LinX/IBT linpack Gulftown >9GB problem size bug.

[zoson]

First of all, if you don't have >9GB ram... You have never experienced this issue. I'm not looking for any sort of speculation, just facts. Also, if you have a Bloomfield, this issue does not affect you.

Here's what we know of the bug so far:
Allocating more than 9216MB causes low GFlops in linpack when HT is enabled.
For example:
vs

You can see that all my timings and settings in these two pictures are identical. The only difference is the problem size, and there is a drastic change in GFlops between 8GB problem size and 10GB problem size.

Some members have reported that they have managed to get full GFlops *some times* but not always, and they have never been able to pinpoint the cause of the issue. Disabling HT does solve the low GFlop problem, but I am looking for the solution with HT on.

So far, the only people that I'm directly aware have this issue are running the ASUS R3E. Are there any users with EVGA or Gbyte mobo's that experience this same issue?

Things we've tried:
Increase/decrease all ram timings/RTL/tRL.
Increase/decrease UCLK(higher UCLK does get higher GFlops, but does not resolve the issue).
Increase/decrease voltages all around(vcore, qpi, ddr, ioh, pll, pwm freqs...)
Updated linpack binaries(I've got the latest from intel's site).

Anyway, if someone knows what the deal is, I'm sure many of us here would appreciate it if you shared your knowledge.

[subaruwrc]

I'm the only who thinks, you should move this thread to " Intel " section instead of " extreme overclocking " ?

[zoson]

If that's where it belongs, please move.

[l0ud_sil3nc3]

I'm the only who thinks, you should move this thread to " Intel " section instead of " extreme overclocking " ?

agreed

[Andreezy]

i also have this problem now and i was able to achieve full bandwidth before.
all test done @ around 4,200MHZ.



and this is the bandwidth of the 12GB kit that currently is capped(screenshot shows full bandwidth before becoming capped).


this is the current speed im running now with HT OFF, system wont run
full bandwidth with HT ON no matter what i try.



and this is the mistery, when i noticed some weird stuff going on.

[skuldarin]

i've been able to go from the lowered Gflops to normal compared to HT on and off. the only thing that seemed to help was have my CR set to 1 instead of 2... only downside is that YES i get my normal 96-98 gFlops at 4.5ghz, but if i use full mem, it stops on error every time... if i test NEAR full mem, it passes everything... and i've done all the adjustments you listed about and it had no effect... still stoped on error if it was over 11gb loaded mem, anything lower passes every time.

[zoson]

Skuldarin, you're on my ignore list for a reason. If you keep posting to my threads with your garbage, you're obviously doing it just to get a rise out of me. Stop it, or you will be reported for harassment.

Or, if you're going to post, actually read the OP and responses so you know what's going on and don't post useless drivel.

Obviously me and Andreezy already know about the HT on/off effect, as Andreezy discovered that and posted it to the R3E forum. Not to mention we both discuss it as an unacceptable solution to the issue in this thread.

I've said it before and I'll say it again. I do not want your help, Skuldarin.

[Andreezy]

i will soon throw my GT chip and 12GB kit on a Asus P6T7 WS mobo.
of course i will document everything and will be more than happy to share.
hopefully it will behave like a real GT chip should.

[Andreezy]

well i have bad news, i loaded the GulfTOWN onto my other motherboard (asus p6t7ws) and tested with 12gb of ram and the results are the same as the rampage 3 extreme.

i guess the special cpus pushed through intels retail channel are duds.
i can only say i have a 970 with a very, very weak IMC.
running AIDA64 i could not reach 20K read/write or copy.


with all same hardware and a Core i7 950 i can reach 23K,22K,25K.
more people really need to come in and chime in.
i was looking though a lot of bench results, and lots of the benchers that
are upthere on the Mhz, are only using dual channel with xtreme cpus.
i didnt catch that before.

i still say intel flocked the 6 core cpu line.
they need to send me a 990X for my pain and suffering, i dont do R&D for free.
well for my fellow Xtreme Members yeah its worth burning 1 chip err 2

[zoson]

I was curious if this was overclocking related, so I dropped my CPU down to 3.3GHz(23x143), but kept my UCLK and memory frequency about the same. Same result at stock speeds. ~66GFlops at 9216, ~56GFlops at 10240.

[Andreezy]

i have a 6GB set of OCZ 2000mhz mem that gave me same cap problems.
i will try it again and we'll see what happens on the asus p6t7ws.

i also had a 12GB set of Patriot 2000mhz 9-11-9-27, which i believe is Hynix chips.



that memory always worked no tweaking required. too bad i sold it in a
system i built for someone.