CPU performance benchmarks

[orzelek]

One question would be:
Are devs willing to prep Ryzen optimized executable of Factorio when there is a toolchain ready for it?

Most stuff is now Intel optimized and Ryzen optimizers will come sooner or later - question is if separate exe would be needed for that or you could make one and switch depending on CPU in runtime.

[bergi9]

I am using this map viewtopic.php?f=204&t=34068 and got at start 34UPS, after a while it went down to 27UPS. Only the Base-Mod.
According to Process Explorer it consumes 6% Cpu usage across all cores.
AMD R7-1800X default clock and 2133mhz memory. Tomorrow i get a new mainboard which i can overclock (the current mainboard cant overclock, buyed wrong one).

[daniel34]

I am using this map viewtopic.php?f=204&t=34068 and got at start 34UPS, after a while it went down to 27UPS. Only the Base-Mod.
According to Process Explorer it consumes 6% Cpu usage across all cores.
AMD R7-1800X default clock and 2133mhz memory. Tomorrow i get a new mainboard which i can overclock (the current mainboard cant overclock, buyed wrong one).

Since I was interested in the Ryzen I compared this to my notebook (faster than my desktop which needs replacement) and was surprised to get almost the same numbers. 35 UPS for the first minute of running the map, then it goes down to 30 UPS at the start of the night (nightvision activates) and around 27-28 UPS at the end of the night.

The notebook is not that new, it has an i7-3537U with 2 physical cores, launched Q1'13 (4 years ago), 2.0/3.1 GHz base/turbo frequency, 17W TDP. It has 6GB RAM that run at 1600 MHz.

I thought the Ryzen would have at least some improvement over a 4 year old laptop processor since it also has a higher frequency (3.6/4.0 GHz base/boost) and uses faster RAM, but apparently Intel is still the preferred choice for Factorio. Or has someone other ideas what the reason could be?

[MeduSalem]

That might not happen since probably it just falls back on generic Am64. That would probably run fine, just a bit slower than it could.

There may still be unknown CPU/microcode bugs not even AMD is aware about yet and which only shine through in the wild in very special situations they never tested for. (Though the chance is really low of something like that happening with generic software)

That or the funky stuff AMD is always doing in their micro-architectures, like the 2x4 Core setup that requires special adjustments of the OS thread scheduler to cope with the otherwise increasing Cache miss rate or thrashing due to moving the threads between two cores of the seperate modules. You never really know with AMD since they are always on the edge of trying experimental/risky stuff or to do things otherwise differently than Intel does... they seldom play it 100% safe and that usually results in software having to adjust.

Same with Bulldozer which also required an update of the thread scheduler to deal with the Bulldozer Modules.

I also remember back when they released the first Athlon X2 dualcore CPUs that each core used to return a different elapsed real time (only the devil knows why they did that instead of a global time that is the same for all cores)... and various applications crashed shortly after starting because of synchronization problems that arised between their threads. It required a microcode/OS driver update to deal with that.

So with AMD there is always those kind of technical teething problems due to them trying to be leading-edge. Over the years I've grown accustomed to that.

[bergi9]

Update: After disabling the High Precision Event Timer (HPET) in BIOS the UPS went to 43, after couple of minutes it went down to 34 UPS, then back to 40 UPS.
According to the CEO of AMD Lisa Su it's an unoptimized software. (for who want read the reddit: https://www.reddit.com/r/Amd/comments/5 ... ?context=3).
Many monitoring softwares can't read the ryzen stats and temps correctly.

[orzelek]

Update: After disabling the High Precision Event Timer (HPET) in BIOS the UPS went to 43, after couple of minutes it went down to 34 UPS, then back to 40 UPS.
According to the CEO of AMD Lisa Su it's an unoptimized software. (for who want read the reddit: https://www.reddit.com/r/Amd/comments/5 ... ?context=3).
Many monitoring softwares can't read the ryzen stats and temps correctly.

Did you also disable SMT?

Try it if not - with it enabled scheduler on windows breaks performance apparently. Results from this post are around what I get on my i7-4790K currently.

[bergi9]

Did you also disable SMT?

Didn't help at all.

After overclocking the CPU to 4ghz and memory to 2666mhz i gained additional 4 UPS. Seems to be i can't overclock the memory further to 3200 because my memory isn't on the support list of my mainboard and i'm using 4x16gb modules. if someone knows the correct timings (for ASUS Prime-X370-Pro and G.Skill Trident Z 3200), please inform me.

[Zeblote]

So basically the ryzen is garbage for factorio. That map runs at 60 ups on my 6700k...

[orzelek]

So basically the ryzen is garbage for factorio. That map runs at 60 ups on my 6700k...

Hmm so my decision to go with 7700K potentially might be a good one for Factorio then since it should be faster. I'm silently hoping that intel will drop prices a bit after Ryzen settles down a bit.
Do you have quad channel memory? That might count for Factorio along with memory speed and latency.

