SeaSonic X750 Gold 750W
MSI X370 XPOWER Gaming Titanium | Ryzen 5 3600X 1924SUT
GSkill TridentZ F4-3600C15D-16GTZ @ 3800/1900 mhz | 1.44v | 15-17-10-15-42-28-266-16-1T
Asus GTX 780 DCII OC @ 1332/3522 mhz | 1.3v | Samsung 960 EVO 500GB
EK-FB MSI X370 XPower RGB Monoblock 1/2" ID | Swiftech D5 MCP655 | EK-XRES 100
Alphacool NexXxoS UT60 280 | 4 x Scythe GT AP-15 @ 5v
Thermochill PA120.3 | 1 x SST-FN123 | 2 x Scythe GT AP-15 | @ 5v
Using Ryzen Master, 'C 04' has a golden star, but while running single core loads, this core is hardly ever used and is almost constantly in sleep mode.
I am using the latest chipset drivers, the latest ryzen master and the latest (sill beta bios from 01/07) BIOS for my motherboard.
I would have at least thought, as per AMD statements, that the OS would at least use the 'fastest' core in the system.
So would be interested to hear from other peeps, does your OS use the 'fastest' cores in your CCXs ???
This is after the OS being up and running for around 30 minutes.
During this period of time I have casually used my PC for simple tasks such as web browsing, playing music, opening some software and I ran a few benchmarks such as CPU-Z CPU benchmark and AIDA64 memory cache benchmark.
As you can see the 'fastest' core peak clock speed is significantly lower than the other cores.
Also if I attempt to pin any threads to core 4 using process lasso, the OS loads the fastest core for a split second then does not respect the CPU affinity that was attached to it.
The latest AMD chipset drivers at the time were v1.8.19.0915
I have seen that v1.09.27.1033 were released a couple of days ago, I will load those up to see if there is any change ......
mongoled, I do not have an answer for you, but do not think you have a problem. I do not know how AMD determines the fastest core. If you look at your Peak speed display, C04 is by far the slowest not the fastest. In any case you have enabled Auto Overclock at set it to 75MHz. Both of your CCXs are set to 4475MHz. Do your cores run that fast when loaded? Please open an AMD Online support request and see what they say. Thanks and enjoy, John.
just a couple of things, of course I know which core is stated as the 'fastest' the title of the thread is "
So did not understand why you had to mention that as its pretty obvious that is something I know ........
Yes, at least two cores show peak speeds of 4475 mhz using these settings, C04 also used to show these peak speeds.
Also, this issue is 'resolved' as after updating the drivers (this issue did not initially exist with previous drivers).
Im going to make a new post in this thread about another anomoly.
It seems that there is alot of work that needs to be done by numerous organisations to get these CPUs to work optimally.
What I am seeing are inconsistencies that rear their head for no apparent reason and this is not just on my setup, the Internet is full of these issues ...........
First of all, Windows (10) CPU drivers schedule the threads--and the OS will always have priority over something like Lasso. I'm running an advanced build of Win10, build 18362.10022, and in this build according to Microsoft, the Win10 CPU drivers are supposed to schedule work to the most capable cores first--as in the fastest cores. Read that as *Windows* does the core scheduling--not the end user...;) Last, be advised this is a new feature Microsoft is working on atm--it's not in your present build. It probably will never always work perfectly--and surely not in the manner some might expect--but I thought you should know they are working on something like you suggest. About sleeping cores you see in Ryzen Master--are you talking about idling? Of course, your cores are going to sleep when you are idling--because they aren't tasked with doing very much...;) Looks like it from your screen shot. If so, that's normal. Also, if you are running a single-threaded app then some cores may sleep. But if you are running a demanding multi-threaded app then all of the cores should be running simultaneously. Try running a "test" of your settings in Ryzen Master and watch how all the cores run at the same time.
Also, remember that for all MHz speeds on every core above the base clock, you are looking at *boost* and all boost MHz are "up to" your max boost. Range of possible boosts runs from no boost at all up to your advertised max boost. My 3600X, for instance, shows 4500MHz where yours shows 4475, but that's because I have set a PBO of 100MHz that is not guaranteed to work--in fact, it never does, really. But occasionally, I do get a max boost of 4.424GHz which is slightly above the normal max for my CPU, which is 4.4GHz (3dMk's Timespy shows me the highest max boost the cpu attains every bench run in its report). So, the short of it is, something that resembles what you are asking for is on the way--but, no, you will never be able manually assign certain threads to certain cores at certain times, imo. I don't think much of anything would work very well if at all if we were able to do things like that for complex multi-core CPU routines operating on a nanosecond timing scale...;)
waltc, Google knows a lot about this. I would like to know exactly which 1903 patch level includes which 'fix'. 1903, when released, contained some or most of the needed changes for the Ryzen Gen3 scheduler support. mongoled's 4475 is the max boost (4400MHz) plus the 75 MHz specified in RM for Auto Overclock. There appears to be little load, so nothing is running fast at the time of the SS.
Please provide a link to the MS discussion of "according to Microsoft" comments above. Thanks and enjoy, John.
The attempt to use process lasso as I am sure you understand was to attempt to 'steer' the OS towards the fastest core.
As you have correctly pointed out, the OS is king and the king has determined that it will not be overridden.
Ive just experienced something that I dont have a explanation for but it fits into what you have described above.
I have two OS, one as my workstation OS and one as my testing OS.
The testing OS and the workstation OS are both the same patch version and the same chipset version and they are both using the same version of Ryzen Master, however, the 'fastest' cores are different !!!
C01 - second fastest core of CCX 0
C02 - fastest core of CCX 0
C04 - fastest core in the system
C06 - second fastest core of CCX 1
C01 - fastest core of CCX 0
C02 - second fastest core of CCX 0
C04 - second fastest core of CCX 1
C06 - fastest core in the system
How is this possible ?
The answer to this could only be provided by an investigation that involved AMD, MSI & Microsoft.
At least if it was made clear to which factors determine the 'fastest' cores we would maybe be in a position to look for a 'solution' to this problem.
No I am not talking about idling
Im also fully aware of how AMD has advertised and marketed these CPUs with all the ins and outs that have been associated with these decisions!
Yes, CPUs are become more and more advanced and complex, governed by algorithms that take time to optimize and yes they may most likely never be fully optimized as the difficulties are many and on across so many different domains.
Just some info,
when the 'fastest' core is used within the confines of the workstation OS, using AIDA64 memory latency test, the characteristics of the boosting algorithm are as follows, the core will have a short burst to say 4414 (+- 10mhz) mhz (according to Ryzen Master) and then will jump to another core momentarily before returning to the 'fastest' core.
This shows in the result of the benchmark, the latency is 64.2 ns.
Now when the same test is run on the test OS, this is what happens, core C01, boosts to 4414 (+- 10mhz), occasionally dropping to just below 4000 mhz, but more importantly it stays on the same core.
This also shows in the result of the benchmark, the latency can be lower than 63 ns.
Im hopeful that future improvements will bring more consistency to the results we are seeing.
So I uninstalled Ryzen Master from the workstation OS using revo-uninstaller portable, rebooted and reinstalled.
Now Ryzen Master is showing same information in both OS's with regards to 'fastest' core...........