Intel 4.5GHz & 5GHz LinX Stable Club

[tool_462]

Pretty nice CPU with 1,32v and 5gig...

Phase for CPU's is like beer goggles, can make an average chip look good

[alex_tpc]

Phase for CPU's is like beer goggles, can make an average chip look good

Lol! This one's on water: 3770k 5.05Ghz 1.315v on bios

alex_tpc | Core i7 3770K [4C 8T] @ 5049.9MHz | 1.312 | ASUS Maximus IV Extreme | Water | L215B620

[zoson]

Am I the only one who thinks 119GFlops for 5GHz Ivy is woefully low?

[C-N]

He has HT on maybe it would look better off

[fuzz3l]

Phase for CPU's is like beer goggles, can make an average chip look good

Oh, i didn't see the "SS"

Anyway still a nice cpu, do you test it under air or h2o and let's say 4500?

[zoson]

He has HT on maybe it would look better off

Nope.
http://www.xtremesystems.org/forums/...=1#post4802911

There's a 5ghz Sandy beating him by >10GFlops at a lower clockspeed. Something isn't right with that submission. Maybe outdated linpack binaries.

[C-N]

Ok next go lol... His mem is also slack & 2T .......

[p3gaz_001]

Oh, i didn't see the "SS"

Anyway still a nice cpu, do you test it under air or h2o and let's say 4500?

yes i can try 4.5ghz on water.

[alex_tpc]

Am I the only one who thinks 119GFlops for 5GHz Ivy is woefully low?

I thought that as well when I posted it but there was not much 3770k 4c8t 5Ghz benches to compare with. I did look at the 2600k 4c8t benches from page 1 (see comments below) and p3gaz's 3770k 4c4t result (I get similar results when I do the same).

He has HT on maybe it would look better off

Yes, it does get better GFlops with HT Off. I mimicked the settings from p3gaz's 3770k 5Ghz 4t4c results and got very similar gflops (stopped at 3 iterations as the official 21-iteration submission is the one above):


Seems I'm on par at least with 4c4t 5Ghz.

Nope.
http://www.xtremesystems.org/forums/...=1#post4802911

There's a 5ghz Sandy beating him by >10GFlops at a lower clockspeed. Something isn't right with that submission. Maybe outdated linpack binaries.

I used the latest linx(0.64)_10.3.10.017 binaries. deachus' 135GFlops peak for 2600K 4c8t 5ghz result is something special, beating the other 2600k 4c8t results on page 1, even those with 5.1-5.3Ghz. It also beat by 20GFlops Dumo's 5.004Ghz results which had close but tighter tRP timing. I took out my hypers from storage and mimicked the settings (5Ghz 2133 c7-9-7-20) from deachus on my 3770k and was able to beat Dumo's results only by a couple of GFlops:


So there's some marginal IB gain from Dumo's SB results but still lagging like everyone else when compared to deachus.

Ok next go lol... His mem is also slack & 2T .......

Yup (ran it slack on purpose to make it easier to get stable results). But tightening CR and subtimings there would give only marginal improvement and still won't catch deachus' results.

So let's have more 3770k >= 5Ghz 4c8t results please ...

[owikh84]

Lol! This one's on water: 3770k 5.05Ghz 1.315v on bios

alex_tpc | Core i7 3770K [4C 8T] @ 5049.9MHz | 1.312 | ASUS Maximus IV Extreme | Water | L215B620

now i'm jealous because I got the same batch as yours but mine needs at least 1.57v for 5GHz then throttled due to overheat issue.

[zoson]

I thought that as well when I posted it but there was not much 3770k 4c8t 5Ghz benches to compare with. I did look at the 2600k 4c8t benches from page 1 (see comments below) and p3gaz's 3770k 4c4t result (I get similar results when I do the same).

Yes, it does get better GFlops with HT Off. I mimicked the settings from p3gaz's 3770k 5Ghz 4t4c results and got very similar gflops (stopped at 3 iterations as the official 21-iteration submission is the one above):
*snip*

Seems I'm on par at least with 4c4t 5Ghz.

I used the latest linx(0.64)_10.3.10.017 binaries. deachus' 135GFlops peak for 2600K 4c8t 5ghz result is something special, beating the other 2600k 4c8t results on page 1, even those with 5.1-5.3Ghz. It also beat by 20GFlops Dumo's 5.004Ghz results which had close but tighter tRP timing. I took out my hypers from storage and mimicked the settings (5Ghz 2133 c7-9-7-20) from deachus on my 3770k and was able to beat Dumo's results only by a couple of GFlops:
*snip*

So there's some marginal IB gain from Dumo's SB results but still lagging like everyone else when compared to deachus.

