4 replies to this topic

[KrEniM MinErK]

tl:dr MWO currently dislikes XFR / Precision Boost. Install AMD chipset driver and use "Ryzen Balanced" Power Plan in Windows 10 to avoid bad frametimes in MWO.
----------------------------------------------------------------

When I got new rig with a Ryzen 7 2700X, MWO performance decreased heavily compared to my old i5-4690K (4 Cores @4GHz), although the Ryzen is very close in terms of clock speed (~4GHz) and IPC performance (~Haswell).

About one year ago, AMD annouced the optional Ryzen Power Plan, which comes with the AMD chipset driver package, was no more necessary as with Windows 1709 "Balanced" Power Plan should include all features of the optional plan.

Long story short:
That was more or less marketing BS. I have yet to find a game that will not suffer from XFR / Precision Boost. Wait. What? More CPU speed should always be better? Well, not exactly. Some software and many games seem to have still trouble with variable clock rates on Ryzen. This might be true for Intel, too. I would be happy if someone could test this here with MWO.

Let's have a closer look with MSI Afterburner (please see link. Unable to embedd the pictures here):
https://www.dropbox...._ryzen.png?dl=0

As you can see, the Power Plans "Ryzen Balanced" and "High Performance" (both disable XFR), resulting in stable 4GHz clock when MWO is running. "Balanced" boosts the CPU to about 4,1GHz and drops the clock from time to time. There is a small impact to avg and total fps with "Balanced", but that is only about a 10% performance loss.

Now let's have a look at the lowest graph: Frametimes
With "Balanced" the frametimes are roughly 50% higher, resulting in sluggish gameplay performance / minor stalls in MWO.
Other games I benchmarked show a smaller performance hit as MWO, e.g. Final Fantasy XV (Benchmark. Full run @Graph): https://www.dropbox....etimes.png?dl=0

FPS seems identical, but look at those framerate spikes ...

Are there any downsides in using the Ryzen Balanced powerplan?
Yes. Power consumption is higher when the CPU runs at idle (10 Watts for my rig), although not as high as with "High Performance" Power Plan. The disabled XFR lowers the potential maximum clock of the CPU, which can slow down other software. For me, XFR only added something like 100MHz, not noticeable and certainly not worth the trouble.

[NARC BAIT]

the motherboard that you use can have a big impact on how well the second gen ryzen really performs in the field, with XFR ... teh way AMD configured XFR works much better than on the first generation, but, its still kinda junk if you care about the actual performance of the chip ( ie, gaming )

XFR boost the chip to its maximum, generally to a level that is not maintainable, that it will not be able to sustain, and the method works quite well for most 'workloads', but generally not for gaming, and the reason why it wont work the best for gaming, is because threads will tend to stay on a singular core more often, which is all good and happy during a 'boost' phase, but before too long, the core holding the main thread will have to backoff, to either P-state 1 or 2, and thats probably the spike you can see in frametime

the real problem comes back to current / wattage / TDP limits, I'm currently running a 2600x, rated to 95w, the second I go above that rating, things go bad, with a 2700x, you get an extra 10w, for the extra core ... problem there, is that a core could max out at about 16w ... so the entire time, your CPU is fighting constraints set by AMD .... anyway, the takeaway piece from this is your wattage, which will be listed in HWinfo64, I cant tell you what its called at the moment, because I have them disabled, but thats another story ...

2600x -  95w / 6 cores = 15.83w per core
2700x - 105w / 8 cores = 13.12w per core 

next, onto SMT, this is where it really all gets convoluted, SMT is equivalent to intels hyperthreading, and in theory allows one core to be execute two threads at the same time, theres been no shortage of caveats over the years, but generally, that 'extra core' per core will tend to yield about a 25% increase to overall number crunching power, on earlier generations, you could get a direct single core thread performance boost, from disabling the 'feature' ... this worked quite well on first gen ryzens, and intel's up to at least the 4th gen ... disabling it on second gen ryzens does not yield a direct boost, however, it lowers the core wattage, alot of the time by 25%, which drastically alters the thermal / performance envelope, which will allow XFR to maintain stupid boosted levels ... disabling SMT should be a case-by-base basis, if your normal workflow include rendering pixar films, you probably want it on, if you play games, optimised for 4-8 threads, you probably dont need 16 .... looking at the two graphs you provided, MWO barely hit a pitiful 25%, whereas final fantasy fourteen million got up to a suspicious 87% cpu usage, I'd be half interested to see that benchmark repeated with SMT off,

