Ametuer efficiency core & performance core benchmark comparisons

[Evelynn]

(Just realized I should've posted this in Software. Apologies!)

Hi everyone!

I wanted to run a few benchmarks of the efficiency cores and see how they compare against the performance cores. It's nothing too in-depth, I simply stand in one spot and allow Intel Presentmon to collect data for a few minutes. In reality the efficiency cores have a higher core-to-core latency, which can result in lower 1% / 0.1% lows than the performance cores. (Source: Anandtech)

Here are some of the averages I collected. I chose these games intentionally for two reasons. I play these games very often, on an almost daily basis if I have the time. Also, they're much older than anything modern, meaning they're very much so bound in memory - latency - and especially IPC. These aren't games that reviewers typically test, as they aren't very popular with their general audience. All tests were done with my 13700k @ 5.3 GHz P / 4.2 GHz E / 7200 DDR5 / 46x ring.

Efficiency cores will always be the first image on the left, with the lower framerate.
Second image will always be the performance cores.
Third image is a picture of the scene.

Guild Wars 2: Lion's Arch (City environment, high NPC / Player count, severely single-threaded)

Final Fantasy 14: Limsa Lominsa (City environment, high NPC / Player count, also very single threaded but memory sensitive as well)

Crysis 1: (Very, VERY single threaded game)

Lastly, Hogwarts Legacy: (Uncertain as to the main bottleneck. Game is modern and multithreaded)

Typically, the performance cores are around double that of the efficiency cores in frame rate. In synthetics, they're a little more than half the IPC of a performance core. Realistically, the core-to-core latency causes very noticeable stutters depending on the title you're playing.

 

[dgianstefani]

Yep e cores are for background tasks and the OS while all eight p cores can focus on the game/3D application.

Intel has thread director which allows fine tuned setting of cores to processes depending on the game too. I expect that allocates some e cores to game processes that don't need P cores, given the tech allows for better than stock performance.

 

[Onasi]

I expect that allocates some e cores to game processes that don't need P cores, given the tech allows for better than stock performance.

It does. Seemingly, if it detects that one of the threads that the game process occupies is a light load it will be offloaded to an E-Core. That’s actually the reason why all the “disable E-waste cores for moar FPS!!111one” cries are wrong - the more multithreaded the games become, the more use out of those E-cores they get. In fact, we’ve seen already examples of games where turning them off hurts performance.

 

[dgianstefani]

It does. Seemingly, if it detects that one of the threads that the game process occupies is a light load it will be offloaded to an E-Core. That’s actually the reason why all the “disable E-waste cores for moar FPS!!111one” cries are wrong - the more multithreaded the games become, the more use out of those E-cores they get. In fact, we’ve seen already examples of games where turning them off hurts performance.

The e cores together with hardware scheduling are excellent.

 

[Dr. Dro]

I wish you had an NVIDIA card on-hand. The behavior you'd see from the CPU would be drastically different, especially in the DX11 games where its driver can make use of command lists and deferred contexts that AMD doesn't support.

Egregious example of this would be Fallout 4, top of the Corvega plant where FPS is known to tank wildly