Yup (ran it slack on purpose to make it easier to get stable results). But tightening CR and subtimings there would give only marginal improvement and still won't catch deachus' results.

So let's have more 3770k >= 5Ghz 4c8t results please ...

There are several of us that don't lag behind deachus. It's because we leave C states enabled, among other technologies that make it harder to overclock. But the end result is our performance at equivalent clockspeeds is better than everyone else. For example, my 108 almost 109 GFlop run at 5GHz vs kitfit's 5.1GHz run that netted him only 110GFlops. Where the response by others was 'wait, that seems low now that I've seen zoson's runs'. =P

In short, the only thing special about deachus's run is an obvious display of actual overclocking skill.

[xarot]

I may be in a wrong thread, but the new LinX with most up-to-date Linpack libraries just doesn't work anymore properly for me. Could some guru help? I was trying to shoot for 4.5 GHz HT on again after a while.

I am getting VERY low gflops with my 990X and Rampage III Black Edition. If I use older libraries version LinX 0.6.4, I get what I expect, but with the newest versions, no go. Here is a screen shot to describe the problem, OS has been installed just a while ago. Removing OC does nothing or running RAM at 1333MHz. I have tried to up and lower QPI voltage and fiddle around with uncore and nb frequency.

The most upper LinX showing 84.2438 GFlops run is an older library version of LinX 0.6.4. Run with max memory (12GB).
The middle LinX showing 56.2561 Gflops is the newest library LinX 0.6.4. Run with max memory (12GB). Very low GFlops. Lower temps reported by RealTemp GT.
The third LinX showing 82.3844 GFlops run is the newest linrary LinX 0.6.4. Run by usin 4096 MB only. GFlops are on the right track again.

Anyone else seen the same? Prime95 27.7 produces the same temps as older LinX.



My OC tested here is stable for 4 hours older LinX, 8 hours Prime95 27.7 and 8 hours Memtest86 without errors.

[MikelMolto]

You must look at the Linx window downstair in the status Line. Here You see, that high GFlops show 4 Threads (even CPU-Z show HT is on) and lower GFlops show 8 Threads.
Here is the example: http://www.xtremesystems.org/forums/...=1#post4802911

[deachus]

There is nothing special about my 5Ghz LinX 2600K run other than alot of tweaking which took alot of time. I still have that setup minus that CPU, have another 2600k in that rig batch# 3122C601.

[zoson]

There is nothing special about my 5Ghz LinX 2600K run other than alot of tweaking which took alot of time. I still have that setup minus that CPU, have another 2600k in that rig batch# 3122C601.

Exactly. As I said... Just an obvious display of overclocking... *SKILL*

@xarot:
Set your problem size to 9216 for stress testing.

@mikelmolto:
ROFL. I hope you were making a bad joke that makes no sense whatsoever.

[xarot]

@xarot:
Set your problem size to 9216 for stress testing.

Thanks! I googled and found your old thread where you hit the same issue. At first I was confused as I tried to change the actual problem size, but I found from your thread that it is actually the memory size that must be set to 9216. Now it works!

[MikelMolto]

@zoson
Why should I joke about this? Repaired the link and please have a look for Your self.
Very easy. Just go at LinX to settings and select the number of Threads.

http://www.xtremesystems.org/forums/...=1#post4802911

one more:
http://www.zhoulander.com/projects/ivy/3770K-4300.png

[alex_tpc]

There are several of us that don't lag behind deachus.

Near 135Gflops peak with HT? Didn't see much of those in the list on page 1. I didn't scan the whole thread though so maybe there were HT results that didn't lag 135Gflops not yet listed on Page 1?

It's because we leave C states enabled, among other technologies that make it harder to overclock. But the end result is our performance at equivalent clockspeeds is better than everyone else. For example, my 108 almost 109 GFlop run at 5GHz vs kitfit's 5.1GHz run that netted him only 110GFlops. Where the response by others was 'wait, that seems low now that I've seen zoson's runs'. =P

In short, the only thing special about deachus's run is an obvious display of actual overclocking skill.

I don't disagree there but I still think it's special because few (none?) of the list on page 1 even came close with HT on. Those other SB 4c8t results and the single IB 4c8t submission (which was on par or better than most of those other submissions) were all "woefully low" in comparison. It may turn out to be low with respect to future IB with HT submissions but I wouldn't mind that at all.

There is nothing special about my 5Ghz LinX 2600K run other than alot of tweaking which took alot of time. I still have that setup minus that CPU, have another 2600k in that rig batch# 3122C601.

