AMD Ryzen 5000 Builders Thread

[B-Riz]

	Cores Threads	Base Freq	Turbo Freq	L3 Cache	TDP	MSRP
Ryzen 9 5950X	16c/32t	3400	4900	64 MB	105 W	$799
Ryzen 9 5900X	12c/24t	3700	4800	64 MB	105 W	$549
Ryzen 7 5800X	8c/16t	3800	4700	32 MB	105 W	$449
Ryzen 5 5600X	6c/12t	3700	4600	32 MB	65 W	$299

5600X comes with cooler, the rest do not.

above from https://www.anandtech.com/show/1614...en-3-on-nov-5th-19-ipc-claims-best-gaming-cpu

 

[Timur Born]

To underline, the drop from 20 watts to 14 watts is induced by C-states, not by the CPU having no load to process. I find this rather strange, since I would expect all C-state related power readings to fall under the Core or SoC moniker?!

 

[B-Riz]

To underline, the drop from 20 watts to 14 watts is induced by C-states, not by the CPU having no load to process. I find this rather strange, since I would expect all C-state related power readings to fall under the Core or SoC moniker?!

In general, to answer your question about "where is the extra power used" go back and read all the Zen1 - Zen3 reviews, plus the server part reviews. Ian does a very good job of showing power usage; what is probably happening is the "missing" power is used for Infinity Fabric and the like; I believe the Zen2 / Zen3 server chip reviews go into great detail about it.

Here is the Zen3 desktop review power usage page, just for easy clicking.

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

 

[Timur Born]

Thanks, I will read the linked article.

Albeit with infinity fabric being part of the IO chip I would have expected it to be part of SoC power. I noticed from another user's screenshot that his 3600 only measures 10 watts Package/PTT over Core + SoC despite the IO chip being the same.

 

[B-Riz]

Thanks, I will read the linked article.

Albeit with infinity fabric being part of the IO chip I would have expected it to be part of SoC power. I noticed from another user's screenshot that his 3600 only measures 10 watts Package/PTT over Core + SoC despite the IO chip being the same.

I believe the 3600 silicon are selected for efficiency over performance, where the X CPU's are selected for performance over efficiency.

When you get to the server reviews, the more cores, the more power the "un-core" power usage goes up, Threadripper vs regular AM4 are interesting reads too.

The bad thing about Zen3 server, is there was a regression in power usage compared to how Zen2 improved from Zen1.

AMD 3rd Gen EPYC Milan Review: A Peak vs Per Core Performance Balance

www.anandtech.com

The AMD Ryzen Threadripper 3960X and 3970X Review: 24 and 32 Cores on 7nm

www.anandtech.com

 

[Timur Born]

The change from Intel GB Master based system to a AMD MSI Creation on definitively included more power consumption and more heat generation. The VRM of the MSI board might be worse than the (very good) GB one, too, with VRM idle temperatures being about 20°C higher (low 50s). On the other hand my current impression is that the MSI VRM is quieter, maybe even considerably. This is worth a lot.

Yesterday I noticed that despite 75% idle times my 5900X never enters C6 during a gaming session of TW:WH2. Now, C1 on AMD seems to encompass stages that mimic C3 on Intel and often I turn off anything higher than C3 anyway. I will have to test that on the Intel system again.

 

[JoeRambo]

The change from Intel GB Master based system to a AMD MSI Creation on definitively included more power consumption and more heat generation. The VRM of the MSI board might be worse than the (very good) GB one, too, with VRM idle temperatures being about 20°C higher (low 50s). On the other hand my current impression is that the MSI VRM is quieter, maybe even considerably. This is worth a lot.

Yesterday I noticed that despite 75% idle times my 5900X never enters C6 during a gaming session of TW:WH2. Now, C1 on AMD seems to encompass stages that mimic C3 on Intel and often I turn off anything higher than C3 anyway. I will have to test that on the Intel system again.

Frankly 5950x is a bad chip if You are after desktop efficiency. It has ~2010 era hardware characteristics in low load situation due to design and AMD boosting algorithms trying to reach 5ghz on ridiculous voltages even without PBO, on motherboard defaults.
All that can be seen in Anandtech review of 5950x:

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

1 core load power - 50W,
4 core load power - 110W

At idle and low load this chip is outright horrible choice for efficiency. AMD does not shy away from burning dozens of watts for extra 5% or so of performance and benchmark wins.

