 95A (TDC) and 140A (EDC).

The only way to see 104.7W ish "Package Power" quoted by Tom's Hardware is to actually manually limit the PPT to 105W, instead of the default 141.75W (which is used when PBO is disabled).
Tom's claim Cinebench R15 nT score of 1859, meaning that obviously they neither did manually set the PPT to 105W nor have PBO disabled.

Silicon variance might normally explain part of the difference, however not in the case of Pinnacle Ridge.
Unless the PPT was limited to 105W (which they claim to be seeing), the CPU would increase it's clocks (and utilize the headroom) until any of the limiters (PPT, TDC, EDC or temps) are reached.
The power consumption wouldn't get stuck to 105W. Obviously the CPU will not increase the clocks infinitely, however the 2700X boosts up to 4.15GHz on all cores (when not restricted by the limiters) and I highly doubt that they hit the limit in Prime95.

Based on that, I said their power measurements were botched.

Thank you for your explanation Are the tools you are using to set power limits "propriety" or can you share them? It would be very useful to be able to lock power use without constant rebooting.

 

[Abwx]

Apparently the cooler thermal resistance is about 0.43°C/W, hence a 90°C CPU temp, FTR Hardware.fr are at 74°C when they got the same power in their overclocking test at 4.3GHz...

Otherwise they are far from 140W with Prime 95 at 4.05GHz@1275 mV wich is the stock setting...

https://www.hardware.fr/medias/photos_news/00/55/IMG0055234.png

Edit : Computerbase Cinebench score is 1790 and the power delta measured point to 90-100W power drain for this...

https://www.computerbase.de/2018-04/amd-ryzen-2000-test/5/
https://www.computerbase.de/2018-04/amd-ryzen-2000-test/3/#diagramm-cinebench-r15

 

[The Stilt]

Apparently the cooler thermal resistance is about 0.43°C/W, hence a 90°C CPU temp, FTR Hardware.fr are at 74°C when they got the same power in their overclocking test at 4.3GHz...

Otherwise they are far from 140W with Prime 95 at 4.05GHz@1275 mV wich is the stock setting...

https://www.hardware.fr/medias/photos_news/00/55/IMG0055234.png

I'm using Wraith Prism, in ~22°C ambient.

 

[The Stilt]

Thank you for your explanation Are the tools you are using to set power limits "propriety" or can you share them? It would be very useful to be able to lock power use without constant rebooting.

They are my own, but they use methods which are under NDA.

I have no real means to protect these methods, so at least for now I'm not willing to share the tools.

 

[mat9v]

They are my own, but they use methods which are under NDA.

I have no real means to protect these methods, so at least for now I'm not willing to share the tools.

Thank you back to Bios for me

 

[The Stilt]

The figures from Hardware.fr are very well in line with the ones I'm seeing:

126W reported package power during X264.

https://www.hardware.fr/articles/974-6/consommation-efficacite-energetique.html

 

[PhonakV30]

So Ignore AMD's TDP and define Ryzen's TDP based on Intel definition ? because they think Intel's definition is most logical they can have.

 

[mat9v]

The figures from Hardware.fr are very well in line with the ones I'm seeing:

126W reported package power during X264.

https://www.hardware.fr/articles/974-6/consommation-efficacite-energetique.html

According to above numbers, 1700X or 1800X also did not respect power limits - is that what you have noticed in your tests too?

 

[The Stilt]

According to above numbers, 1700X or 1800X also did not respect power limits - is that what you have noticed in your tests too?

I didn't test 1800X with X264 or Prime95, just in CB15 nT.
In CB15 nT the performance of 1800X didn't change when the PPT was limited to 95W, compared to stock.

 

[mat9v]

I was just curious about your opinion, if their results came from specific configuration of 2700X CPU and if so why did 1800X also cross power limit - if it was a standard case for those Ryzen CPUs.

