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

 

[Magic Carpet]

I did, dropped power by about 10 watts under my Wow test-load, which still corresponds to 20-25% more than the 9900K doing the same thing.

Well Timur, if you're that concerned about power consumption, you should have probably stayed with 9900K.

 

[kognak]

When tuning 5950x i've found it runs 4.4ghz static 1.1375V core, DDR4 3600C15, ~1.1V SOC, 100% high perf plan with C states enabled:
at idle it uses 30W of package power, while showing CPU spends >98% in C6 power state. WTH really.

It's downside of chiplet design. External SoC and high speed/low latency links on substrate aren't most power efficient way to build a cpu. Threadrippers have more chiplets and bigger SoC, they use even more power when idling. Monolithic design is much superior in this regard, that's why all mobile chips are monolithic. Zen cores themselves however are very power efficient, in idle they can switch to sleep mode and consume practically zero watts. Tweaking cores have no real effect on idle consumption either nor have clock speeds. But tuning memory and fabric clock speeds does. If anyone wants very low idle consumption and AMD, it needs to be APU. They do under 5W package power.

 

[Timur Born]

Not sure about being concerned. I share experiences and compare to what I have available.

 

[JoeRambo]

Definitely, and things will improve big time when AMD moves to ZEN4 and new design of IOD on new process.

What i find disappointing is the lack of information about this in actual reviews. I do understand that reviewers don't want to compare Apple to Oranges and so on, but they do bad job here highlighting this very substantial difference in idle power use.

And on topic of bad job by reviewers - the fact than not one of them noticed USB problems is disgrace and puts a stain on their reputation going forward. I mean during my limited time 5950x with pre 1.2.0.2 AGESA USB was acting up already ( mouse + keyb + HyperX usb audio crackling), there is no chance in hell that all of them got lucky and ran into no problems.

 

[B-Riz]

Not sure about being concerned. I share experiences and compare to what I have available.

I understand your point and frustration, but, you are comparing a non-monolithic 12c/24t part to a monolithic 8c/16t part made on a super refined process that Intel targeted at having very low idle power usage.

For many of us who have been reading reviews and playing with Zen since the Zen1 launch, your points about idle power usage and low thread loading power usage have been known, and are highlighted in the Anandtech reviews;

"Moving down to a single chiplet but will the full power budget, and there is some power savings by not having the communications of a second chiplet. However, at 8-core load, the 5800X is showing 4450 MHz: the Ryzen 9 processors are showing 4475 MHz and 4500 MHz, indicating that there is still some product differentiation going on with this sort of performance. With this chip we still saw 140 W peak power consumption, however it wasn’t on this benchmark (our peak numbers can come from a number of benchmarks we monitor, not just our power-loading benchmark set). "

Also, see below, that is why I asked if you read reviews before buying, the loading and power usage is shown:



If the 5900X performance in WoW is such a horrible thing, sell it and get a 5600X.

 

[B-Riz]

Definitely, and things will improve big time when AMD moves to ZEN4 and new design of IOD on new process.

What i find disappointing is the lack of information about this in actual reviews. I do understand that reviewers don't want to compare Apple to Oranges and so on, but they do bad job here highlighting this very substantial difference in idle power use.

And on topic of bad job by reviewers - the fact than not one of them noticed USB problems is disgrace and puts a stain on their reputation going forward. I mean during my limited time 5950x with pre 1.2.0.2 AGESA USB was acting up already ( mouse + keyb + HyperX usb audio crackling), there is no chance in hell that all of them got lucky and ran into no problems.

The USB issues are very unique, and, I do not think a reviewer would ever think to load up the USB bus and have at it.

Like any complicated technical issue, AMD gathered information and are working towards a fix; it's not like it can be fixed in a week or two. The root cause has to be identified, a fix developed, tested, and propagated to how many board vendors?

And, to be on a cutting edge halo product and complain there are issues, is kinda funny on an enthusiast msg board.

 

[Makaveli]

And on topic of bad job by reviewers - the fact than not one of them noticed USB problems is disgrace and puts a stain on their reputation going forward. I mean during my limited time 5950x with pre 1.2.0.2 AGESA USB was acting up already ( mouse + keyb + HyperX usb audio crackling), there is no chance in hell that all of them got lucky and ran into no problems.

I'm on 1.2.0.1 Patch A and have no usb issues.

