SR-2 Optimization Thread

Memory dividers playing into my slow frames, you think? I set it up exactly as Z2D had his, but I wasn't sure since I had no -bigadv units ever get this far. I do know that my standard A3 times that used to be around 1:15 with my old settings dropped off to around 1:37 under these new settings though....

You are running L5640's right? If so, can you post your memory settings / uncore, etc for reference, or compare them to Z2D's?
 
My memory is at 1066 (2:8), so 1488 on the slow one and 1584 on the fast one. Uncore is on Auto for both, which is 2133 - 16x - 2x memory. My Vtt is 1.35 on both, and Vcore is around 1.35V for both. IOH is 1.4 for both. Memory timings are 8-8-8-20-2T for the faster one, and 9-9-9-24-2T for the slow one. I have a bad memory chip in the slow one, which is hurting my overclock on it. I'll get some new memory on order this weekend for it.

Edit: Also, no signal tweaks for either, MCH is auto for both, memory voltage it 1.669 actual for the fast one and 1.658 for the slow one.
 
Do you think it's hurting me only having one 2Gb chip per each processor. Not getting double speeds (single vs. double)?
 
Yes, that is probably the main reason. If at all possible, get to triple channel somehow. I am going to do 6 x 1Gb chips.
 
I figured out my issue, its heat, once my CPU's hit 65c or above WHILE THE AMBIENT ROOM TEMP IS ABOVE 28c, my overclock gets flakey. If the ambient is 26c or below, the CPU's can hit up 75c no issue. I tested this by varying the speed of my fan on my radiator and room using a floor fan.

Unfortunately due to my broken window(see died in line of duty thread), I cannot fold, as ambient temps in my room QUICKLY rise without the ability to open a window(floor fan isn't enough). If the SR-2 was the only system in the room, the floor fan can keep up, but I do use my office space to run alot of electronic equipment and do not have anywhere else CURRENTLY to put my SR-2.

Anyways, here are my system specs and currently used settings;

EVGA SR-2
Dual X5650's running at 4.114GHz;
CPU0 = 2x4GB Dual Channel DDR3 @ 1496 MHz (8GB)
CPU1 = 1x1GB Dual Channel DDR3 @ 1496 MHz (2GB)

BIOS Revision/Version: A46
187 BCLK x 22 multiplier (with Turbo and SpeedStep enabled)
VDroop: WITHOUT VDroop
VCore: 1.38125v
VTT: 1.4v
DRAM Voltage: 1.59V (shows as 1.63v in e-leet)
DRAM: 2:8 / 1066mhz > 1496Mhz, 9-9-9-24-74-1T
IOH: 1.425v (my IOH droops by over .026v, shows up as 1.39v in e-leet at this setting)
QPI: 4.8GT/s
Uncore: doesn't seem to work, but set to 16x
MCH Strap: 1333mhz

Now, remember my lower than should be expected at my speed PPD is due to my asynchronous dual channel memory setup

Code: 
 Project ID: 2684
 Core: GRO-A3
 Credit: 8955
 Frames: 100


 Name: SR2 BIGADV
 Path: C:\Windows\SysWOW64\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:18:30 - 77,397.9 PPD
 Avg. Time / Frame : 00:19:09 - 73,490.9 PPD

Code: 
 Project ID: 2686
 Core: GRO-A3
 Credit: 8955
 Frames: 100


 Name: SR2 BIGADV
 Path: C:\Windows\SysWOW64\
 Number of Frames Observed: 100

 Min. Time / Frame : 00:14:13 - 114,892.1 PPD
 Avg. Time / Frame : 00:14:43 - 109,086.9 PPD

And a 6701 just for good measure

Code: 
 Project ID: 6701
 Core: GRO-A3
 Credit: 921
 Frames: 100


 Name: SR2 BIGADV
 Path: C:\Windows\SysWOW64\
 Number of Frames Observed: 232

 Min. Time / Frame : 00:02:52 - 51,428.7 PPD
 Avg. Time / Frame : 00:02:59 - 48,441.6 PPD
 
Last edited:
Axdrenalin said:
Is this the only problem / issue that is being corrected in this update? If so, I'm not going to mess with it for the time being.
Click to expand...

Event 46 is known as and what you would see as a 124 BSOD, its a very good fix in otherwords. Its also reported that the OC profiles are fixed with this BIOS.

As for the Event 46 fix, it should allow people to lower their vtt(as it was a "common fix" to raise the vtt, even at stcok settings to fix this BSOD) and ultimately might get you a better overclock.

