Linpack Challenge

[luv2increase]

One of the most major problems with using LinX to determine GFlops is that it is impossible to determine the "true" GFlop value if HT is enabled.

 

[Idontcare]

One of the most major problems with using LinX to determine GFlops is that it is impossible to determine the "true" GFlop value if HT is enabled.

What do you mean? What is a "true" GFlop?

 

[Rubycon]

What do you mean? What is a "true" GFlop?

The GFLOP number is higher with HT OFF. I noticed some people getting into the 60's with slower chips than mine and was like WTF. Then someone pointed out the number of processes at 4 instead of 8. So I turned off HT and 64 GFLOPS!

Of course I leave HT ON. Encoding takes a hit with it off. My PC is in an area where the average temp is 65F, I have a Megahalem Shadow with a pair of Gentle Typhoon 1800RPM fans in push pull. Life is good.

 

[lopri]

38.1 Gflops
Athlon II X4 @ 3.2ghz
2GB OCZ Gold @ 1660ish 8-8-8-30 1.7v

i was at 32gflops at stock clocks (2.6ghz) with 4gb of ram, but i have to RMA it due to a faulty stick.

29gflops @ stock clocks/2gb ram...so i should be able to break 40 Gflops when i get my RAM back next week.

Also +200 MHz on PII's NB will yield extra 0.5~0.7 GFlops at the same core frequency. You have to keep an eye on heat, though.

 

[lopri]

The GFLOP number is higher with HT OFF. I noticed some people getting into the 60's with slower chips than mine and was like WTF. Then someone pointed out the number of processes at 4 instead of 8. So I turned off HT and 64 GFLOPS!

Found the AT article on this.

http://it.anandtech.com/weblog/showpost.aspx?i=528

 

[lopri]

Come on, people. I know everyone here overclocks. Give us your Conroe scores and K8 scores.

 

[JFAMD]

I can't believe they upgraded 37k+ CPUs that along would take months. good thing AMD designed the 6cores as a drop in replacement for the socket, no need to upgrade cooling solutions.

but it says: "AMD says it will offer the first eight-core and 12-core x86 processors for high performance computers early next year." I have never heard of this been said anywhere. If it can be done 8/12 build on 45nm would require lot more cooling.

It took on average, ~5min per blade. Each blade has 4 compute nodes with 2 processors each, so 8 total processors.

So that is ~4265 total blades. At 5 minutes each, that is ~385 total hours.

The 8 and 12-core processors will be in similar power and thermal as the 6-core today.

 

[aigomorla]

Ima smash your contest Lopri.

CPU: Gulftown
Core Frequency: 4380
NB/Uncore Frequency: umm i dont remember.. 3400ish i think...
Memory Frequency: DDR3-1600 (8-8-8-20-2T)
Peak Performance: 77.45 GFlops (can i haz cookie for being #1?)

Guys who are running into errors... ease up on the ram.

Run it with like 1024megs for benchmarks...

However running it with more ran usually gets higher scores... however good luck matching that.

 

[n7]

We've had this discussion before...running a minuscule amount of RAM during LinX/IBT testing pretty much defeats the point of stresstesting with LinPack...

You can have your cookie when you run max RAM stably & post screenshot for me :twisted:

Till then, no cookie.

The numbers are nice though
I assume HT off, right?

 

[Rubycon]

Guys who are running into errors... ease up on the ram.

Run it with like 1024megs for benchmarks...

What's the point if it's not stable?

If I cannot get 25 passes with max ram it serves the same purpose as titties on a bull. Sorry.

If you err out NEVER drop the ram and repeat the test, FIX your unstable OC!

That is like bragging you have a 30,000 hp locomotive but no track to run it on!

 

[nyker96]

on a newly built x4 620@3,25//8GB DDR2@800+ it shows 39.2 GFplops.

