Intel 4.5GHz & 5GHz LinX Stable Club

[zalbard]

Updated! Very nice job, guys!
@ pet168: what were your ambients? That's very impressive for air cooling...

[pet168]

Updated! Very nice job, guys!
@ pet168: what were your ambients? That's very impressive for air cooling...

my ambient temp is around 20 ~ 22c

[LennyRhys]

Submission removed because LinX was not up to date. Resubmission to follow.

[zoson]

haha 49gflops. weak. You're not really stressing your CPU with that low performance.
That's why your temps are so low, also.

Edit: with HT on your temps should be ~10C HIGHER than with HT off.

[kitfit1]

Knock knock.........................................can i join the over 100 Gflop club ?

kitifit1 | Core i7 990x [6C 6T] @ 5009Mhz | 1.611 | EVGA Classified E760 | Single Stage | 3044B119

Attachment 117407

[LennyRhys]

haha 49gflops. weak.

Amazing contribution to the thread

You're not really stressing your CPU with that low performance.

yeah, that's why zalbard decided on LinX, because it doesn't really stress the CPU

Edit: with HT on your temps should be ~10C HIGHER than with HT off.

Oh should they? OK, I'll see if I can somehow alter the laws of physics and invert the temperature differential for you.

[donmarkoni]

@LennyRhys:

Zoson can be very blunt in his comments, but he is absolutely right.

Check others' results to see what your performance and heat output should be at those settings.

[LennyRhys]

My performance is more-or-less the same as angelreaper (to name but one) who also used the 25 multi for his i7 970 4.5GHz run. What's more, I have low memory frequency so the throughput just isn't going to be as high. But this thread has nothing to do with performance... it's a frequency club.

And what zoson says is not true at all - his remark that LinX "doesn't really stress" a CPU is about the stupidest thing a person could say in a thread like this - LinX is a monster of a test for any CPU, HT or no HT.

As far as heat output is concerned, this is an interesting one. For starters, nobody can point at a result and say "your temps should be different." That's crass - my temps are what they are; I didn't choose them or make them myself, so how can I be expected to change them?! LOL really... I did read around a bit and found that the linpack library stresses the CPU very differently from Prime95. The Prime95 mindset is a very logical one: more threads equals more heat; however LinX doesn't appear to work like that.

And it's worth mentioning that other people have used very high bclk for their runs, and that will make a chip much hotter due to the necessary increas in VTT.

[newhit]

Before it all kicks off.............

I've got a 980x Gulftown submission.....Phase cooling.
batch 3003B287. Mbd EVGA Classified E760.
Thanx to Kitfit for helping me to squeeze a bit more out of it.

Attachment 117420

[zoson]

My performance is more-or-less the same as angelreaper (to name but one) who also used the 25 multi for his i7 970 4.5GHz run. What's more, I have low memory frequency so the throughput just isn't going to be as high. But this thread has nothing to do with performance... it's a frequency club.

And what zoson says is not true at all - his remark that LinX "doesn't really stress" a CPU is about the stupidest thing a person could say in a thread like this - LinX is a monster of a test for any CPU, HT or no HT.

As far as heat output is concerned, this is an interesting one. For starters, nobody can point at a result and say "your temps should be different." That's crass - my temps are what they are; I didn't choose them or make them myself, so how can I be expected to change them?! LOL really... I did read around a bit and found that the linpack library stresses the CPU very differently from Prime95. The Prime95 mindset is a very logical one: more threads equals more heat; however LinX doesn't appear to work like that.

And it's worth mentioning that other people have used very high bclk for their runs, and that will make a chip much hotter due to the necessary increas in VTT.

What I was saying is there is something wrong with YOUR machine (and angelreaper's too, along with anyone else who has low performance for their frequency) and you are not actually stressing your cpu fully with YOUR linx run. Done right linx stresses the hardest. Done wrong you see low GFlops like you and angelreaper.