 

[B-Riz]

Frankly 5950x is a bad chip if You are after desktop efficiency. It has ~2010 era hardware characteristics in low load situation due to design and AMD boosting algorithms trying to reach 5ghz on ridiculous voltages even without PBO, on motherboard defaults.
All that can be seen in Anandtech review of 5950x:

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

1 core load power - 50W,
4 core load power - 110W

At idle and low load this chip is outright horrible choice for efficiency. AMD does not shy away from burning dozens of watts for extra 5% or so of performance and benchmark wins.

And that is why Eco Mode exists, try it out, is great!

 

[B-Riz]

The change from Intel GB Master based system to a AMD MSI Creation on definitively included more power consumption and more heat generation. The VRM of the MSI board might be worse than the (very good) GB one, too, with VRM idle temperatures being about 20°C higher (low 50s). On the other hand my current impression is that the MSI VRM is quieter, maybe even considerably. This is worth a lot.

Yesterday I noticed that despite 75% idle times my 5900X never enters C6 during a gaming session of TW:WH2. Now, C1 on AMD seems to encompass stages that mimic C3 on Intel and often I turn off anything higher than C3 anyway. I will have to test that on the Intel system again.

What do your VRM readings look like in HWinfo?

I set my PBO to + 125 Mhz and PPT to 130W.

Below is just regular idle stuff, nothing running / processing in the background, like 3% cpu usage right now.

 

[JoeRambo]

And that is why Eco Mode exists, try it out, is great!

Thanks but no, I took the matter into my own hands and properly undervolted the chip



As You can see my chip is using ~half* of the power versus stock tested by Anandtech. All cause I don't think pumping 1.5V to achieve >4.4ghz is only good for AMD marketing department.

* different load, no HT, different mobo etc

 

[Kenmitch]

Thanks but no, I took the matter into my own hands and properly undervolted the chip
View attachment 44475


As You can see my chip is using ~half* of the power versus stock tested by Anandtech. All cause I don't think pumping 1.5V to achieve >4.4ghz is only good for AMD marketing department.

* different load, no HT, different mobo etc

What is the above trying to show us? Viewing the readings it's looking like your using 1.119v's while you have your CPU gimped to 1143.2MHz running Cinebench?

 

[B-Riz]

Thanks but no, I took the matter into my own hands and properly undervolted the chip
View attachment 44475


As You can see my chip is using ~half* of the power versus stock tested by Anandtech. All cause I don't think pumping 1.5V to achieve >4.4ghz is only good for AMD marketing department.

* different load, no HT, different mobo etc

So what do you expect out the the 5950X?

An efficient CPU or a performant CPU?

Why no HT?

AMD already has said, touching Zen3 beyond a few fiddly bits with PBO / PPT etc. can not beat their baked in auto on the fly adjustments...

 

[JoeRambo]

What is the above trying to show us? Viewing the readings it's looking like your using 1.119v's while you have your CPU gimped to 1143.2MHz running Cinebench?

Gimped to 4.4ghz static OC @ 1.13V. AMD marketing department shoves 1.5V to achieve 4.75Ghz freq @ 110W with 4C load. Average effective clock = 4C 100% load, 12C 0% load results in ~1.1Ghz average.

So what do you expect out the the 5950X?

An efficient CPU or a performant CPU?

Why no HT?

I expect an efficient CPU, one that does not use 110W under light load of 4 cores, cause that is what my i7 920 was using in 2009.
I don't care about HT, cause my workloads don't really play well with HT, not everything is Cinebench.

AMD already has said, touching Zen3 beyond a few fiddly bits with PBO / PPT etc. can not beat their baked in auto on the fly adjustments...

Sadly my practice with 3950X and now 5950X has found that both stock and those PBO/PPT things are plain retarded both in low load regime, and when under or near full load.

Low load regime is already obvious - using 50W extra to gain less than 10% extra clock.
Under full load the "almighty" stock either drops clocks to ~3.8Ghz, or with PBO and expanded power limits goes ballistic with power to achieve the all core clocks of my undervolted setups.

