Linpack Challenge

[Idontcare]

linX does not do bulk processesing across all threads and adding to the total gflop number.

it loads up as many cores as you want it to load up, an then averages the gflops across all your working cores.

The HT cores counting for half your total threads will down the average on linX.

If you disable HT, you will get a better number on LinX because ur not averaging the slower cores into your total gflop count.

I'm still not following.

The application under discussion for this topic very much does have explicit thread-count spawn control, which is the tool used to generate the data presented in the graph from the post you quoted.

So I'm still just scratching my head here.

Yes HT encumbers the gflops value, I thought that was well-addressed by the numerous posts in this thread that have already stated as much. The post with the graph you quoted merely adds data to support those statements.

So I'm gonna have to fall back to my earlier question - what am I missing?

 

[Dufus]

its because when linX does work on the HT cores, the HT is only half to 1/3rd the actual speed of a physical core.

aigomorla there are not physical cores and HT cores only physical cores. In the case of hyperthreading each of those physical cores has 2 threads which are seen by the OS as 2 logical cores. Both threads are equal but have to share the resources of one physical core. With Linpack that leads to contention and degrade in performance due to the way it utilizes cores resources.

 

[Lorne]

I would also assume that with HT the general memory bandwidth begins saturation to a point that a available core might starve.

I belive there was the same conclusion with a recent testing between some Xeons in order of single CPU and dual using 2 and 4 core Xeons w/wo HT and a I7 equivelent running a bunch of apps and CS5 benching.
Also seeing more users complain about upgrading from an X4 to an X6 on some DDR2 platforms about lower benchmarks on memory intensive apps.

A test with HT by adding available cores to LinX one at a time might be something to look at.

 

[WildW]

This thread is too entertaining to let it die. Time for new scores

Athlon 64 X2 4200 (skt 939), 2GB DDR @ 400MHz, WinXP

@stock (2.2GHz) --> 5.something GFLOPS - can't actually remember exactly now.

Overclocked (240x11 = 2640MHz) --> 6.48 GFLOPS

I seem to remember that LINX gives much higher scores the more memory you give it, so I imagine that 2GB is holding me back here. Still, a score this low reminds me just how far we've come in 5 years.

 

[Mir96TA]

Where I can D/L it from ?

 

[WildW]

Where I can D/L it from ?

http://www.xtremesystems.org/forums/showthread.php?t=201670

 

[Mir96TA]

Dead Links

 

[PreferLinux]

My i5 2500K at stock gets ~47-48 GFLOPS under Win Vista and 80.1 GFLOPS under Linux (runlevel 3, i.e. no graphics or mouse: it gets ~79-80 in KDE).

 

[Makaveli]

This is my score HT on

 

[Mir96TA]

http://www.xtremesystems.org/forums/showthread.php?t=201670

Dead Links

It works now

 

[smakme7757]

I got some stupid numbers when i had the 2600K overclocked with SP1. I don't have a screen shot, but it was 120ish Gflops. Insane stuff. AVX really boosts the Flops!

 

[FordGT]

My Q6600@3Ghz gets 41GFlops using a problem size of 20000. That seems about right, no?

 

[Dadofamunky]

My Q6600@3Ghz gets 41GFlops using a problem size of 20000. That seems about right, no?

Not bad at all.

 

[Mir96TA]

Here is my dinky AMD P320 base Laptop

 

[Eug]

My iPhone 4 usually peaks at 38-41, with a few times at 45, but once peaked at just over 49.

That's Megaflops though. (Problem size 200, 50 runs)

 

[The2ff]

AMD 1055T 6 core
8gigs Gskill PC800 DDR2
Gigabyte 785G AM2 board
14x multi
3724MHz core
2402Mhz HT
444MHz Mem
5-5-5-15 timings
HT ref Clock 266

61.4219 GFlops

 

[Bryf50]

AVX

 

[PreferLinux]

My i5 2500K at stock gets ~47-48 GFLOPS under Win Vista and 80.1 GFLOPS under Linux (runlevel 3, i.e. no graphics or mouse: it gets ~79-80 in KDE).

UPDATE: That was under Fedora 12 with the kernel upgraded to 2.6.31. Now, under openSuse 11.4 (kernel version 2.6.37), I get 94.769 GFLOPS (still runlevel 3). Still stock. I don't know what gave the increase, but it is quite a significant improvement! The anly reason I can come up with is either I'm now running my RAM at 1600 MHz (possibly) instead of 1066, or more likely because of reduced overhead from the newer kernel or different software.

 

[Bryf50]

UPDATE: That was under Fedora 12 with the kernel upgraded to 2.6.31. Now, under openSuse 11.4 (kernel version 2.6.37), I get 94.769 GFLOPS (still runlevel 3). Still stock. I don't know what gave the increase, but it is quite a significant improvement! The anly reason I can come up with is either I'm now running my RAM at 1600 MHz (possibly) instead of 1066, or more likely because of reduced overhead from the newer kernel or different software.

The newer version of linux probably supports AVX. Look at my post above.

 

[PreferLinux]

The newer version of linux probably supports AVX. Look at my post above.

The other difference I missed is the old one was 32-bit with PAE. The new one is 64-bit.

The older version supported AVX too: on Windows I get ~48 (no AVX) as I said before, and the older Linux got 80.1 (which fits with the thread on here comparing with and without AVX). And yes, I see you got 100.8 with AVX @ 4.4 GHz (and 64-bit, so that isn't the cause of the difference for me), but I'm getting 94.769 with AVX @ 3.3 GHz (or possibly 3.4 from TurboBoost).

 

[lopri]

CPU: Core i5-2500K
Core Frequency: 4800 MHz (1.45V)
NB/Uncore Frequency: No clue
Memory Frequency: DDR3-1600 (8-8-8-24)
Peak Performance: 132.92 GFlops

 

[Eddie313]

CPU: Core i5-2500K
Core Frequency: 4800 MHz (1.45V)
NB/Uncore Frequency: No clue
Memory Frequency: DDR3-1600 (8-8-8-24)
Peak Performance: 132.92 GFlops

Really 4.8ghz oc that's crazy very nice over clock there.

 

[lopri]

I was more surprised by the IMC handling 4x4GB @1T. It runs 5GHz fine in Prime95 but I stopped Linpack as the temp shot up to 90C at that clock. At 4.8GHz it's ~75C while running Linpack. (~65C Prime95)

Edit: OK so this score looks to be "inflated" due to the use of AVX. (thought it was unrealistically high) How do I run it w/ SSE to know the comparable score?

Edit 2: I retract the statement on IMC. 1090T seems to handle 4x4GB @1T fine as well. (with better timings) I guess I've gotten a good set of sticks.

 

[lopri]

Bump. Anyone knows whether/how this chip (2500K) runs Linpack using SSE2? I tried old binaries but It doesn't run properly. (~25% CPU usage, uneven among cores)

 

[PreferLinux]

Easy: use Windows 7 without SP1 or earlier! Or do it in Linux with and old kernel, but I'm not sure what version.