CPU Cache and MMOs Discussion

[Agall]

Has anyone else had similar experience when upgrading to a 13900k, 5800x3D, 7000x3D from any previous CPU and seen disproportionate gains in performance in MMOs specifically? As in higher % gain than expected overall in these games, especially in high player density environments?

 

I've done two system upgrades with drop in 5800x3D, both of which users were primary MMO players at the time (WoW, Warframe, Planetside 2)

 

First system was a 3950x (tuned since it was my system) to 5800x3D CPU swap with a kit of 2x16GB 3600MHz CL18. A primary control was a relatively low population scenario, defense mission in Warframe in Orokin Void. On a 6900XT at 4K, the framerate was ~160fps which went up to ~240fps, being a 50% gain. In WoW (which at the time I played, Shadowlands) sitting in Oribos also netted ~50% increase. Planetside 2 overall experience is far more difficult to measure, since the player count is particularly limting on that game, but going from a 3950x resulted in close to double the framerate, ~45 fps to ~90 fps, though that's subjectively gauging a similarly sized battle. Though with this game, the test wasn't immediately after a CPU swap like the other two.

 

Second system was my brother's R5 3600 (factory, he's not an overclocker), which got a CPU and RAM swap to my previous 5800x3D and 2x16GB 3600MHz CL18 kit, once I got a 7950x3D. This was recent (obviously) and in the initial testing in WoW, sitting in the new city in Dragonflight, his framerates were doubled. He doesn't play Warframe and has only launched Planetside 2 twice, so I didn't have data before or after for these comparably. 

 

The theory being that in high CPU draw call games like MMOs with tons of player assets to handle, CPUs with markedly higher cache, being 3D v-cache CPUs and even CPUs like the 13900k, should disproportionately benefit MMOs. In my testing so far, when I swapped both an R5 3600 (mind you with a RAM upgrade) and the 3950x, the gain in performance in MMOs was disproportionally higher than the gain in performance for other games.

 

Overall between the 5800x3D vs 7950x3D in these games did not net nearly the performance gain as from upgrading from non 3D processors, providing the theory that 3D v-cache disproportionately benefits these titles. I could very easy test this theory with the 7950x3D by disabling CCDs in WoW specifically, though I don't play that game anymore, and Planetside 2 is very hard to control for. Warframe only occasionally has a scenario where there's a high player count to test this, though that's neigh impossible to control since each session instances differently.

 

The 13900k should be relevant in this discussion, since the amount of L2+L3 cache between 10, 11, 12, 13th gens was quite massive, tripling between 10th and 13th gen and a 1.5x gain between 12th and 13th gen.

 

The obvious problem is no major benchmarking source is going to do extensive CPU benchmarking in MMOs, since they're very difficult to control in terms of variables. In my experience, this is independent of resolution, whether 1080p, 1440p, 1440p UW, or 4K, which indicates a heavy CPU limitation. That heavy CPU limitation in particular being in high player count areas like cities, raid groups, large battles, etc.

 

I've always seen it as a missed opportunity since the academic benchmarking is generally consisting of AAA games that I would imagine most gamers aren't actually playing, versus the ever popular MMOs. The data I've gathered myself and seen from various sources has driven me to differentiate between academic benchmarks like such and practical benchmarks like what I've gathered in particular games, as someone who will disproportionately stares at framerates while gaming.

 

From Ryzen 5 3600 to 5800X3D: The Big Upgrade | TechSpot

 

From this article, one particular value being of note, the 1% lows. This value in games like MMO can be crippling in some areas with lesser CPUs, something that 3D v-cache seems to mitigate. I would assume MMOs overall would fall quite high on this tier list, probably >100%.

 

[Kunt]

This is extremely interesting. I think testing this with MMOs are quite possible, even in situations like 40-man raid or world pvp. Because I already see greater deviation in performance in games like PUBG. Hardwareunboxed would definitely be interested in this.

[Agall]

1 hour ago, Kunt said:

This is extremely interesting. I think testing this with MMOs are quite possible, even in situations like 40-man raid or world pvp. Because I already see greater deviation in performance in games like PUBG. Hardwareunboxed would definitely be interesting in this.

 

If it puts into perspective, it really took over a dozen different sample benchmarks in a single game to get a somewhat accurate measurement of the difference. Warframe is variable enough to have anywhere from no difference to nearly 3x.

 

I've also done testing with each CCD to get a 1:1 comparison between 3D v-cache and not, and the framerate consistency in any game was noticeable.