Worth playing around with in any case, as I was getting the Event 46 error all the time, but I just assumed my OC wasn't stable and kept pumping the vtt up was, it was only when I read about the BIOS fix and how others solved their issue by raising their VTT(even at stock, BCLk 133).

I'm going to try it anyways, plus remember the SR-2 has 3 BIOS Chips, in fact I currently have 3 different versions flashed on each of them right now, the original(A41), the one I am using now(A46), and what was the newest but no longer is(A47). You don't have to worry about issues the new BIOS might cause, as you can just flip the switch to your older BIOS. :D
 
LOCO LAPTOP said:
Yes, I'm the one. :)
Click to expand...

Most of us don't need this mod. Anyone with a pair of older Gainstown ES chips might. I know of one at the moment, and I think he already found the modded bios. The L5640 crowd should be fine with the EVGA-released version, although it definately won't hurt anything to use this one. Loco's last bios worked like a charm, so I am sure this one is fine as well.
 
musky said:
Most of us don't need this mod. Anyone with a pair of older Gainstown ES chips might. I know of one at the moment, and I think he already found the modded bios. The L5640 crowd should be fine with the EVGA-released version, although it definately won't hurt anything to use this one. Loco's last bios worked like a charm, so I am sure this one is fine as well.
Click to expand...

No but we DO NEED LocoLaptop!!!!!!!! :)

I have mine now rock solid at 182 Bclk x 22 multiplier. Unfortunately my x5650 chips need a fair bit of VCore (1.34375) and VTT (1.40) to reach this speed, but temps are in the 50s so I think I will leave it here.

Funny enough the speed difference on P2684s is not very different between 3.8 and 4.0ghz.

At 3.8 (190x20) it was doing 17:54 a frame and at 4.0 (182x22) it's going between 17:50 and 17:51. I might try and tighten timings from the loose 9-9-9-25-2T and check stability then, but I know my max BCLK is 195 right now and I need 1.425V VTT to get it there.

Wowzers.
 
have you tried 190x21?

I'm wondering if your lost bclk and memory bandwidth is counteracting your increased CPU clock speed.
 
10e said:
No but we DO NEED LocoLaptop!!!!!!!! :)
Click to expand...

Not disagreeing at all. He saved my butt when I was sitting with 2 x E5530s and a brand new SR-2 and couldn't do anything with it. I'm just saying that the L5640 ES, which is the processor in probably 80%+ of the [H] SR-2 rigs, does not need a custom bios.
 
Has anyone with L5640's and the SR-2 played around with the A49 BIOS update? I loaded it tonight - first Loco's version, then the EVGA version, and I'm in the process of going back to A47 now. I can't get the same clocks with the newer A49 that I was getting with the A47 (at least with the Z2D configuration / settings). :( I bet I've reset this crap at least a dozen times over the past couple hours tooling around with it...
 
Hmmm, I haven't got L5620s, but I went to A49 yesterday and it certainly hasn't hurt my clocks with my new x5660 build

USEFUL NUGGET: use the additional power connectors at 1.4v vtt and above.

I was experimenting with higher vcore and vtt, and I found that even though EVGA say you do not need the additional power connectors to each CPU unless you are doing extreme cooling OC, they do help.

I found that over 1.4v vcore, going from 1.375vtt to 1.400vtt would cause bluescreen #124 STOP errors in the first or second frame of Intelburntest2.5 Very High 32 threads. At the end of the unit, when the power usage spikes another 80w or so, bang, reboot. This got my attention, because STOP #124s are the ones where you are supposed to add more vtt or vcore, not less. I plugged in the extra power connectors and things were stable for 1.5 hrs plus. Removed them and instant reboots in one or two frames of IntelBurn. So they stay in.

Now the question is - is that beneficial if you were running say 1.375 vcore and 1.4+ vtt. Possibly not, as there would be more power left over from vcore for vtt.

But I guess I would say try it - costs nothing if you have the plugs, so go for it.

My x5650 and x5660 seem to be more vcore hungry and less vtt hungry that you guys running the lower multi cpus, but I thought this nugget might be of help.:)
 
Good info. I should give those connectors a try.
Posted via [H] Mobile Device
 
Are you talking about the 6 pin connectors by the 8 pins, or something else?
 
Why would anyone try to OC without plugging in every available power plug? The only one I think people could avoid is the one solely for Tri-Quad SLI setups
 
Vaulter98c said:
Why would anyone try to OC without plugging in every available power plug? The only one I think people could avoid is the one solely for Tri-Quad SLI setups
Click to expand...

