itsmydamnation
Diamond Member
- Feb 6, 2011
- 3,020
- 3,779
- 136
Did you ever run any of those custom bios builds out there?
T
- The combination of Windows 10's new Ryzen-aware scheduler and AMD's chipset drivers allow the operating system to schedule single-threaded tasks into the fastest cores (thread pinning).
I wonder if the OEM-provided chipset drivers provide that functionality? On the stock UEFI for my x570 Aorus Master + OEM-provided chipset drivers, it sure looks like CBR20 (at least) isn't pinning itself in ST mode to any particular core. It jumps around like crazy.
Are you running the latest Windows 10 build? Last time I heard Windows 1809 doesn't include the new Ryzen 3000 optimized task scheduler.
Maybe all your cores are created equal and boost properly that's why none of them is the chosen one? ;-)
f the OEM-provided chipset drivers provide that functionality? On the stock UEFI for my x570 Aorus Master + OEM-provided chipset drivers, it sure looks like CBR20 (at least) isn't pinning itself in ST mode to any particular core. It jumps around like crazy.
Which power plan are you running?
Are these issues really a concern for a "set and forget" type of person in any case? These seem more like minutia that enthusiasts worry about. What issues are affecting peoples day to day usage in a perceptible way, not people poring over HWiNFO numbers?
I don't think it can be stated enough that reporting bias is a factor here especially when using reddit as some sort of basis for you being "shocked by the number of issues". "Nothing to report, everything working fine" posts don't get the upvotes that more controversial posts might.
Now I feel like an apologist so I'll say AMD should fix everything they can!
This PDF from AMD is worth a read.
I've read the entire PDF - it looks and sounds like an excuse for a broken piece of silicon which warrants a new fixed revision. "Light" loads must not peg the CPU at maximum voltages and frequencies no matter what. You see that they even went ahead and released a "fixed" version of Ryzen Master which will no longer report the true data about the CPU - this is simply unacceptable. In my over 25+ years of running various CPUs I've never seen a situation when light monitoring tools like CPU-Z or HWiNFO would cause any discernible CPU clocks/voltages ramp up.
If I were AMD I'd admit I've f***ed up and released a new revision of Ryzen 3000 CPUs which Intel does all the time. Still, I've been under the impression that both Intel and AMD nowadays have a big chunk of the CPU logic completely programmable via firmware, so I wonder why it's not the case right now at least for AMD.
What's light load? You even put it in quotes yourself."Light" loads must not peg the CPU at maximum voltages and frequencies no matter what.
What's light load? You even put it in quotes yourself.
Whatever the load, it's a valid policy to execute the load as quickly as possible. The PDF even mentions and linked why that's commonly accepted as a valid strategy, race-to-idle. AMD increased the action granularity to 1ms and allowed this behavior to be used unrestricted. Now it turns out "idling" applications are essentially abusing this behavior and having loads for the CPU to process again and again, making a fool of the whole idle concept. AMD's reaction is to draw a line and reintroduce the previous 15ms window for short load bursts that stay within so the CPU actually gets a chance to idle.
I'm familiar with the efforts through powertop (a tool by Intel) under Linux which tracks devices and processes waking the CPU from sleep, thus decreasing power efficiency. The result of having this tool over the years was that such wake calls were vastly reduced, and if still necessary, combined in batches. To me the idle behavior under Windows looks like it could use such an optimization as well.
It's a case of PEBCAK imo. The sampling is a highly specific snapshot value that can change, at a granularity of 1ms, 1000 times per second (duh!). AMD now changes this to a running average across time and the whole package to fix the issue of perception.I dont even think its broken, the problem is a error in sampling , both for temperature and for voltage.
It's a case of PEBCAK imo. The sampling is a highly specific snapshot value that can change, at a granularity of 1ms, 1000 times per second (duh!). AMD now changes this to a running average across time and the whole package to fix the issue of perception.
Perception in general is an issue with Ryzen 3k, somebody did a huge rant on Reddit about that. Unlike him I do think the changes AMD did so far as a result of that are mostly alright though.
It doesn't .. I mean there are scope traces in there and everything. It absolutely does not and never did peg high voltages and frequencies at idle. Did you even read the pdf?"Light" loads must not peg the CPU at maximum voltages and frequencies
It doesn't .. I mean there are scope traces in there and everything. It absolutely does not and never did peg high voltages and frequencies at idle. Did you even read the pdf?
I do agree that the high idle temp thing was a bit of an issue that needed to be fixed -- especially when putting this arch into laptops, but it seems like their solution is good. They were being a bit too aggressive in coming from idle states.
Once I finally get some RAM for my build I would like to do some scope traces myself too for additional data. Just need to find some good probe points..
Why would you want to build a Ryzen 3000 system? It seems you would be a nervous wreck, with all the issues you're investigating.I also hope to build a Ryzen 3000 based system but currently cannot afford it. The reason I didn't like that PDF is again because AMD decided they would no longer show true data from their CPUs. If they have lots of sensors I want to see multiple data points. If there are spikes I want to see them as a graph. I don't want them to dumb down their software to hide the data.
Why would you want to build a Ryzen 3000 system? It seems you would be a nervous wreck, with all the issues you're investigating.
I prefer my silicon to be cool, low voltage and quiet. My last CPU which is Intel Core i5 2500 is still in a perfect condition eight years after I purchased it. My Nexus 5 phone still works perfectly six years after I bought it. I would like my Ryzen 3000 system to serve me for at least the next five years. I do understand that most middle-class Europeans and Americans (who I guess prevail at these forums) can afford a new PC every month but it's not my case. And I'm not content with my future CPU running at 1.45V at stock which doesn't sound like a safe figure.
And thanks for calling me a nervous wreck. Much appreciated. Being concerned about HW equipment longevity is a trait of a madman. Right, I get it.
Simple enough. Set a modest pstate OC, leave cool & quiet enabled, fix your fan curve, and you're golden.I prefer my silicon to be cool, low voltage and quiet.
While I can fully understand your OCD and tend to feel the same with wanting complete access to interesting data, this change is actually an improvement wrt "true data" if AMD's explanation that the new rolling averages are what their PB2 algorithm is actually acting upon is correct.The reason I didn't like that PDF is again because AMD decided they would no longer show true data from their CPUs. If they have lots of sensors I want to see multiple data points. If there are spikes I want to see them as a graph. I don't want them to dumb down their software to hide the data.
I also hope to build a Ryzen 3000 based system but currently cannot afford it. The reason I didn't like that PDF is again because AMD decided they would no longer show true data from their CPUs. If they have lots of sensors I want to see multiple data points. If there are spikes I want to see them as a graph. I don't want them to dumb down their software to hide the data.