 

[B-Riz]

Gimped to 4.4ghz static OC @ 1.13V. AMD marketing department shoves 1.5V to achieve 4.75Ghz freq @ 110W with 4C load. Average effective clock = 4C 100% load, 12C 0% load results in ~1.1Ghz average.




I expect an efficient CPU, one that does not use 110W under light load of 4 cores, cause that is what my i7 920 was using in 2009.
I don't care about HT, cause my workloads don't really play well with HT, not everything is Cinebench.



Sadly my practice with 3950X and now 5950X has found that both stock and those PBO/PPT things are plain retarded both in low load regime, and when under or near full load.

Low load regime is already obvious - using 50W extra to gain less than 10% extra clock.
Under full load the "almighty" stock either drops clocks to ~3.8Ghz, or with PBO and expanded power limits goes ballistic with power to achieve the all core clocks of my undervolted setups.

So why did you get this if a 4C load and no HT is the most important thing?

How does the chip down clock if you have a static OC? Isn't that a fixed frequency it is constantly running at?

Do you have posts about testing with a negative offset, or did you just go for a fixed voltage when you started the under-clock tuning?

I did negative offset with Ryzen 7 1700 and Ryzen 7 2700X, but have left Zen2 and Zen3 largely to their own devices as all the under-clock testing posts resulted in reduced performance versus leaving most things on auto.

 

[JoeRambo]

It was in response to another user complaining about AMD system with 5950x using a ton more watts versus his Intel one on idle and "gaming" load.
So i have looked up data from Anandtech testing, that had 4C load point with wattage north of 100W ( no need to rely on Anandtech either, our own Det0x has screenshots with CB power / threads scaling in this thread) and provided my own data for my static overclock of 4.4Ghz.
And my power usage is actually overstated already, due to memory OC and non stock voltages to support it. The cores are using ~5.5W @4.4ghz when fed with 1.13V, the rest of the chip chews 30W, would go down with casual 3200 mem.

I did try negative offsets, PBO, CTR 2.1 and so on. The result is always the same, horribly inefficient compared to under volt to the region where the chip is still efficient.
Offsets or not, AMD is showing 1.4-1.5V to rise frequency, per core wattage goes to 12-18W to extract 10-15% more clock. Then either PPT limit is hit and downclocking/downvolting gradually starts, or if limits allow, chip goes ballistic to 300W.

 

[DrMrLordX]

Not gonna lie, if I had a 5950X burning 110W on a ~4c workload I would not really be complaining much, especially since I know the chip can't pull more than 32W on top of that as the workload intensifies (assuming I'm running stock).

 

[Timur Born]

5900X, not 5950X. I made another interesting observation: Running few threads of high load increases CPU temperature when C-states (sleep) are enabled vs. disabled, despite wattage being lower.

 

[JoeRambo]

Not gonna lie, if I had a 5950X burning 110W on a ~4c workload I would not really be complaining much, especially since I know the chip can't pull more than 32W on top of that as the workload intensifies (assuming I'm running stock).

I would not mind either, if they had anything to show for it. According to Anandtech testing, at 4c load they are achieving 4.725Ghz at the cost of 20W per core and 110W total.
While with undervolt and fixed OC i am achieving 4.4Ghz at the cost of 5.5W per core and 55W total. All cause of AMD marketing department deciding it is OK to shove 1.5V into the chip and hit as high freqs as possible.
Btw 5.5*16 => 88W, so ~ 120W of peak power usage when all cores are active.

AMD has incredible chips in 3950x and 5950x. Just that if left to default behavior they have atrocious power usage when left to the mercy of stock algorithms. The trade off seems to be 15W per core for maybe 5% more performance @stock.
Combine such nasty power usage with overhead of IOD and you end up with CPU, that competes with Intel's "finest" in burning power and only become efficienty in throughput/power mid-late in load regime.

Of course i understand AMD's position, more Mhz is good for marketing, and they need high voltages to have enough safety margin to account for load transients, frequency ramp from power saving, users running 16C on that motherboard with 2 phase VRM and so on. + my stability is without HT and probably would not hold Prime testing. Would it burn 15W more per core for 4.4? Nope.

I am running 3950X @4.05Ghz since last February, now 5950x joined compute resources @4.4ghz, very happy with these CPUs. Not happy with their brain dead hunt for mhz way out of efficient region.