 

[JoeRambo]

1.2.0.1 Patch A

Patch A is 1.2.0.1 + backport of USB fixes from 1.2.0.2. 1.2.0.2 fixed USB problems for me as well.

 

[Muadib]

Patch A is 1.2.0.1 + backport of USB fixes from 1.2.0.2. 1.2.0.2 fixed USB problems for me as well.

I wasn't having any issues with USB, but I upgraded the BIOS anyway.

 

[Timur Born]

I understand your point and frustration, but, you are comparing a non-monolithic 12c/24t part to a monolithic 8c/16t part made on a super refined process that Intel targeted at having very low idle power usage.

No frustration here and no need for you to defend the CPU and keep pointing to reviews, I would have bought the CPU anyway for testing drivers (especially USB and possible PCIe issues). I am reporting experiences and comparing to my Intel system. And I was an early adopter of Ryzen 1 and did some crazy tests with that (like proving that AIO hoses can explode off defective pumps due to default BIOS settings disallowing thermal shutdowns).

The main disappointment so far is the lack of Lua script performance despite the claimed IPC improvements (which Lua does not seem to benefit from at all). And that a heavily threaded application like Topaz Gigapixel AI still is not parallelized enough to make good use of the 5900X. Throwing cores and cache at problems still is not an easy solution for everything.

More power draw for the same load also means more heat production, which in turn means more noise. I will check that more properly once I applied TIM to the CPU/cooler.

 

[Timur Born]

On a side-note. I specifically switched all PCIe 4.0 down to 3.0 trying to lower the power consumption. Unfortunately it did not make a difference. Not exactly a surprise since none of my device makes use of PCIe 4.0 anyway, but worth a try knowing that the IO and chipset chips need the active cooling.

That being said, is there a temperature sensor for the IO chip of the CPU? My Arctic Liquid Freezer 2 offers an optional offset placement to better cool the CPU (CCDs), but I suspect that this leaves the IO chip cooled worse.

 

[Det0x]

The main disappointment so far is the lack of Lua script performance despite the claimed IPC improvements (which Lua does not seem to benefit from at all)

Some programs have reached the limit and almost dont scale with "IPC" anymore, only frequency.. CPUmark99 is one example of this:

Page 18 - Post your Cpumark99 scores

Page 18 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[DrMrLordX]

The main disappointment so far is the lack of Lua script performance despite the claimed IPC improvements (which Lua does not seem to benefit from at all).

Has Lua performance ever really been a big problem?

 

[B-Riz]

No frustration here and no need for you to defend the CPU and keep pointing to reviews, I would have bought the CPU anyway for testing drivers (especially USB and possible PCIe issues). I am reporting experiences and comparing to my Intel system. And I was an early adopter of Ryzen 1 and did some crazy tests with that (like proving that AIO hoses can explode off defective pumps due to default BIOS settings disallowing thermal shutdowns).

The main disappointment so far is the lack of Lua script performance despite the claimed IPC improvements (which Lua does not seem to benefit from at all). And that a heavily threaded application like Topaz Gigapixel AI still is not parallelized enough to make good use of the 5900X. Throwing cores and cache at problems still is not an easy solution for everything.

More power draw for the same load also means more heat production, which in turn means more noise. I will check that more properly once I applied TIM to the CPU/cooler.

Not defending, trying to point out your performance observations vs the 9900K and the 5900X's higher idle power usage are known items, while providing references where I recall seeing relevant data.

 

[Timur Born]

Has Lua performance ever really been a big problem?

Load times for WoW and Fantasy Grounds are well over half a minute due to Lua scripts being loaded and initialized. CPU bottlenecks in Lua are also regular causes for dramatic frame-rate drops (GPU load decreases when these happen). Lua is single-threaded and many engines still use old Lua versions (like v5.1), so they don't benefit from the performance improvements of v5.4.

 

[Timur Born]

Is it normal for CCX quality to differ so much on a single CPU (5900X)? The best core of CCX 2 is worse than the worst core of CCX 1. Is this lottery or done on purpose by AMD?

 

[B-Riz]

Is it normal for CCX quality to differ so much on a single CPU (5900X)? The best core of CCX 2 is worse than the worst core of CCX 1. Is this lottery or done on purpose by AMD?

View attachment 44081

5900X is one "good" 5600X and one "bad" 5600X stuck together, guess what a 5950X is?