It's a known fact by anyone who is anyone that HT on causes your temps to be ~10C hotter than HT off. If your HT on temps are LOWER than your HT off temps, there's something wrong. With HT on you stress your cores more, not less. More stress = more heat.

Memory frequency does make a difference, but I've posted many shots showing that the actual difference in GFlops between 1600mhz and 2000mhz is small. At most you'll see a 5GFlop difference, with the average being 2-3GFlops.

Yes, this is a club for frequency stability. You're not stable if you're not performing right. Sure if you're running at 4.7ghz and you're only using 50% of your clock cycles, you're going to appear to be stable. Your cpu won't heat up as much as it should, and as a result the internal resistance of your transistors doesn't go up as much as it would if every cycle were used. If you fixed what's causing your low GFlops, you'd probably crash or overheat.

[LennyRhys]

There's nothing wrong with my machine - just realised that I don't have the updated linpack libraries, so I apologise for the confusion.

If you fixed what's causing your low GFlops, you'd probably crash or overheat.

Since updating the linpack executables I did a quick run at 4.5GHz (180x25) and I get 73 gflops consistently; it does run hotter but it's stable.

I'll see how far I can push on air and resubmit later this evening.

Edit: Here is 4.5GHz. Air temp is nearly 5C higher than yesterday which made a significant difference to load temps, but my trusty Delta still kept them under 80C. If air temp gets lower I'll give 4.6GHz a bash.

Cire i7 970 batch #3012B113 @ 4500MHz | HT on | 1.384v | Asus P6X58D-E | Air cooled with Copper TRUE @ 14C ambient

[kazukun]

■CPU ：core i7 2600k ：L102B549
■M/B ：GA-Z68X-UD7-B3
■BIOS ：F7
■BCLK ：100
■CPU倍率 ：50
■Vcore ：BIOS1.55V(Load1.548～1.608V)
■LLC：7
■off set　0
■QPI/VTT V(VCCIO) ：1.22V
■Vdimm ：1.62V
■MEM ：CMT8GX3M2A1866C9　4GB*2
■FSB：1:7
■DRAM Frequency ：DDR3-1872MHｚ
■DRAM Timing ：9-10-9-27 1T
■HT：ON
■RoomTemp ：26℃
■Core（RearTemp3.67）：Max72-83-87-79(Min30-33-34-33)

LinX0.6.4Load
http://img98.imageshack.us/img98/8213/61393386.jpg
End
http://img6.imageshack.us/img6/7712/65230832.jpg

[donmarkoni]

@kazukun:

Why so high VccIO? It is more likely to kill a chip then high vCore.

[acebmxer]

I see alot of people arguing about Gflops vs performance at a given speed/cpu. Have you guys read this. http://www.overclock.net/intel-cpus/...test-linx.html I have done my testing and with my cpu it has been true.

Also have notice while overclocking and testing. If a system is unstable, I have had it pass 15 runs of linX with LOW gflops then what it should be. Stopped the test and rerun it and had it crash after second round. Readjust voltages and this time would pass with the correct gflops and pass.

[donmarkoni]

... http://www.overclock.net/intel-cpus/...test-linx.html...

Yeah, it's a good guide, have been mentioned before. Nonetheless, it's good to remind everybody of it.

[zoson]

Zalbard needs to read it, i think. Otherwise all these super low GFlop people wouldn't be on the list as stable overclocks.

[LennyRhys]

I don't think it's fair to say that every LinX run with low gflops is indicative of instability; many of the hexcore CPU users (like me) have run the older version of LinX, either because that's all there was at the time or because they were unaware of the updates, so although the gflops are low the system isn't necessarily unstable. I updated the library and ran LinX at 4.5GHz without changing voltages from my previous 4.5GHz run and passed no problem, albeit with higher temps.

At the end of the day, zalbard decided on 20 passes with LinX as a measure of stability, not gflops or anything else. I think it should be for each person to decide themselves what they submit; 20 passes is 20 passes, stable or not.

