-
Re: SMT off and temperature
Anyone ? Would really appreciate any input, asked it in several places and for some reasons no one has an answer. It really seems like something is off with temp sensing when disabling smt.
-
Re: SMT off and temperature
bump!
can someone maybe kind enough to tell me if he has the same issue? just disable SMT in the bios and run Prime95 or realbench etc...
-
Re: SMT off and temperature
Well first of all, you're not going to get an instant answer... try to be more patient next time.
With this said., anyone who assumes that SMT Disabled would result in Lower Temperatures is almost certainly assuming that SMT =/= HT.
It doesn't.
Hyper-Threading on Intel Processors works by combining 2 Threads into a Single Thread, to Maximise the Efficiency under 100% Utilisation.
Keep in mind that while 100% Utilisation means All Stages are Populated on the CPU, it doesn't mean that the ALU is Fully Populated because most Threads don't actually use every single Operation the Pipeline can actually Perform Simultaneously.
This means HT will result in Higher Temperatures because more Transistors are "In-Flight" at any given time in the same Area.
SMT however doesn't work like this. It isn't Combining 2 Threads,. instead the Processor is made up of 4 "Full" ALU Pipeline and 2 "General" ALG Pipeline.
"Hyper-Threads" are offloaded to the ALG ... but so too are General Workloads via the Thread Balancing., on top of this Precision Boost will actually scale back Clocks rather than increase Voltage. So what you'll see are slightly Lower Clocks, Higher Performance and Lower Temperatures while seeing similar Power Consumption.
It's generally a good idea to always have SMT Enabled, as when it's leveraged you get Free Performance., when it isn't you get more Efficient Power Usage.
-
Re: SMT off and temperature
First thank you for the detailed answer.
Though, while it may explain the lack of lower temps, it still doesn't explain why the temp is actually higher despite less power consumption , and why it generally behaves in an erratic way, unlike smt enabled. That's pretty counterintuitive and therefore I'm curious as to why it happens, and also whether it's an isolated case that I have or its an issue with ryzen. If I had another separate ryzen it would be easier for me to check it myself, but I don't.
Regarding smt and free performance, while generally true, unfortunately , whether due to windows 10 scheduler issue with thread hopping when the workload is lightly threaded (which amd claimes functions properly, but it doesn't...) or lack of optimization by developers ,or a combination of both, currently most games shows an improvement with smt disabled, and since I mostly game, I don't see a reason to leave it enabled for now. But what's holding me off is this weird temperature behaviour when you disable smt and I was hoping someone would shed some light on it.
-
Re: SMT off and temperature
binder87 wrote:
First thank you for the detailed answer.
Though, while it may explain the lack of lower temps, it still doesn't explain why the temp is actually higher despite less power consumption , and why it generally behaves in an erratic way, unlike smt enabled. That's pretty counterintuitive and therefore I'm curious as to why it happens, and also whether it's an isolated case that I have or its an issue with ryzen. If I had another separate ryzen it would be easier for me to check it myself, but I don't.
Okay, well I can see how it might seem counter-intuitive but nothing a quick refresher on how Electronics work can't fix.
A good way to conceptualise Electricity is to imagine that it's Water... we can literally say that Voltage is Flow Rate, while Ampere is the Maximum Pipe Volume.
As the Flow Rate increases, so too does the Total Volume of Water., increasing the Pressure of the System.
Now there are two ways to alleviate pressure... either we reduce the Flow Rate *or* we increase the Length of the System., as the larger the System we have the more Total Volume of Water is required to achieve an Ideal Pressure.
Even if we're beneath said Ideal Pressure though, the Total Volume of Water in the System still Increases by the amount of Piping that we've added.
If we were to say have a totally closed system going above the Ideal Pressure is actually a bad idea, because what you'll be doing is creating undue stress on the Pipes until they eventually just burst., and as you can imagine that isn't exactly useful for Electronics. As such they use a more "Open" Approach, where Excess Pressure is bled off in the form of Heat., in fact they're designed explicitly to have a much higher operational Pressure Threshold because of this; plus it cuts down on the need for additional materials in construction reducing costs.... blah, blah, blah.