 

[Kenmitch]

Is it normal for CCX quality to differ so much on a single CPU (5900X)? The best core of CCX 2 is worse than the worst core of CCX 1. Is this lottery or done on purpose by AMD?

View attachment 44081

It's the norm for one strong and one weaker CCX.

Maybe the app just makes them up? My 5900x has the same exact readings as yours, but on different cores as yours.



I wouldn't give it much thought in the end.

 

[Timur Born]

I would rather suspect that the numbers are fixed by AMD, but the quality differences may well still be real. You can switch the availability of the core classification via UEFI and Windows Thread Scheduler bases its decision on that exact classification (mostly preferring good cores for high load and bad cores for low background load).

 

[Timur Born]

5900X is one "good" 5600X and one "bad" 5600X stuck together, guess what a 5950X is?

So it is a deliberate decision by AMD to get rid of worse dies instead of pure lottery?

 

[B-Riz]

So it is a deliberate decision by AMD to get rid of worse dies instead of pure lottery?

They only need to have one core boost to max 4.8 and the rest meet minimum at 3.7 on load, so, they only need one chip with one core that meets the max boost specification.

https://www.amd.com/en/products/cpu/amd-ryzen-9-5900x

https://www.amd.com/en/support/kb/faq/cpu-pb2

 

[Det0x]

For comparison:

5600x


5800x


5900x


5950x

 

[Timur Born]

Thanks for the comparison, that puts the numbers into perspective and demonstrates that different numbers can be had within the 5900X series. Maybe the numbers are set by BIOS and thus mainboard manufacturers have a hand in this rather than AMD setting them for each CPU in the factory. Or maybe AMD sets them for each run of dies coming from the fab or just position on the wafer?! 1usmus may know more as coder of CTR?!

I like how CPPC allows Windows to properly schedule the threads to the most fitting cores. In one test-run I did see Prime95 1 thread load being shifted from the two best cores to the very worst core for some time, though. That seemed strange, on the other hand P95 uses lowest idle thread priorities, so it may confuse the scheduler.

 

[Yosar]

So it is a deliberate decision by AMD to get rid of worse dies instead of pure lottery?

I would not call them worse. They bin them as worse but that's not the same. The worse die on my 5900X would be probably good die on 5600X.
Every core on my worse die boost well over 4.9 GHz (on my better die every core boost over 5 GHz). Some of those cores on worse die would probably boost higher with no problem but AMD limited them down I guess. Almost every core on my worse die can do -30 in Curve Optimizer. And probably much better.
Only Curve Optimizer is limited to max -30. That could allow them boost higher but they are not able with curve set on them by binning. Just by setting -30 on most cores I raised their boost about 200 MHz.

On the other hand my best core on better die can not do anything in Curve Optimizer. It has already maxed curve of voltages. Thanks to that actually my best core is not my best core. My best core now is core 0 where setting over -20 on Curve Optimizer gives me better boost than on my 'best' core 3.
I am not complaining definitely. But it looks like they mark cores quite arbitrary. They must meet probably some specification to be marked as best core, or the second best core. Not actually being really best core on the die.

When I speak about setting voltage in Curve Optimizer, I don't speak about going to BIOS and put some numbers, boot into Windows and call it a day if it doesn't crash for a few hours. I mean I really tested them for few hours with Core Cycler and other programs on one core.
That's why I know my best core 3 can not do anything in Curve Optimizer. But core 0 with setting -20 on Curve Optimizer is just plainly better (and it's not even my second best core).
Of course there is always chance that even my heavy testing didn't show that -20 on core 0 is just too aggressive. But so far so good, testing was done as much as I could.

Personally I suspect they have mostly good or very good dies. Chiplets are small, it's their second generation on 7 nm process. In the meantime TSMC probably also learned quite a bit about clocking high chips on their 7 nm (look at the clocks of RDNA2).
Basically I was very surprised when I put my CPU into mainboard. After all the hoopla with boosting on Zen 2 processors, my 5900X (and probably most others either) was boosting to 4.95 GHZ right out of gate with default BIOS. So well over specification.
Basically AMD reached 5 GHz barrier just silently not making too much fuss about it.

 

[Timur Born]

Makes sense, yes. Does anyone know for sure why Windows schedules lower load threads to the "worst" cores first, only to elevate them to "best" cores once they max out a core? My speculation is that it tries to keep the "better" cores cool/unused to allow for higher frequency spikes once they are needed.