 

[E4300]

- WXP Pro SP3 32-bit
- PII X3 710 unlocked to PII X4 @ 3.60GHz/1.42V BIOS with 46C peak in 80F room
- 2GB DDR3 ganged mode @ 1108MHz/7-6-6-15-2T (1760MB tested)
- HT and NB frequencies: 2216MHz

36.48 GFlops peak

The performance of this overclocked/unlocked PII X3 710 @ 3.6GHz should be identical to the OP's X4 9XX @ 3.6GHz. Ganged/unganged mode does not materially affect GFlops. Note that GFlops went from 36.48 to 45.33 when the amount of tested memory increases from 1760MB to 6826MB.

 

[Hyperlite]

on a newly built x4 620@3,25//8GB DDR2@800+ it shows 39.2 GFplops.

That is interesting, and indicative of the performance advantage DDR3 gives you. I hit 38.1 Gflops with 2GB DDR3 1599 @ 3.2ghz CPU, 2ghz HT. I'm surprised to see your HT so high, that's awesome...I have to RMA my ram and lost 3Gflops when i pulled the bad stick out... Were you testing with All your ram used? i was usually benching with <1GB, as it's hard to get more than that free with only a 2gb stick in there.

 

[lopri]

@Aigo:

Hehe. 77.45 GFlops. That's crazy.

No cookies for you, yet, though. ^_^ Besides what n7 has pointed out, there was this as well.

2) Dual-core CPUs - at least 10 loops, Quad-core CPUs - at least 20 loops.

I should have added "at least 30 loops for hexa-core CPUs" :biggrin:

 

[aigomorla]

well i can do that but with all the memory thats really not fair for me.

Reason is im first at a disadvantage running 12GB of DDR3 vs 6GB of DDR3.
Also, like the others i see a problem when i run maximum memory.

I can do 30 passes if you like, however you might see numbers in the 80 GFlops.
That last screenie i showed u was a quick run without a fresh reboot.

 

[Idontcare]

@Aigo:

Hehe. 77.45 GFlops. That's crazy.

No cookies for you, yet, though. ^_^ Besides what n7 has pointed out, there was this as well.


I should have added "at least 30 loops for hexa-core CPUs" :biggrin:

Would be awesome if you could get that guy on XS who has the 2S MagnyCours to run this app and report his results. That would be a real coup d'&#233;tat for the thread.

Edit: Didn't know I was running an older version of LinX. I don't think it'd make a difference but I guess I'll double check the channel log of the latest version.

Is the link in the OP for the most recent version? A quick google on LinX did not lead me to any seemingly singular authoritative source on LinX and its versions.

 

[aigomorla]

i just use agentgod's burn test.

Its the same thing.

Would be awesome if you could get that guy on XS who has the 2S MagnyCours to run this app and report his results. That would be a real coup d'état for the thread.

I already know the result IDC.
Let me just tell you this.
Nadeshiko @ 4ghz beat a dual gainestown 8 core system @ 3.3ghz
Nadeshiko @ 4.3 also equals, but i get slightly more points over my friends dual gulftown system with 12 cores @ 2.66.

AMD's 2S MagnyCours gets raped by a 3.07 ghz dual gainestown system.

AMD, my wishes are on a 400 dollar Hexcore Please, and you still might win in value, otherwise if they price it near gulftown, prepare for slaughter.

 

[zsdersw]

AMD's 2S MagnyCours gets raped by a 3.07 ghz dual gainestown system.

Uh oh....

 

[aigomorla]

Uh oh....

it does.. :\

You guys are gonna scream 12 Core system... but the chips each dont = an i7 itself.
So you got 12 working threads on the AMD, vs 16 working threads on the gainestown with HT ON.
And the virtual cores on the Intel are almost the same speed as the AMD's physical cores, so its a land slide loss. :\

So a dual i7 system = gainestown system, will beat it.
A dual i9 system will slaughter it.

Ives seen enough benchmarks on it. AMD's only win in this market is to release the cheapest hexcore possible.

And 1 up intel that way. Because as in raw computation power, the AMD's Hexcore is not near par to Intel's.

Also Gulftown is not intel's most powerful cpu. They have a chip called the beckinton. 8 core monster which only fits in a special board that takes 4 cpu's.