[Zeblote]

So basically the ryzen is garbage for factorio. That map runs at 60 ups on my 6700k...

Hmm so my decision to go with 7700K potentially might be a good one for Factorio then since it should be faster. I'm silently hoping that intel will drop prices a bit after Ryzen settles down a bit.
Do you have quad channel memory? That might count for Factorio along with memory speed and latency.

No quad channel, though it is 3000mhz at cl15. The latency is about half what I see in ryzen reviews with comparable memory, so that may have something to do with factorio running up to twice as fast. There definitely is something wrong with that ryzen...

[aober93]

Did you also disable SMT?

Didn't help at all.

After overclocking the CPU to 4ghz and memory to 2666mhz i gained additional 4 UPS. Seems to be i can't overclock the memory further to 3200 because my memory isn't on the support list of my mainboard and i'm using 4x16gb modules. if someone knows the correct timings (for ASUS Prime-X370-Pro and G.Skill Trident Z 3200), please inform me.

I did some numbers, and the 25% increase in memory speed yielded 10% more fps. Thats very high imo. Most games gain either nothing or like 1% here.

Well i realize the 1800x is only 3,6ghz and 4boost, but my 3,5ghz (3,6boost) intel 3570k gets 47fps in that map and doing so with 1600mhz memory, albeit not ddr4 so thats not really a point for intel i guess, its got cl 11. Hard to compare that really.
Nothing really surprising but like on the negative end of all negative reviews of the ryzen game performance i guess.

Btw you can read out the memory profile with cpuz right? (This is for G.Skill Trident Z 3200 MHz C16 DDR4 (2x 8 GB) F4-3200C16D-16GTZ)

[striepan]

Don't forget that multi-threaded update is coming in 0.15. I'm curious how it changes the game regarding Ryzen which is supposed to be good on multi-threaded workloads.

[Rseding91]

Don't forget that multi-threaded update is coming in 0.15. I'm curious how it changes the game regarding Ryzen which is supposed to be good on multi-threaded workloads.

No it's not. It didn't give any worthwhile gains that couldn't be gotten by optimizing it without threads while also complicating the code and making it harder to debug.

[striepan]

Don't forget that multi-threaded update is coming in 0.15. I'm curious how it changes the game regarding Ryzen which is supposed to be good on multi-threaded workloads.

No it's not. It didn't give any worthwhile gains that couldn't be gotten by optimizing it without threads while also complicating the code and making it harder to debug.

Did I miss some mention of it or is it the first? It's still in the road map.
Anyway, we shouldn't be surprised. From what we know about current architecture, there isn't much to be gained by parallelisation.

[Loewchen]

Don't forget that multi-threaded update is coming in 0.15. I'm curious how it changes the game regarding Ryzen which is supposed to be good on multi-threaded workloads.

No it's not. It didn't give any worthwhile gains that couldn't be gotten by optimizing it without threads while also complicating the code and making it harder to debug.

Did I miss some mention of it or is it the first? It's still in the road map.

It has been mentioned several times before but not in a FFF afaik, the last time the roadmap has been updated was in august, so some information is obsolete (like the GUI update).

[ChoMar]

I dug this one up because id still want to know if Factorio will be compiled or even optimized for maximum Ryzen performance.
I dont know the switches to set in the compiler, but im sure their are some that Factorio could use.

[Mehve]

AMD Community Update: BIOS Updates, Patches, Performance Improvements(TPU)

Promising, if not earth-shattering news. If anyone with a Ryzen rig who regularly updates their BIOS happens to get a before and after reading, that might be interesting.

[bergi9]

AMD Community Update: BIOS Updates, Patches, Performance Improvements(TPU)

Promising, if not earth-shattering news. If anyone with a Ryzen rig who regularly updates their BIOS happens to get a before and after reading, that might be interesting.

well, thats too late for me. I have returned the AMD Cpu and Mobo and changed to intel i7-6850k. Way better speed in Single-Threading and much higher NVMe speed.

[MrHick]

R7 1700 @3,9 ; ddr4 @3200 cas16; MSI GTX 1080; MSI B350M mortar
4k Screen
Loaded that reference map.
When I load the Map I get 40-44 FPS/UPS (Standing there not moving the mouse to make sure it is the same as everyone else)
When the night comes in it goes down to 33-35 FPS/UPS then comes up to 40+ ish.

2 Cores are running 75-80% rest are 0-10%

Lowest I seen it dip was 31 and highest was 48

[Loewchen]

R7 1700 @3,9 ; ddr4 @3200 cas16; MSI GTX 1080; MSI B350M mortar
When I load the Map I get 40-44 FPS/UPS (Standing there not moving the mouse to make sure it is the same as everyone else)
When the night comes in it goes down to 33-35 FPS/UPS then comes up to 40+ ish.

I got slightly better numbers with my six year old i5-2500