Yes, that's what's special about it (and the other HT submissions not coming close).

edit: corrected mistyped 4c4t to 4c8t.

[zoson]

Near 135Gflops peak with HT? Didn't see much of those in the list on page 1. I didn't scan the whole thread though so maybe there were HT results that didn't lag 135Gflops not yet listed on Page 1?



I don't disagree there but I still think it's special because few (none?) of the list on page 1 even came close with HT on. Those other SB 4c8t results and the single IB 4c8t submission (which was on par or better than most of those other submissions) were all "woefully low" in comparison. It may turn out to be low with respect to future IB with HT submissions but I wouldn't mind that at all.


Yes, that's what's special about it (and the other HT submissions not coming close).

edit: corrected mistyped 4c4t to 4c8t.

Zalbard hasn't updated the front page for many, many pages of results. He's been too busy, which is totally understandable.

I started a discussion in this thread about optimizing GFlops in LinX for the highest performance results. We ended up with a ton of really impressive GFlop/MHz ratios.

[alex_tpc]

Ok, I got 144 GFlops on Ivy with 5Ghz 4c8T HT on in bios but not due to any special oc or tweaking skills:



Yup, it's by forcing Linx to run linpack with 4 threads even though the setting is hyper-threaded in bios. Which means it's really running with HT off despite the bios setting (and it's pretty much the same Gflops +/- as the 4C4T result I posted earlier). Which is actually how the 135GFlops run with SB was obtained. Should have seen that earlier but was only looking at the CPU-Z screen when checking for Hyper-threading. Should have checked the Linx screen first time round. Oh well.

Incidentally, fired up my old SB again (2600K batch 3101A482) using 4GB hypers with 4T on linx, 8T on bios and got 141GFlops at 5Ghz with 2133 7-9-7-20 with full stable run:

alex_tpc | Core i7 2600K [4C 4T*] @ 5000.7MHz | 1.512 | ASUS Maximus IV Extreme | Water | 3101A482



Also tried a bunch of tweaks (C-states, page file disable, real-time prio, background prio, LSC, most devices disabled in bios, etc.) but they didn't really help much on SB/IB (that I tested). Best tweak was still tightening mems of course but diagnostic mode did help with a couple of fractional GFlops.

BTW, the IB chip was getting only around 138GFlops with the same 4GB Hypers and OS at 5Ghz 2133 7-9-7-20, 3Gflops less than SB on same settings and OS. Could it be that the linpack binaries that came with linx(0.64)_10.3.10.017 could still use additional optimizations for IB? Or is it because the IB run was also on a P67 board instead of Z77? In any case, is that 119GFlops IB run with true hyper-threading still "woefully low"?

[owikh84]

owikh84 | Core i7 3770K [4C/8T] @ 4999.94MHz | 1.440 | ASUS Maximus V Formula| Water | L220B296

[zoson]

BTW, the IB chip was getting only around 138GFlops with the same 4GB Hypers and OS at 5Ghz 2133 7-9-7-20, 3Gflops less than SB on same settings and OS. Could it be that the linpack binaries that came with linx(0.64)_10.3.10.017 could still use additional optimizations for IB? Or is it because the IB run was also on a P67 board instead of Z77? In any case, is that 119GFlops IB run with true hyper-threading still "woefully low"?

Intel releases a new binary for new chips. If you're running on the binary that came with LinX posted in this thread, you're doing it wrong.

[dr_dx]

Dr_Dx | Core i7 3770K [4C/8T] @ 5000.1MHz | 1.296v | ASUS Maximus V Formula| Water | L221B009 | LinX 0.6.4 AVX w/ Linpack 10.3.11.019 | stock TIM





@ 1.288v using Liquid Pro/Liquid Pro

[st0ned]

That's a hell of a CPU Should be amazing on cold too.

[alex_tpc]

Intel releases a new binary for new chips. If you're running on the binary that came with LinX posted in this thread, you're doing it wrong.

Not really. Got it from the Linx thread here at XS and crossed check the version at the Intel site: only 10.3.11.017 was available at that time. Have since tried .019, same result. I think I might have used a leaner OS install from my raid card on the 2600k run (just noticed the desktop background was different). have dismantled that array after getting a new board -- will try to get it back again later and try it if it indeed gives better linpack throughput.

A decent piece of silicon
Dr_Dx | Core i7 3770K [4C/8T] @ 5000.1MHz | 1.296v | ASUS Maximus V Formula| Water | L221B009 | LinX 0.6.4 AVX w/ Linpack 10.3.11.019 | stock TIM

decent indeed, if I could say so myself