Intel Core i7 4770K "Haswell" Benchmarks On Ubuntu Linux

[phoronix]

Phoronix: Intel Core i7 4770K "Haswell" Benchmarks On Ubuntu Linux

This past weekend I shared the first experiences of running Intel's new Haswell CPU on Linux. While Intel Haswell is a beast and brings many new features and innovations to the new Core CPUs succeeding Ivy Bridge, there were a few shortcomings with the initial Linux support. It still appears that the Core i7 4770K is still being finicky at times for both the processor and graphics, but in this article are the first benchmarks. Up today are benchmarks of the Intel Core i7 4770K when running Ubuntu 13.04 with the Linux 3.10 kernel.

http://www.phoronix.com/vr.php?view=18754

[frign]

That, considering the fact, Haswell hasn't been around long enough to be actually well-optimized for in the compilers. It might be only a new generation, but I am sure there is more potential.

[coats]

compiler settings?

For at least some of these benchmarks, compiler settings should make a substantial difference: it would be interesting to see the comparison between SSE4 vs AVX vs AVX2 enabled executables for Himeno on i7-4770K, for example. I would not make serious scientific use of a Haswell without compiling my programs for that architecture!

[mendieta]

I thought, as for CPU performance, it is expected to be only marginally better than IB. The real improvements are in graphics.

@Michael: it would be great to see a three way graphics comparison: Haswell vs Richland Open Source vs Richland Catalyst, now that the Richland APUs are also out.

My home desktop is an AMD phenom first gen rig with ATI entry level graphics. The ATI GC is no longer supported by Catalyst (thanks!), and the open source stack works great for multimedia and ok for 3D, but a lot of the humble bundle games will not run (and run instead on my son's $200 intel's chromebook, which makes you wonder).

I think Richland is a great value for an APU (mobo's are also cheaper), and I would definitely buy an a10-6800k if I were a Windows user, but as it stands, as a Linux user, my only sane option is Intel. I try to buy AMD when I can, because I want to see competition in the desktop CPU space, but hell, are they making it hard for me.

[coats]

I thought, as for CPU performance, it is expected to be only marginally better than IB. The real improvements are in graphics [snip...]

Except for scientific codes, where the AVX2 instruction set offers the potential of doubling the floating-point performance for codes doing LINPACK-style (array AX+B) computations. I would expect Himeno to be such a code; I know that the environmental modeling/forecasting codes I develop and use have those characteristics for a substantial chunk of their execution-profiles.

[chris200x9]

it's over intel is finished. Long live AMD!

[user82]

x264 Benchmark

There has been a bunch of new AVX2 additions to x264 on May 20th.: https://github.com/DarkShikari/x264-...commits/master

Could the test please be repeated with a newer version? I would be very interested in the results!

[smitty3268]

That was kind of weird to start the power efficiency section with a gpu test

After the entire article was all about cpu performance...

[mendieta]

Except for scientific codes, where the AVX2 instruction set offers the potential of doubling the floating-point performance for codes doing LINPACK-style (array AX+B) computations. I would expect Himeno to be such a code; I know that the environmental modeling/forecasting codes I develop and use have those characteristics for a substantial chunk of their execution-profiles.

Oh, great point. For what you are describing, it seemed like media encoding/decoding would benefit, so I googled a bit, and so seems the case:

http://software.intel.com/en-us/articles/intel-avx-new-frontiers-in-performance-improvements-and-energy-efficiency

Exciting stuff!

[FourDMusic]

Scaling Governor

The PTS table on page 1 shows different scaling governors being used for the different processors (intel_pstate powersave and acpi-freq ondemand). From the documentation, the "powersave" mode of the intel_pstate "sets the CPU statically to the lowest frequency within the borders of scaling_min_freq and scaling_max_freq" (governors.txt).

Unless scaling_min_freq to the maximum frequency, then there is going to be greater performance for those processors.

However, the benchmarks don't look too far off, so I don't know what's up. Thought I would point this out.