Far Cry 4 and AMD cpus added to AT linux-bench

[Essence_of_War]

Far Cry 4 CPU
i5's look like the limit of scaling at 1920x1080 w/ a 980 but notably, even the 4170 and i3 are able to keep their nose above 40 fps mins.

Anandtech added Linux-Bench for a bunch of AMD cpus finally!
c-ray
7z Compression
7z Decompression
Molecular Dynamics
Fluid Dynamics
OSSL sign
OSSL verify
redis 1x
redis 10x
redis 100x

 

[Abwx]

Far Cry 4 CPU
i5's look like the limit of scaling at 1920x1080 w/ a 980 but notably, even the 4170 and i3 are able to keep their nose above 40 fps mins.

Anandtech added Linux-Bench for a bunch of AMD cpus finally!

Fluid Dynamics

The scores are dubbious, since it s a score per thread this imply that a i7 core + HT has twice the throughput of a core without HT, an i5 core for exemple, should i conclude that at equal frequency an i7 should score twice the i5 score.??.

 

[Essence_of_War]

HT scales really well (up to ~2x!) in float/double intensive workloads where there are lots of thread stalls. It's plausible that CFD workloads might have this sort of scaling (lib calls, waiting on RAM, etc). I don't do any CFD personally, and I don't have the source, so maybe someone who knows the workload better could comment.

 

[Abwx]

HT scales really well (up to ~2x!) in float/double intensive workloads where there are lots of thread stalls. It's plausible that CFD workloads might have this sort of scaling (lib calls, waiting on RAM, etc). I don't do any CFD personally, and I don't have the source, so maybe someone who knows the workload better could comment.

Scaling between a i5 and a i7 is certainly not 90-100% more perfs because of HT.

 

[Enigmoid]

Scaling between a i5 and a i7 is certainly not 90-100% more perfs because of HT.

Depends on the code and the compiler. If calculations have to be fetched out of main memory it certainly could be.

 

[Essence_of_War]

HT scales really well (up to ~2x!) in float/double intensive workloads where there are lots of thread stalls.

Scaling between a i5 and a i7 is certainly not 90-100% more perfs because of HT.

I don't think you're hearing what I'm saying.

The absolute best case scenario for hyper-threading is a workload with a ton of cache misses, library calls, data dependencies, etc. In cases where threads are constantly stalling due to this stuff, hyperthreading can come close to doubling throughput. An excellent example of this is anandtech's 3DPM benchmark. Consider the i5-4690k and the i7-4770k. They have the same base clocks and turbos and the i7 has a slight cache advantage. HT is the only real difference:

3D Particle Movement: Single Threaded

i5 - 129.44
i7 - 129.37

3D Particle Movement: MultiThreaded

i5 - 417.8
i7 - 727.64

Single threaded they perform identically, but when they can go full multi-threaded you get +75% throughput JUST FROM HT. Moreover, if you read Ian's explanation of how this bench works, it lines up exactly with expectations. It simulates brownian motion that requires generation of random numbers at every time step that creates thread stalls. Lots of thread stalls = lots of benefit from HT.

 

[jhu]

Wish they'd add something more useful like Blender.