Thuban and 4 sticks of RAM @ 1600 @ 1T ?

[Mad-Marty]

Hi,

is anybody running 4x4 GB Sticks of RAM successfull on a Phenom ii x6 at >= 1600 Mhz with command rate 1T?
If so, i would run 4x4 AMD Performance Edition (Patriot) RAM, rated for 1600 8-9-8-24 @ 1.65 V.


I wonder if the IMC can handle that ... at 1.3 V with 2600 MHz on the NB.

[SkullCracka]

i rub 4x2gb 1600cas6 on my x6. nb 2600 auto voltage.

[Mad-Marty]

1T or 2T?

[ga1ve1an]

I am running 16gb with my THuban at 1600mhz. THe timings I am using are 7-8-8-24 with command rate 1t. Been running it for quite some time now.

[ga1ve1an]

FOrgot to mention NB is at 3000mhz using 1.27v

[SkullCracka]

ya 1T.................

[ga1ve1an]

Here some screenshots so you can see settings...

[Albertini]

my x6 1055t won't run stable at 1t with 4x4gb kingston hyperx 1600 dimms.

cpu: 3500@1.3v
nb: 2500@1.15v

ram: 1333mhz, cl9, 2t @1.5v (voltages up to 1.7 didn't help making 1t absolutely stable)

btw: errors were ONLY found when testing with http://hcidesign.com/memtest/ or s&m memory test. even goold old memtest86+ couldn't find errors @ 1t.

So may I suggest testing with one of those two tools to assure your stability!

[ga1ve1an]

my x6 1055t won't run stable at 1t with 4x4gb kingston hyperx 1600 dimms.

cpu: 3500@1.3v
nb: 2500@1.15v

ram: 1333mhz, cl9, 2t @1.5v (voltages up to 1.7 didn't help making 1t absolutely stable)

btw: errors were ONLY found when testing with http://hcidesign.com/memtest/ or s&m memory test. even goold old memtest86+ couldn't find errors @ 1t.

So may I suggest testing with one of those two tools to assure your stability!

I will run some tests... I have been at these settings for more than half a year now gaming, working, multitasking different programs and some occasional video photo editing of family stuff. No crashes or stalls or resets. This was a run through of futuremark 11 at these settings also.



When I was configuring my memory I noticed that the physics test under futuremark would crash with bad settings. Dont know if this matters but this is 2 kits of 2x4gb G. SKILL Ripjaws 2133mhz underclocked. I saw better results with Phenom and tight timings then with high bandwidth settings. Let me run some of those tests you listed.

[ga1ve1an]

Here is a screenshot of it running 5 sessions of the memtest.. I ran six but couldnt do anything because the system was so loaded from all the memory usage and cpu load. Was pretty cool test. So as you can see 5 sessions still running fine and after i got the programs open i needed to do a screenshot for you all, I started the 6 session. That was using all my memory as you can see with what windows7 was using also.. Left me nothing afterwards... No problems as I mentioned. The 16GB memory at 1600 7-8-8-24-1t and NB solid at 3000mhz w 1.275v.



P.S. I did see that this program stress my cpu more than I guess other tests did.. SO i am playing with the voltage to see what I would need to stabilize with this test.. But it might be for no reason since this type of test is not a real world type stress. And as I mentioned earlier the system has been solid for over 6 months with the way it is. Again this was for the memory question started above anyways just to show the 4 sticks of 4gb are solid for me with thuban.

[Mad-Marty]

Thanks Guys,
so lets see if it will work for me ... additional 2x4 gb are ordered

Too bad it doesn't work for all combinations at 1T ... they (CPU + Memory Manufacturer) should really fix this together.

[ga1ve1an]

As i mentioned above. ONe thing I have always done is bought the highest bandwidth memory that I can afford and if I cant run at the rated bandwidth, I have noticed that the underclocks are always solid with tight timings.

[Albertini]

Hi, glad to hear yours is stable!

Finally got my 4x4gb kingston to run @1t, too

entered the spd timings manually (tRas and tRC were set far lower by my crosshairIV) and voila - 1t is stable now @1333mhz-9-9-9-25-34-1T (unfortunately there's no way of running them at the specified 1600mhz, not even two sticks did so btw

[ga1ve1an]

Hi, glad to hear yours is stable!

Finally got my 4x4gb kingston to run @1t, too

entered the spd timings manually (tRas and tRC were set far lower by my crosshairIV) and voila - 1t is stable now @1333mhz-9-9-9-25-34-1T (unfortunately there's no way of running them at the specified 1600mhz, not even two sticks did so btw

I think you could stablize it at 1600mhz if you lowered the FSB or put it back at 200mhz. I have played with my thuban quite a bit and realized that 200mhz gives best stability for tight memory timings. Of course I understand you will lose CPU performance because you dont have a black edition, but just something for you to know.

With my testing I have found the performance % increases follow this order(from largest gain to smallest)...
1.CPU Clock (obviously)
2.NB Clock (closer to 3000mhz the better)
3.Memory timings
4.Bandwidth (upping FSB at cost of memory timing stability the higher it is)

If you have black edition you should just mess with the first three.. if not black edition then obviously 1, 2, and 4 would be what you need to balance at the cost of number 3.
Based on that you probably got pretty good timings on your memory with a 261mhz FSB.

