Linpack Challenge

[Idontcare]

4.8 GHz and 97 Gflops, that is impressive!

So we are looking at circa 1992 world-class supercomputer processing speed packed into a $1k single piece of silicon. Not bad.

http://en.wikipedia.org/wiki/Supercomputer#Timeline_of_supercomputers

 

[VirtualLarry]

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

I can't get to that link on Comcast, using IE 8 on XP SP3. Brings me to a comcast page, saying www.it.anandtech.com is not found.

How it is deciding to translate it.anandtech.com into www.it.anandtech.com is beyond me. Is this some new IE8 "feature", that's going to start blocking websites?

Edit: Typing "it.anandtech.com" and "http://it.anandtech.com/" lands me on the same Comcast "not found" page. WTF?

Edit: Tried Firefox 3.6.3, still no dice. I just get the Comcast not found page.

Edit: Just got home, trying that top link on Verizon DSL, using IE8 on win7 64-bit hp, using OpenDNS, no go. It also says www.it.anandtech.com is not found. Perhaps it's it.anandtech.com that is automatically redirecting to www.it.anandtech.com.

 

[Idontcare]

I can't get to that link on Comcast, using IE 8 on XP SP3. Brings me to a comcast page, saying www.it.anandtech.com is not found.

How it is deciding to translate it.anandtech.com into www.it.anandtech.com is beyond me. Is this some new IE8 "feature", that's going to start blocking websites?

Edit: Typing "it.anandtech.com" and "http://it.anandtech.com/" lands me on the same Comcast "not found" page. WTF?

Edit: Tried Firefox 3.6.3, still no dice. I just get the Comcast not found page.

That is your DNS fucking up, not your browser. Change your DNS to the new google ones.

http://code.google.com/speed/public-dns/

 

[lopri]

I think my DNS is messed up, too. Anyway, I believe the linked page is this:

http://www.anandtech.com/show/3470

So now ~100 GFlops are seemingly within the reach of a single-socket platform (albeit via overclocking). I'd like to put that in a perspective. Does anyone know how many GFlops an Opteron 146 (2.0 GHz K8, DDR-400) can do? How many would be needed to reach 100 GFlops?

 

[Unheard]

5.2 MFlops

Droid Eris @ 806mhz

 

[Idontcare]

Does anyone know how many GFlops an Opteron 146 (2.0 GHz K8, DDR-400) can do? How many would be needed to reach 100 GFlops?

According to this compilation table an overclocked Opteron 146 @ 2.6GHz turned in 4.3GFlops but that was with on 512MB of ram allocated...

So multiply that GFlop by 0.77 to undo the overclock (assuming linear scaling with GHz) and we have 3.3GFlops for 2GHz. Not sure how much higher that would become if one geeked out the ram.

So...100/3.3 = 30 Opteron 146's give or take your ram setup.

 

[lopri]

Incredible. So that's like a cluster of 8 x 4 socket Optern 846 setups. (think how much that would have cost!)

 

[Idontcare]

Pricing
The AMD Opteron processor Model 846 is priced at $3,199 in 1,000-unit quantities.

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543~74126,00.html

So >$100k just in processor costs alone.

Let alone the quad-socket mobo + ram costs.

Makes that $1k 980X look cheap

edit: just for further example of how ridiculous performance/$ is these days, in 2000 I spent ~$25k on parts to build a 24-node beowulf comprising 800MHz Athlons which in aggregate netted around 4 Gflops for the entire cluster.

10yrs later and for ~$3k you can build a nicely overclocked rig that gives you 25x more performance. Now that's crazy D:

 

[lopri]

5.2 MFlops

Droid Eris @ 806mhz

How did you get the results?

 

[Mir96TA]

Here is 100 %Stock

Here is 2 Gig with 4 Core

With MAx Mem and Quiet and Cool Enable

It is amazing, my HTPC has left my Intel in a DUST :twisted:

Man My Workhorse Computer Suck Arse compare to my
HTPC AMD.
I understand Intel is ruuning lesser cores and running with lower Cache and Ram.
Any rate here it is

I might come back with OC version of Intel

I guess Intel Core is much more productive then Intel at similiar speed
Check out this following LinX I am running AMD 1055T with 2 core only at 3 Gig.