so, lets assume you try it all, and disable SMT, suddenly the XFR profile is actually quite over the top, and will be feeding the CPU more than it needs, probably 10% more voltage than it needs, depending on the motherboards ability to deliver power, and how the manufacturer felt like configuring the BIOS ... I cant really say with certainty how much juice your CPU will need, but I ended up tuning mine down to 1.37v @ 4300 mhz with a full load temp around 65c .... at these numbers I'm actually happy enough with the very minor difference between upgrading from a 1600x to a 2600x .... the defaults with XFR, would go up over 1.5v for 4250 mhz

I'm using a p-state overclock, with the high performance plan, this configuration does not downclock at all and draws around 30% of the current limit at idle ... heres a CPU-Z validation https://valid.x86.fr/fjt0lm

[KrEniM MinErK]

I still wonder why the XFR/Frametimes-Topic wasn't more thoroughly tested by the media. Only thing I've ever found regarding that topic was one(!) user comment at the computerbase (German HW Site) forum. Most people have no idea how bad the general "out of the box" performance of their gaming rigs really is

When I selected the CPU for my system, I was aware about the maximum wattage restriction for each core compared to the 2600X, but found the single threaded performance difference (in a XFR-Enabled scenario) neglectable. Plus the general consensus in the media: "No need to (manually) overclock your Ryzen2, as XFR+Precision Boost will do a better job in nearly all cases".

When warrenty runs out for my Ryzen, I am going all core oc like you did. Best thing one can do for gaming, hands down. You might want to have a look at the Ryzen balanced power plan if you want to save some power/heat/noise without losing performance. I'm not sure which P-States are used for this profile, but even a super sensitive application like MWO shows no negative impact.

Regarding SMT, I couldn't find any difference in any benchmark I conducted. Going to link that graph later tonight.
Graph with disabled SMT: https://www.dropbox....en_smt.png?dl=0
SMT off seems to disable XFR. Those graphs are 4 months old, couldn't find the time to write something in detail until now. Going to test this again. Maybe I can get those graphs logged somehow to calculate an avg. delta of the frametimes.

btw. Final Fantasy XV can use up to 8 Threads and requires a 4C/8T CPU for a smooth gameplay experience.
See Graphes "Frame Time Variance" and "Unevenness" here: https://www.tomshard...sis,5513-6.html

PS
Big thx for your comprehensive thread about user.cfg tuning, really enjoyed reading about your findings and recommendations

Edited by KrEniM MinErK, 10 September 2018 - 12:16 PM.

[NARC BAIT]

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

I still wonder why the XFR/Frametimes-Topic wasn't more thoroughly tested by the media.

because of the pace of the medium the media uses these days ... on release day, everyone wants to know performance benchmarks, being slow to release those, could hurt the number of views on a site like youtube ... most viewers are not looking or dont understand the technical nitty gritty ... also the motherboard can have a bigger than expected impact with ryzens, my old 1600x, I paired that with a simple A320 motherboard, and XFR has worked much better than it ever did on my X370, pretty much manages to pull the same numbers that I had to work for, through XFR and on air ...

another thing with my board is that some configurations can actually cause weird microstutters, that alot of benchmark stuff wont even see properly ... I think its a synchronicity issue with the RAM configuration ... maybe its more evident on this modified bios without the HPET timer, that you normally cant disable on a ryzen ...

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

"No need to (manually) overclock your Ryzen2, as XFR+Precision Boost will do a better job in nearly all cases".

completely true for the 'average' user who does not want to learn anything about any of it, I'd think maybe 5% of people would manage to get a better result than it, because it pushes the CPU's to their limits pretty well ...