[lopri]

This chip isn't anything special. Bought at a local MC when it came out. (like 2 years ago?) I have a 955BE C2 and I stuck the 4GB sticks just for the sake of it and it also handles 16GB @1T just fine. IMC gets hot so try to keep the chip cool and set the vNB (IMC) as low as possible while maintaining stability. FWIW, it did not need any voltage adjustment going from 8GB to 16GB. Just had to loosen the timings. (vNB was around 1.25V measured by a digital multimeter)



Edit: Memory sticks are Corsairs and Gskills. I don't think you can tell anything by the brand these days anyway. You just need a good chemistry between the memory IC's and the board's BIOS. And some trial and errors if necessary.

[Mad-Marty]

Wow,
thats quite impressive, lopri.

4 Ghz @ 8-8-8-24-1T @ 250 Mhz TRef @ 3Ghz NB AND 4x4 Mixed Memory ...

My memory has not arrived yet ... *sigh*
Order Status is still in progress ...

[Albertini]

I think you could stablize it at 1600mhz if you lowered the FSB or put it back at 200mhz. I have played with my thuban quite a bit and realized that 200mhz gives best stability for tight memory timings. Of course I understand you will lose CPU performance because you dont have a black edition, but just something for you to know.

With my testing I have found the performance % increases follow this order(from largest gain to smallest)...
1.CPU Clock (obviously)
2.NB Clock (closer to 3000mhz the better)
3.Memory timings
4.Bandwidth (upping FSB at cost of memory timing stability the higher it is)

If you have black edition you should just mess with the first three.. if not black edition then obviously 1, 2, and 4 would be what you need to balance at the cost of number 3.
Based on that you probably got pretty good timings on your memory with a 261mhz FSB.

well, unfortunately I can't even run better timings at 250fsb, so how should I be able to do so at 261?

that's why I tried to maximize cpu and nb clock while running at stock vcore or even UV (1000mhz @0.8v), as my memory or IMC or both seems to be rather weak compared to the examples listed in this thread

[Albertini]

Wow,
thats quite impressive, lopri.

4 Ghz @ 8-8-8-24-1T @ 250 Mhz TRef @ 3Ghz NB AND 4x4 Mixed Memory ...
.

unbelievable, what are your voltages for cpu and cpu/nb(IMC)?

[ga1ve1an]

well, unfortunately I can't even run better timings at 250fsb, so how should I be able to do so at 261?

that's why I tried to maximize cpu and nb clock while running at stock vcore or even UV (1000mhz @0.8v), as my memory or IMC or both seems to be rather weak compared to the examples listed in this thread

I apologize, I swear I saw your sig say 261 FSB.. I got to lay off the caffeine!!!

[Mad-Marty]

unbelievable, what are your voltages for cpu and cpu/nb(IMC)?

You should read Lopri's post
He said, its 1.25V for vNB.

And the crazy thing is, he doing this on a 955BE C2

... this must be the best 955 BE on earth ...

[Albertini]

I apologize, I swear I saw your sig say 261 FSB.. I got to lay off the caffeine!!!

caffeine is never to blame - can't live without it

my signature wasn't up to date - so I have to apologize


@ Lopri's unbelievabe chip: I thought Lopri was talking about the 955BE, as the picture shows an x6 1090t

[ga1ve1an]

caffeine is never to blame - can't live without it

my signature wasn't up to date - so I have to apologize


@ Lopri's unbelievabe chip: I thought Lopri was talking about the 955BE, as the picture shows an x6 1090t

We wouldnt be xtreme if we didnt overclock ourselves with a little caffeine..

Here is a little update.. Running this stress test on the memory and it should pass.. I noticed if I can get 6 sessions running it is usually good. But as I mentioned the system is difficult to control since all the resources are being consumed. I tried to stress everything out so you can see the flexibility you should have with the nb and memory. Think you just have to play with it a bit. This is thuban at 4200mhz, NB speed at 3120, Memory pushed farther and harder at 1920mhz with the tightest timings I could get stable of 9-10-9-28 Command rate 1t. HT ref or FSB i pushed up to 240 for this.. didnt need much more voltage to stabilize the cpu at 4200. Each memory session is at 20% coverage now.



Hope this gives some confidence and help settings wise for anyone trying to stabilize their 4 sticks.

[BeepBeep2]

btw: errors were ONLY found when testing with http://hcidesign.com/memtest/ or s&m memory test. even goold old memtest86+ couldn't find errors @ 1t.

So may I suggest testing with one of those two tools to assure your stability!

Memtest86+ is pretty useless IMHO.
Prime95 Blend for ~4 hours would probably show error.

Guys, 1T isn't as important as you think. If you can run higher frequency with 2T at same timings then I would do so.

[ga1ve1an]

Memtest86+ is pretty useless IMHO.
Prime95 Blend for ~4 hours would probably show error.

Guys, 1T isn't as important as you think. If you can run higher frequency with 2T at same timings then I would do so.

I am prime95 blend stable.. Tested that the other day when i was pushing the system further. I started up another session now, though to ask how prime95 will show error. It is stressing the cpu much more than the memory. Even looking at the comparison of the task bar you can see the memtest stresses all your memory which is what you need to do if you are finding weaknesses in the memory (also what the topic was about). Prime95 used at most 3.61GB of memory. Memtest up to 14GB. Also I am guessing that 6 sessions of memtest using 14GB is stressing the NB much worse than prime64 is with 3.61GB. Only makes sense. But correct me if I am wrong

[BeepBeep2]

First select blend, then select "custom" and change the amount of ram used to be whatever you want...
As for how it will show error, a thread will fail with something akin to "ROUNDING ERROR: Rounding was 0.x, expected less than 0.4".