Clear Linux Squeezes More Performance Out Of Intel's Core i9 7960X, 7980XE

[waxhead]

I’m not so sure about your point. “Squeezing” refers to getting everything out of it that is (already) in it. Not to altering it in order to produce even more. So to me, it seems to fit.

Ok I admit that I am arguing mostly for the sake of arguing here, but isn't this what these forums are all about anyway?! - A processor is not comparable to a sponge. It is a hard uncompressable object that (at least historically) have no lower or upper speed limit. The CPU will require a certain amount of clock cycles to process each and every instruction. Some instructions require more clock cycles than others and the clock cycle rate is determined by an external crystal. Ergo it is therefore impossible to extract more performance of a CPU itself even if you change the external clock (the CPU processes instructions at the same rate e.g. same number of cycles regardless of the clock cycle speed). It is however possible to write smarter and more optimized programs that accomplish the same result using simpler or less number of instructions and therefore using less clock cycles as well. This have nothing to do with the CPU's ability to process instructions at a certain rate tough.

[waxhead]

I’m not so sure about your point. “Squeezing” refers to getting everything out of it that is (already) in it. Not to altering it in order to produce even more. So to me, it seems to fit.

The point being that the title should be Clean Linux runs a tad faster on Intel's Core i9 7960X, 7980XE - the way it's written now makes it appear that software can "tune" the CPU to run faster. It is as I said nitpicking since I assume most here will understand the difference. That being said these days you can tune the frequency of a CPU in software and with careful programming you can even get certain threads to run faster (turbo). The point is a CPU can only run at a certain speed at max performance and there is nothing software can do with that. Software can only be written more efficient to make the best out of the CPU as you have written yourself.

[lxuser81]

Yet again i must cry about blender test, The BMW test You are using bmw27_gpu.blend file, the tile size is 256x256 so if You got 36 thread cpu You still are only using 8 threads (the render resolution is half fullHD) ...
<Value>../benchmark/bmw27/bmw27_gpu.blend</Value> from test-definition.xml file

So there must be "fix" if cpu is used for rendering change bmw27_cpu.blend file or set
scene.render.tile_x = 32 scene.render.tile_y = 32 Other way the blender cpu test results are not usable, that way high clock 4core/8thread cpu can get wery close to 18c/32threads monster (like it happened if You tested i9 and TR).

[chithanh]

Well, "no so legal" as in forbidden by the C99 standard. I don't think there is a police who you can call against this.

[nils_]

Clear Linux tunes the compiler with all sorts of legal (and sometimes not so legal) optimizations to squeeze more performance of the software running on the hardware.

Illegal compiler flags? Call the Police!

[chithanh]

Clear Linux tunes the compiler with all sorts of legal (and sometimes not so legal) optimizations to squeeze more performance of the software running on the hardware.

[indepe]

Nitpicking but I feel this is a bit wrong "Clear Linux Squeezes More Performance Out Of Intel's Core i9 7960X, 7980XE"

Software can't squeeze more performance out of a processor - this means altering it. What software can do is run more efficiently

I’m not so sure about your point. “Squeezing” refers to getting everything out of it that is (already) in it. Not to altering it in order to produce even more. So to me, it seems to fit.

[waxhead]

Nitpicking but I feel this is a bit wrong "Clear Linux Squeezes More Performance Out Of Intel's Core i9 7960X, 7980XE"

Software can't squeeze more performance out of a processor - this means altering it. What software can do is run more efficiently

[mlau]

I used the standard Blender offered by Gentoo and the bmw27_cpu file, changed no settings in blender, just ran it. it does max out all 16 cores though, and the result is always the same.

[chuckula]

The Blender result seems way off: My 1800x CPU-renders the BMW27 scene in 04:41 (281 seconds), I'd expect any of the tested CPUs to do much better than that.

The Blender results are very weird since while Blender doesn't scale perfectly with core count it does generally scale much better than what Phoronix shows. There may also be render settings like the number of samples to use that are different in Phoronix's test compared to a vanilla test.

One easy way to diagnose the issue would be to look at CPU utilization during the rendering process. If only a few cores are being used heavily it could point to a configuration problem.