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

You might want to have a look at the Ryzen balanced power plan if you want to save some power/heat/noise without losing performance.

most of the time my idle has been 30 - 45 celcius, and when the cores are not doing much, they dont draw anywhere near as much current as at full load .... I really dont spent much time in a full load scenario, the game that I'm playing lately only tends to use about 40% of the total, an 8 hour game session might push the temp up to 50c ... I've only got a 120mm AIO cooler on it ... in comparison to the abuse I was putting the 1600x through, everything in my case is having to work less than 6 months ago

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

SMT off seems to disable XFR.

it doesnt actually disable it, but it changes how the profile is interpreted, senseMi can have an impact here also, its one of those things with no right way for all, you'd turn it off for an all core OC, but probably need it on with XFR

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

btw. Final Fantasy XV can use up to 8 Threads

you'd hope after the umpteenth iteration that they were really starting to use some multi core environments ... you know the sad bit, is that the change is really coming about because of consoles ... where its not uncommon to find 8 cores in a 4+4 configuration, just at much lower clocks ( and voltages ) than desktops are pushing ...

the linked results seem a bit all over the place, and they dont seem to have mentioned which methods they used to disable cores .... you'd think there might only be one way, but on my board, I'm presented with two basic options, I can use cores on one side of the CCX, or use cores on both ... with the 1600x plugged in, amongst its options for 4 core modes was 4+0 or 2+2 ... implying to me that that CPU had at least 7 working cores ... anyway, the two different modes basically optimised differing memory channel configurations, potentially skewing results ...

I've ramped my 2600x up over 4.5 ghz in asynch mode ... crashed pretty easily under load in windows ...

[Alastor The Radio Demon]

KrEniM MinErK, on 10 September 2018 - 06:46 AM, said:

I still wonder why the XFR/Frametimes-Topic wasn't more thoroughly tested by the media. Only thing I've ever found regarding that topic was one(!) user comment at the computerbase (German HW Site) forum. Most people have no idea how bad the general "out of the box" performance of their gaming rigs really is

When I selected the CPU for my system, I was aware about the maximum wattage restriction for each core compared to the 2600X, but found the single threaded performance difference (in a XFR-Enabled scenario) neglectable. Plus the general consensus in the media: "No need to (manually) overclock your Ryzen2, as XFR+Precision Boost will do a better job in nearly all cases".

When warrenty runs out for my Ryzen, I am going all core oc like you did. Best thing one can do for gaming, hands down. You might want to have a look at the Ryzen balanced power plan if you want to save some power/heat/noise without losing performance. I'm not sure which P-States are used for this profile, but even a super sensitive application like MWO shows no negative impact.

Regarding SMT, I couldn't find any difference in any benchmark I conducted. Going to link that graph later tonight.
Graph with disabled SMT: https://www.dropbox....en_smt.png?dl=0
SMT off seems to disable XFR. Those graphs are 4 months old, couldn't find the time to write something in detail until now. Going to test this again. Maybe I can get those graphs logged somehow to calculate an avg. delta of the frametimes.

btw. Final Fantasy XV can use up to 8 Threads and requires a 4C/8T CPU for a smooth gameplay experience.
See Graphes "Frame Time Variance" and "Unevenness" here: https://www.tomshard...sis,5513-6.html

PS
Big thx for your comprehensive thread about user.cfg tuning, really enjoyed reading about your findings and recommendations

Most people who are jumping into Ryzen already know XFR is bad, Intel turbo is bad, overclocking AND having XFR/turbo enabled is bad, all forms of power saving at the BIOS level (C1E CnQ, enhanced C states) are bad, v-droop in general is bad. Get good ram, adjust the timings to get above 3200MHz, turn off XFR/turbo, I got my TR 2990WX running 4000MHz ram, XFR disabled, C1E and other power savings disabled, clocked to 4.4GHz across all cores and can game in mwo all day at a steady 160fps+ @10320x1440.