[newhit]

Got another, if it's OK?

Attachment 117456

[LennyRhys]

Update:

i7 970 #3012B113
4625MHz @ 1.44v (1.275v QPI)
6C 6T
Air Cooling - TRUE Copper & Delta AFC1212DE, 13C air temp
Asus P6X58D-E



And 5GHz in windows for the heck of it!

http://valid.canardpc.com/show_oc.php?id=1901503

[newhit]

Update:

i7 970 #3012B113
4625MHz @ 1.44v (1.275v QPI)
6C 6T
Air Cooling - TRUE Copper & Delta AFC1212DE, 13C air temp
Asus P6X58D-E

And 5GHz in windows for the heck of it!

http://valid.canardpc.com/show_oc.php?id=1901503

Looks very good to me m8

I had low girlyplops on my 980x until I upgraded to AVX.

[kitfit1]

Ok, just got my 100 Gflops clock up a bit.

kitifit1 | Core i7 990x [6C 6T] @ 5100Mhz | 1.682 | EVGA Classified E760 | Single Stage | 3044B119

Attachment 117497

[CryptiK]

Hang on, zoson hit 100 GFLOPS at ~4.6GHz on the 970 (Link), and this is 5.1GHz and its at 110 GFLOPS, seems like some performance is lacking for some reason or is it just not scaling up in that CPU speed range?

Really impressive OC though, you are thrashing that 990x

[LennyRhys]

Zoson got 100 gflops because of the relatively high bclk & memory frequency - compare it to my run, and the lower bclk/memory costs about 3-4 gflops, even though my CPU clock was higher.

If kitfit1 had 5GHz on a 23x CPU multi, gflops would be through the roof

[donmarkoni]

Hang on, zoson hit 100 GFLOPS at ~4.6GHz on the 970 (Link), and this is 5.1GHz and its at 110 GFLOPS, seems like some performance is lacking for some reason or is it just not scaling up in that CPU speed range?

You are right, it's not scaling very good over 4.5-4.6GHz. I don't know why, maybe some internal throttling mechanism, OVP, OCP, or something.
But we must remember that we are giving those chips up to 20% more voltage then absolute max recommended.
Here is the scaling from my experience 4560MHz/100GFLOPS, 4800MHz/103GFLOPS, 4900MHz/105GFLOPS, 5000MHz/107GFLOPS (an you won't see this picture, cause I never managed to pass all 20 loops )

If kitfit1 had 5GHz on a 23x CPU multi, gflops would be through the roof

No they would not, they would be the same. There are many topics about running higher BCLK with lower multis or lower BCLK with higher multis.
Common conclusion is that there is no difference if clocks are the same (or very similar).

[kitfit1]

You are right, it's not scaling very good over 4.5-4.6GHz. I don't know why, maybe some internal throttling mechanism, OVP, OCP, or something.
But we must remember that we are giving those chips up to 20% more voltage then absolute max recommended.
Here is the scaling from my experience 4560MHz/100GFLOPS, 4800MHz/103GFLOPS, 4900MHz/105GFLOPS, 5000MHz/107GFLOPS (an you won't see this picture, cause I never managed to pass all 20 loops )
No they would not, they would be the same. There are many topics about running higher BCLK with lower multis or lower BCLK with higher multis.
Common conclusion is that there is no difference if clocks are the same (or very similar).

Yep......................just about what i found as well. So far i've had a 920,975 and the current 990x. All of them scale the same as far as Gflops are concerned.
The only way to push for higher Gflops on the 5.1ghz clock i posted would be to raise the the NB multi. But like a lot of i7's, my 990x doesn't like anything much above 4000mhz with the cpu clock much over 4.9ghz.
Obviously i could post a higher Gflop value at a lower cpu clock, but as the clock was done just to get higher than 100 gflops and over 5ghz it would be pointless.