Fallout 4 CPU scaling

[cbn]

http://gamegpu.ru/rpg/rollevye/fallout-4-test-gpu.html

Pleasantly surprised to see respectable core scaling here, particularly the i7 3970X (Sandy Bridge 6C/12T, 3.5/4.0 turbo) vs. i7 4770K (Haswell 4C/8T, 3.5/3.9 turbo). Interesting to see an older architecture with 6C/12T able to beat a more modern 4C/8T configuration. This especially when we see how the i5 4670K (Haswell 4C/4T) pulls ahead of the i5 2500K (Sandy Bridge 4C/4T) by 16% in both avg. and min. FPS (indicating a strong IPC advantage for Haswell over Sandy Bridge in this game).

i7 2600K (Sandy Bridge 4C/8T) vs. i5 2500K (Sandy Bridge 4C/4t) shows a very strong gain of 20% greater avg. FPS and 26% greater minimum FPS in favor of the i7. (Hyperthreading scaling well for the quad core in this game)

Even the Bulldozer FX-8150 @ 3.6 is almost 10% faster than the Vishera FX-6300 @ 3.5 in minimum frame rates.

FX-6300 @ 3.5 shows a marginal improvement over the FX-4300 @ 3.8. But not bad considering the clock difference.

P.S. With those results noted, I do wonder how much the cache differences between processors factor in?

 

[Headfoot]

P.S. With those results noted, I do wonder how much the cache differences between processors factor in?

My guess is fairly substantially, some user testing here on the forums has shown massive frame rate differences based on memory speed with DDR3 1600 showing a huge deficit versus DDR4 3000. My guess is that cache is helpful as the game seems to be memory bound

 

[frozentundra123456]

Another cpu bench http://www.gamersnexus.net/game-bench/2182-fallout-4-cpu-benchmark-huge-performance-difference

FX fares much more poorly here.

 

[Dave2150]

Odd that they don't include the 6700K in these results.

 

[dark zero]

Another cpu bench http://www.gamersnexus.net/game-bench/2182-fallout-4-cpu-benchmark-huge-performance-difference

FX fares much more poorly here.

Actually people noted that the reason was RAM Latency.... Pretty weird how Bethesda did the game

 

[SPBHM]

Another cpu bench http://www.gamersnexus.net/game-bench/2182-fallout-4-cpu-benchmark-huge-performance-difference

FX fares much more poorly here.

because it's a huge game, you can compare CPUs while doing CPU limited things or not, also maybe gamegpu lowered some settings? (shadows distance have a big effect on CPU perf)

 

[ShintaiDK]

Gamegpu doesn't test at ultra settings and not at a hard location, that's why.

 

[Leyawiin]

Gamegpu doesn't test at ultra settings and not at a hard location, that's why.

Which makes their findings pretty much useless.

 

[frozentundra123456]

They certainly seem to be falling behind. They dont even have a 4790k, much less skylake. Used to be my go to site.

 

[toyota]

yeah depends on where you test in this game as some parts need insane memory speeds and cpu power. even I drop to just under 50 fps in spots even though my cpu nor gpu appears to be getting taxed.

even in Fallout New Vegas, a 2500k could not maintain 60 fps in some spots so that just shows how crappy this old engine is.

 

[cbn]

Gamegpu doesn't test at ultra settings and not at a hard location, that's why.

Shintai, in the translated text Game GPU did mention testing 720p at maximum quality settings:

And turns out "maximum" did specifically mean ultra. English to Russian translation of ultra here:

https://translate.google.com/#auto/ru/ultra

Ultra = ультра (and notice it listed in the first image below)

CPU test

CPU-testing we have conducted on 16 models of base configurations that are relevant today. The test was conducted in areas where the value of video games for the minimum and its load was less than 99%, this time at a resolution of 1280x720 at a given quality setting.

Testing at maximum quality settings 1280x720

CPU performance in the game enough for all models.

Loading processor cores at maximum quality settings 1280x720 Intel%

Loading processor cores at maximum quality settings 1280x720 AMD%

With that mentioned, moving from ultra to medium is going to remove bottleneck from the GPU.

For example, notice how in the following benchmark i7 5930K (Haswell 6C/12T, 3.5 GHhz base/3.7 Ghz turbo) makes a stronger gain in .1% low frame rate compared to the i7 4790K (Haswell 4C/8T, 4.0 Ghz base/4.4 Ghz turbo) when comparing 1080p ultra to 1080p medium.

i7 4790K: 57 FPS to 65 FPS (.1% low frame rate, 1080p ultra to 1080p medium)
i7 5930K: 57 FPS to 72 FPS (.1% low frame rate, 1080p ultra to 1080p medium)

http://www.gamersnexus.net/game-bench/2182-fallout-4-cpu-benchmark-huge-performance-difference

 

[SPBHM]

for this game, you really need to know how they tested it, the part of the map and so on, gamegpu surely tested it in a less than good enough place to test CPU performance considering their results (to ridiculously high for all the CPUs),
and as usual, most of the graphics settings affect more the GPU, but not all, some have a clear effect on CPU load.