Now with All Core Enable at 3 Gig it puts some serious numbers

 

[Juddog]

Does it make a difference with these Linpacks if Hyper Threading is running or not? I'm hitting 54.9 or so Gflops on my i7 w/ hyper threading enabled, but I need to find a good screenshot host to show it.

 

[Idontcare]

Does it make a difference with these Linpacks if Hyper Threading is running or not?

yes

http://forums.anandtech.com/showpost.php?p=29002347&postcount=82

 

[Juddog]

yes

http://forums.anandtech.com/showpost.php?p=29002347&postcount=82

Ah ok... but has that chart been updated since the newer 6 core CPU's have come out? Just curious since it seems like that 6 core Gulftown seemed to pull some pretty high Gflops.

 

[Idontcare]

I haven't seen anyone post thread-scaling data for the Intel hexcores.

 

[ehume]

Just to record this for later reference:

With HT on (didn't read that post until after the test had started):

Peak = 49.8579 GFlops in 20 runs, 64-bit OS, 8 threads
Intel i7 860 182x22=4004MHz
Mem=1820MHz
In the log, CPU frequency listed as 3.822 GHz (!)

Edit: HT off, 5 runs -> Peak = 58.2912 GFlops
. . . and my temps were 7-8c less.

Edit2: HT re-enabled, but mem multi pushed to 12x, so mem clock = 2184 MHz. In 20 runs of LinX, peak = 51.3141 GFlops.

 

[lopri]

Your scores seem just right.

50 GFlops for PII X4 @4.0 GHz
60 GFlops for i7 @4.0 GHz (no HT)
90 GFlos for i7-980 @4.0 GHz (no HT)

 

[Mir96TA]

I guess Intel Core is much more productive then Intel at similiar speed
Check out this following LinX I am running AMD 1055T with 2 core only at 3 Gig.

Now with All Core Enable at 3 Gig it puts some serious numbers

This is runining in Win7 64 Bit same CPU and Mem
It is amazing who things get to run better in Win 7
BTW I have Cool and Quiet and CIE state enable

 

[coffeejunkee]

I'm thinking it has more to do with 64-bit vs 32-bit than vista vs win7. Impressive numbers nonetheless.

Here's my little contribution:

i5 750 4GHz: 54.7586 GFlops peak

My cpu needs a healthy amount of extra juice to remain stable at 4GHz so things get a little toasty. Powerdraw is remarkable as well.

 

[Markfw]

Well, I am disappointed. dual E5520 Xeon's get the same 54 gflops that my single 920@3.9 did. I was hoping to beat it my at least 10-20%. And I had to take off HT, otherwise the system was horrible under load.

 

[Idontcare]

Well, I am disappointed. dual E5520 Xeon's get the same 54 gflops that my single 920@3.9 did. I was hoping to beat it my at least 10-20%. And I had to take off HT, otherwise the system was horrible under load.

Mark, it would be very cool if you could generate thread-scaling data on your dual-socket rig for both cases of no-HT and with HT.

Checkout this post: http://forums.anandtech.com/showpost.php?p=28985223&postcount=68

And this post: http://forums.anandtech.com/showpost.php?p=29002347&postcount=82

I'm curious to see how well the thread scaling deals with the dual-socket sitation.

(dual-socket rig, i7-920, AMD hex-core...you got all the sexy toys!)

 

[Tsavo]

Well, I am disappointed. dual E5520 Xeon's get the same 54 gflops that my single 920@3.9 did. I was hoping to beat it my at least 10-20%. And I had to take off HT, otherwise the system was horrible under load.

@stock cpu speed?

 

[Markfw]

@stock cpu speed?

ahhh, yes. I don't have the $600 mobo to overclock....

lowly 2366 mhz. (each)

 

[Markfw]

one thread 8.0352 gflops ( will update as I can)
two threads 15.65 gflops
3 threads 23.2 gflops
4 threads 30.46 gflops
5 threads 36.37 gflops

OK, I went to 6 threads... All of a sudden it went to 12 gflops !!!

WTF ??

 

[Idontcare]

Mark be sure and set your problem size such that your results will be comparable.

http://forums.anandtech.com/showpost.php?p=29002321&postcount=80

Different problem size equals different GFlops value even with same computer.

 

[Markfw]

It was the same every run, but I killed it. I will do more later, very weird....