32 physical cores... OUCH...

 

[n7]

well i can do that but with all the memory thats really not fair for me.

Reason is im first at a disadvantage running 12GB of DDR3 vs 6GB of DDR3.
Also, like the others i see a problem when i run maximum memory.

I can do 30 passes if you like, however you might see numbers in the 80 GFlops.
That last screenie i showed u was a quick run without a fresh reboot.

It's not fair?

Why, because it shows your OC isn't stable?

You don't see me bitching about it not being fair i have to stress 6+ GB vs. people doing only 3 GB (because they only have 4 GB).

The "unfairness" is actually you, since you have 12 GB, which will give you higher scores than someone with 6 GB could ever get.

I know, it forces you to lower your [not really stable] OC to something less impressive, but unstable OCs don't impress me, sorry.

I can run 4.4 GHz w/ my system on pretty average air cooling, big whoop...it's not remotely stable.

As i said, cookie for you when you test with all your RAM. :hmm:

 

[Phynaz]

I have a couple of nice boxes to run this on...Anybody have it compiled for POWER running AIX 5.x?

Edit - well at least one, now that I think of it, the big boys are all running production apps, that unfortunately means no playing on them.

 

[aigomorla]

It's not fair?

Why, because it shows your OC isn't stable?

You don't see me bitching about it not being fair i have to stress 6+ GB vs. people doing only 3 GB (because they only have 4 GB).

The "unfairness" is actually you, since you have 12 GB, which will give you higher scores than someone with 6 GB could ever get.

I know, it forces you to lower your [not really stable] OC to something less impressive, but unstable OCs don't impress me, sorry.

I can run 4.4 GHz w/ my system on pretty average air cooling, big whoop...it's not remotely stable.

As i said, cookie for you when you test with all your RAM. :hmm:

its not fair because all my slots is populated.
Thats what im stating at n7.

If you want i'll pull out 3 sticks, and go 5ghz and i can run it @ 5ghz for 30 passe with max ram when only using 3 sticks.

Ive done 20 passes with max ram on only 3 sticks. 6 sticks and i have problems past 75&#37; ram.

If all your ram slots are populated, you have a disadvantage. Unfortunately for me, i require all that ram because i have 0 HD caching.

 

[nyker96]

That is interesting, and indicative of the performance advantage DDR3 gives you. I hit 38.1 Gflops with 2GB DDR3 1599 @ 3.2ghz CPU, 2ghz HT. I'm surprised to see your HT so high, that's awesome...I have to RMA my ram and lost 3Gflops when i pulled the bad stick out... Were you testing with All your ram used? i was usually benching with <1GB, as it's hard to get more than that free with only a 2gb stick in there.

yes I set to use all my ram. I wouldn't think DDR3s can give you such an edge in computationally intensive task such as Linpack, but if you hit 38.1 then it must be. you are running at DDR1600 vs my DDR800 that's massive bandwidth advantage. Some poster before mentioned that if you use more Ram, the score goes up, try it with more than 2GB I wonder if you can hit it much higher. I'm assume you are using PII x4, maybe the 6mb cache helped too.

 

[lopri]

@Aigo: I thought Magny Cours will have 12 physical cores w/ quad-channel DDR3? 2S Magny Cours will have 24 cores total. 2S Gainestown platform has 8 physical cores and 16 logical cores. I think you were thinking of dual Istanbul (12 cores) which indeed lose flat out to Nehalem-EP (8 physical cores / 16 logical cores). Nehalem-EX will be a completely different beast, of course.

I could be wrong on the Magny Cours.. I haven't really been a big fan of many cores.

P.S. There is a timely article by Johan.

http://it.anandtech.com/IT/showdoc.aspx?i=3681

 

[lopri]

on a newly built x4 620@3,25//8GB DDR2@800+ it shows 39.2 GFplops.

Very good result you've got there, nyker96. Especially with DDR2-800. Where is your CPU-NB frequency at?