I have found an interesting review - http://www.pcgameshardware.de/Ryzen...2600X-Review-Benchmark-Release-Preis-1254720/ and would love your opinion on the following screenshot - http://www.pcgameshardware.de/screenshots/original/2018/04/Ryzen-Master-Tool-R7-2700X-pcgh.png - look at the core "7" - it states that the clock is 4500Mhz, further "control mode" is set to manual. I do not have 2700X CPU and was wondering - should I understand that I may be able to manually set maximum clock per core in RyzenMaster (or in bios of an X470 board), so the "golden core" would work at 4500Mhz for maximum single core performance?

 

[The Stilt]

There is some indication that AMD is planning to provide a control, which would allow the user to set the maximum boost clock.
i.e. on 2700X alter the 4.35GHz boost ceiling.

Last time I tried it, it wasn't working so I'm not certain if I've understood it correctly.

Per-core clocking isn't possible on Pinnacle Ridge, at least as far as I know.
However if the ability to increase the maximum boost clock would materialize, then practically it should behave the same as the stock XFR 2 does (i.e. the highest boost is only achievable on the best core(s)).

 

[Abwx]

The figures from Hardware.fr are very well in line with the ones I'm seeing:

126W reported package power during X264.

https://www.hardware.fr/articles/974-6/consommation-efficacite-energetique.html

That s with an Asus Crossair that overvolt by 50mV, beside I pointed that the CPU consume more in X264 but not as much in the test you re using wich is Prime 95.

He has 129W7 at the ATX and he got 126W and 135W on a MSI and a Gigabyte MBs respectively, at a 256K FFT, what is the FFT size in your test..?..128K.?

BTW, his CPU run this FFT at 1275mV VID (1264mV measured) /4.05GHz , that s more than in your test and still, your sample consume more.?

https://www.hardware.fr/articles/974-7/overclocking-pratique.html

 

[IRobot23]

Weird. Yeah, you did all the job and now we are "criticizing" you... Sorry. Just keep up the good work. I always like to see your results, basically one of the best reviewers.

 

[IRobot23]

The figures from Hardware.fr are very well in line with the ones I'm seeing:

126W reported package power during X264.

https://www.hardware.fr/articles/974-6/consommation-efficacite-energetique.html

Well i7 8700K is certainly not 100W at 4.3GHz.
You said that R7 2700X is first one... this site shows every ryzen out of TDP.

Tomshardware did same test and they got 105W in hwinfo64. Maybe ryzen 2000 builders could help in that thread. Also 12nmLP is probably rushed and since we are just below freq limit there can be huge differences...

I saw some R7 1800X 4-4.1GHz driking only 130-140W (peak) doing prime95 and monitoring by HWinfo64... another did 180-200W.

 

[mat9v]

There is some indication that AMD is planning to provide a control, which would allow the user to set the maximum boost clock.
i.e. on 2700X alter the 4.35GHz boost ceiling.

Last time I tried it, it wasn't working so I'm not certain if I've understood it correctly.

Per-core clocking isn't possible on Pinnacle Ridge, at least as far as I know.
However if the ability to increase the maximum boost clock would materialize, then practically it should behave the same as the stock XFR 2 does (i.e. the highest boost is only achievable on the best core(s)).

Well, they might have been able to do it like this: i.redditmedia.com/--EzP2t4YMf9TCYeibbG29tu3Ajfeje5QjrvCk8g1uI.png?w=572&s=d4d21987be37de7cfc9679f1aff4d5ea - but I do not know if RyzenMaster would correctly show it as 4500Mhz.

 

[The Stilt]

Well i7 8700K is certainly not 100W at 4.3GHz.
You said that R7 2700X is first one... this site shows every ryzen out of TDP.

Tomshardware did same test and they got 105W in hwinfo64. Maybe ryzen 2000 builders could help in that thread. Also 12nmLP is probably rushed and since we are just below freq limit there can be huge differences...

I saw some R7 1800X 4-4.1GHz driking only 130-140W (peak) doing prime95 and monitoring by HWinfo64... another did 180-200W.

The 8700K is actually able to sustain the advertized all-core turbo frequency (4.3GHz) in most workloads, even when the power limits are configured as they should (95W/119W).
I tested this after 8700K was launched and basically everything else but highly 256-bit workloads maintained 4.3GHz at 95W.

But I agree, 100W from the EPS is ~10% too low for the 8700K.