The point here is that because AMD SMT works by enabling a Larger System rather than trying to Maximise the Existing System to operate at it's near Maximum Pressure Constantly., this means that while the Voltage remains the same as does the Amps; the Total Volume (TDP) increases inline with the additional Pipelines that have been Added... it also means that Overall System Pressure DECREASES by the same Margin.
In other words you're trading more Power Efficiency for Performance., the Lower Temperature is merely a side effect of their Lower Efficiency of Performance / Watt.
Hopefully that explains it a bit better.
binder87 wrote:
Regarding smt and free performance, while generally true, unfortunately , whether due to windows 10 scheduler issue with thread hopping when the workload is lightly threaded (which amd claimes functions properly, but it doesn't...) or lack of optimization by developers ,or a combination of both, currently most games shows an improvement with smt disabled, and since I mostly game, I don't see a reason to leave it enabled for now. But what's holding me off is this weird temperature behaviour when you disable smt and I was hoping someone would shed some light on it.
I would like to see some proof of Performance Degradation with SMT today., as from Internal Tests it hasn't been an issue since April / May 2017.
Now as to what exactly was causing this, well neither Microsoft or AMD would actually admit that there was even an issue., instead it was simply resolved shortly after the Ryzen 5 Launch.
As far as I could tell it was never an issue with the Windows 10 Scheduler, but rather Power Management / Optimisation.
If you're still experiencing said issues, I'd make sure you have the latest Motherboard BIOS and Chipset Driver... especially setting Power Management to Ryzen(tm) Balanced or Windows Performance, will generally resolve it.
Still Motherboard BIOS updates to support SMT / Auto Mode., does help as it'll automatically decide when to disable it to maintain performance.
-
-
-
-
Re: SMT off and temperature
So, while I liked your analogy, I disagree with the suggested explanation. No need to overcomplicate, its pretty simple. Running six cores and 12 threads consumes more power, that's a fact. More power consumption raises the tdp, and that increases the heat...there's no way around it. If with smt disabled for example the CPU plus its soc consumes 10-15w less than with smt on, it should produce less heat, its the same system. That's why power consumption and tdp are related, after all tdp scales with power consumption, and tdp is used as a quantifying factor for thermal solutions. There's no reason why the same system which works in the same way should produce more heat while consuming less power, it just doesn't happen (on purpose I say the same system , not different systems )
Do a test , disable 2 cores, make a 1600 a 1500x (2+2 downcore control ) , use the exact voltage for both , and see how temperatures drops significantly ...its simply because it consumes less power, its all all there's to it. There's something going on with the smt and temps, and i don't know if its only my rig or not, but something is off, it behaves completely differently in terms of temp sensing that with smt enabled.
Regarding the issue with smt and gaming, it's indeed still on, it's just not being discussed anymore because gamers are a small market share and they kind of just accepted it probably. The latest benchmark I saw was from august I believe made by ancient gameplay on YouTube, well after may, with an updated hardware/windows.
If its still wasn't an issue by the way, programs like project mercury by svent_bent weren't being actively updated still...
You can benchmark by yourself and see that in lightly threaded applications like many games, smt disabled yield better performance. Sometimes its marginal, but it points to some issues.
Dont get me wrong, I love my ryzen, a lot, and I think the 1600 is the best value CPU of 2017, but it is important to bring potential issues to amd's attention to get a better platform , and not just dismiss it .
If its an isolated case in my system only, obviously this discussion is irrelevant , but if its something which people can reproduce, it either needs to be addressed ,or to have a formal , scientific explanation, and not an assumption .
-
Re: SMT off and temperature
Update to whoever is interested; one of my suspicions was confirmed, it was my bios. With a bios update, the issue is gone, temps appear stable, and not all over the place as it was. vrm temps are noticeably lower, with CPU temps roughly the same, with maybe a minor decrease. The vrm temps are decreased by around 6-8c.
Since the imc is less stressed, the memory timings could be lowered further (while maintaining stability), which has proven to benefit games a lot. so currently I am going to keep these settings as I practically only game on this ryzen system .