 

[ElFenix]

i'm suspicious because the 4670k is behind the 2600k. that doesn't seem like it should happen. ever.

 

[cbn]

P.S. With those results noted, I do wonder how much the cache differences between processors factor in?

My guess is fairly substantially, some user testing here on the forums has shown massive frame rate differences based on memory speed with DDR3 1600 showing a huge deficit versus DDR4 3000. My guess is that cache is helpful as the game seems to be memory bound

Some FO4 core scaling in the following review using a 2.5 Ghz i7 -5820K with 2C/2T, 2C/2T 2C/4T, 4C/4T, 4C/8T, 6C/6T, 6C/12T configurations:

https://translate.google.com/transl...hmark-vor-Release-1177284/&edit-text=&act=url

(I'm assuming all the core configurations have the same 15MB L3 cache?)

The 6C/12T configuration gains 6.5% higher average FPS and 5% higher minimum FPS over the 4C/8T.

So fairly modest gains.

With this pointed out, I do wonder what would have happened if the cores were clocked higher? (re: This to remove single thread as a possible bottleneck holding back FPS gains with the 6C/12T vs. 4C/8T)

 

[cbn]

i'm suspicious because the 4670k is behind the 2600k. that doesn't seem like it should happen. ever.

Look at the core scaling results from pcgameshardware.de I linked in post #14.

4C/8T is 10% faster than 4C/4T in average FPS and 8% faster in minimum FPS in the benchmark they used. And I do wonder if this gap would have been even larger if a higher than 2.5 Ghz clock speed was used.

Furthermore, also consider i7 2600K has 2MB more L3 cache than i5 4670K.

 

[Erenhardt]

FX seems to have some bottleneck. The scaling with clock speed seems to fall after 4GHz:

Maybe it is RAM speed, maybe cache latency, maybe even both. Hard to guess. Game scales very good (even too good!) with RAM speeds, but not with fx:

Also, would love to see CPU usage pattern of fx4300 on fx6300 and fx8300. That would be amazing.

 

[Erenhardt]

In the case of i7 it scales 1:1 with memory clocks, which is strange. Even IGPs do not show such a benefit in games.

 

[SPBHM]

would be interesting to overclock the L3/IMC on the AMD CPUs.

 

[ElFenix]

Look at the core scaling results from pcgameshardware.de I linked in post #14.

Wait, I didn't notice the 4670k is an i5

/grumbles about useless inconsistent model numbers

 

[ShintaiDK]

Wait, I didn't notice the 4670k is an i5

/grumbles about useless inconsistent model numbers

It is consistent as such. Remember the 2700K

 

[MajinCry]

I don't trust these benchmarks. I remember reading up on Skyrim's, with a huge disparity between actual framerates and what was shown in the graphs.

For testing, they should just stand at the top of Diamond City's stairs, overlooking the entire city. Makes around 6,200 draw calls with shadow distance @ 8000 (Medium = shadow distance 3000, ~4.5k draw calls, High = shadow distance 13000).

 

[TheELF]

For testing, they should just stand at the top of Diamond City's stairs, overlooking the entire city. Makes around 6,200 draw calls with shadow distance @ 8000 (Medium = shadow distance 3000, ~4.5k draw calls, High = shadow distance 13000).

How did you get these numbers?

 

[Head1985]

if you want cpu bottleneck scene try this save and post here screenshot.Important thing is FOV must be at default!(default is 80 BTW)
Also max details are important.Lower resolution if needed(if you have slow video card like GTX760 but keep max details)
http://www.ulozto.net/xxfqHqjH/save...19-2-fos?_ga=1.61009964.1857298745.1440807563

 

[Face2Face]

if you want cpu bottleneck scene try this save and post here screenshot.Important thing is FOV must be at default!(default is 80 BTW)
Also max details are important.Lower resolution if needed(if you have slow video card like GTX760 but keep max details)
http://www.ulozto.net/xxfqHqjH/save...19-2-fos?_ga=1.61009964.1857298745.1440807563

I'll try this later tonight. I assume this is Diamond city? I haven't made it there yet, but everyone says it's brutal on CPUs.

 

[Head1985]

diamond city in not brutal at all at cpu.That scene is far more demanding than diamond city.
So far only skylake was able more than 60fps in that scene.