 

[DrMrLordX]

While with undervolt and fixed OC i am achieving 4.4Ghz at the cost of 5.5W per core and 55W total. All cause of AMD marketing department deciding it is OK to shove 1.5V into the chip and hit as high freqs as possible.

You do realize that was a response to everyone complaining that Matisse could not sustain max boost on more than one core, and not in any realistic scenario? It is exactly what AMD's critics wanted from them. Also, I see no indicator that the voltages required to achieve boost in excess of 4.7 GHz on four cores are at all dangerous at those current levels.

 

[B-Riz]

So the current topic of 4C load usage made me test my 5900X at current settings of 130W PPT and +125Mhz PBO, I ran 4T mixed testing on prime95 and pinned them to the first 4 physical cores in task manager; my test run errored out, so I need to change settings.

But the chip used up to the 130W PPT limit on those 4 cores on load, so, if you want a low power chip, set a lower target C temp or a lower PPT and leave PBO off.

I am going to test a target C of 70 now and leave everything else stock, to see how it does in prime95 mixed 4C load.

 

[JoeRambo]

But the chip used up to the 130W PPT limit on those 4 cores on load, so, if you want a low power chip, set a lower target C temp or a lower PPT and leave PBO off.

With PBO off it will still use 20W per core, not much improvement. The only way to have efficient 12-16C CPU is to manual OC in efficient voltage range and giving up last 5% of ST performance.

 

[Kenmitch]

With PBO off it will still use 20W per core, not much improvement. The only way to have efficient 12-16C CPU is to manual OC in efficient voltage range and giving up last 5% of ST performance.

What is going to be the main use of your 5950x? There's nothing wrong with tweaking it to your liking as long as your using loads that will be similar to your use case scenarios.

I have a 5900x and am currently experimenting with clocks/voltages as the summer months are coming up quickly. I recently found out that my current uEFI doesn't generate WHEA errors when I run my memory at 3800MHZ with 1900FCLK so I've been side tracked stress testing for WHEA errors.

 

[JoeRambo]

What is going to be the main use of your 5950x? There's nothing wrong with tweaking it to your liking as long as your using loads that will be similar to your use case scenarios.

Same as 3950x and 9980XE, they sit in the corner of my working room, a compute pool of DYI sort, running specialized software. The load varies with time, but after adding such monster as 5950X they spend even more time in low load regime.
I have put my "review" of 5950x @4.6ghz here:

Page 45 - AMD Ryzen 5000 Builders Thread

Page 45 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

but i have since settled on 4.4ghz since it required almost 0.1V less.

 

[B-Riz]

So the current topic of 4C load usage made me test my 5900X at current settings of 130W PPT and +125Mhz PBO, I ran 4T mixed testing on prime95 and pinned them to the first 4 physical cores in task manager; my test run errored out, so I need to change settings.

But the chip used up to the 130W PPT limit on those 4 cores on load, so, if you want a low power chip, set a lower target C temp or a lower PPT and leave PBO off.

I am going to test a target C of 70 now and leave everything else stock, to see how it does in prime95 mixed 4C load.

4C, mixed prime95, 70 C max temp test started.

 

[Kenmitch]

but i have since settled on 4.4ghz since it required almost 0.1V less.

My 5900x more or less a toy in the end. I'm on 1440p so I don't think the end clocks for me will make much difference in gaming, but I still have to check the fps when I'm underclocking my cpu.

I believe my uEFI settings are as follows. PBO on advanced, all sub-menus are auto, curve optimizer disabled, load line setting 4 the memory is running at 3800/1900. I found no reason to disable SMT on my rig.

Using CBr20 as a quick test.

Default:



Using the latest free version of CTR 2.1 it came up with this profile. It's the 1st time that it's actually given me a tried and true stable profile.



I guess it does prove your point in the end as far as clocks vs voltages go. I'm more or less concerned with temps in the end as Summer is just around the corner. 20c drop was impressive without loosing any performance in the above scenario.

 

[B-Riz]

4C, mixed prime95, 70 C max temp test started.

prime95 crashed with that temp limit, put everything back to default and am just testing Cores 0, 2, 4 and 6 smallFFT to see if I have something else unstable.