Simple, not enough plugs with my PSU, until I consolidated and borrowed one from my first SR2 build.

I agree with you, always plug in the optionals if you can. But when the EVGA product manager specifically said on the evga forum you would not need them unless you are doing liquid nitrogen cooling, I thought I would be ok. And as my final overclock is only 1.35 vtt, i probably don't need them, but why leave them in the box eh?
 
SR2 #2 x5660 Overclock settings

Same config at SR2 #1 (see Post #7 and #142) but with the following differences:

**************
CURRENT 4.3GHz BIOS settings - Jan 2011

See here: http://www.pbase.com/davexl/x5660
**************

  • X5660 rather than x5650
  • Only 12GB RAM - only red slots filled.
  • 1020 watt Enermax PSU, not 1250.
  • No GPU worth mentioning - headless render box
  • No case = savings upgraded the CPUs.:D
  • HDD, no ssd.

The most important difference with the second SR2 is the higher multi on the x5660, which means a ~4.5% lower baseclock than my x5650 for the same Ghz.
Compared:
SR2#1: x5650: 191 x 22 = 4202Mhz, at 2:8 mem = 12:21 bench on P2686 = 142,700 ppd
SR2#2: x5660: 185 x 23 = 4255Mhz, at 2:10 mem =11:54 bench on P2686 = 150,000 ppd

This means I have been able to go to 2:10 (1333 ddr) divider on the ram, which has meant a good performance boost. This was a bit of a risk, because if I hadn't been able to do 2:10 I would have been WORSE off than with the cheaper CPU at the same clock speed. So the 25% boost in DDR speed was somewhat reduced by looser timings, and 5% slower baseclock, but still enough to come in for the big win.

This makes me want to go and retry my main rig at 2:10, but it will be 12 DIMMs at a higher baseclock, which would be pushing things at current 191 bclck = 1910DDR. Plus I have heard of 2000DDR 12 dimm setups frying CPUs, which is sobering. It is quite possible I could get better folding performance at a lower CPU clock given the memory boost.

Lessons:

  • The x5650 and the x5660 seem to need exactly the same voltage for the same clockspeed. The difference is that the x5660 has an easier multi, and can keep going higher a little bit better.
  • I don't seem to need high vtt like those fighting lower multi cpus.
  • Plug in your secondary 6 pin power cables. You never know.
  • I did go up to 4.324Ghz and pass a few hours HCI memtest, so I know my memory has a bit of headroom.
  • MCH of 1067 boots with only 8gb, MCH of 1867 won't post. 1333 is best performance, while being stable.
  • Mem timings on auto at this strap are 11 11 11 24 etc, but tightening them to 9 10 9 24 etc was possible.
  • Caseless naked on the floor - a few degrees cooler than table height - is about 10c cooler than open case upright SR2#1. We will see how dust and wandering curious two year old fingers become issues.
  • I can get folding stable at 4.3 Ghz at about 1.43 vcore, but I don't want to run that 24/7, even though temps are absolutely fine at 24c ambient.
  • I basically dropped to whatever speed could bench my 2686 at 11mins:54sec a frame, which is the magic 150,000 ppd. ;)
  • Also, being on the ragged edge, there was very little difference in performance between 4255Mhz and 4300.

A note on screenshots: Settings 1.4125 vcore, 1.375 vtt passed 8 hours memtest, 8 hours Intelburntest2.5 at very high 32 thread. I usually do 12 hours, but this is a headless renderbox, not my main rig. I have since dropped to 1.40625 vcore 1.35 vtt.

Another 8 hr+ stable combo is: 182 x 23= 4186 Ghz 1.39375 vcore 1.375 vtt 1.4 IOH QPI -82 - although in hindsight you could drop the vtt and IOH probably.


Idle = 268 watts C-state disabled.
Folding 6701 = 578w
Folding 2686 = 572w
Folding temps are 70c on CPU0, 61c on CPU1 at 22c ambient. woot.

original.jpg

original.jpg

original.jpg


Voltage is set to Without Droop.

original.jpg

original.jpg

I am not convinced uncore does anything at all anymore. I thought you had to set it at 2x mem divider speed, but I think it is locked.
original.jpg

Voltage is set to Without Droop.

The following settings 1.40625 vcore 1.35 vtt are at least 4 hours stable Intelburntest - I cut short my usual validation given the race to 4 billion etc. They are folding stable so far.

(1.4125 vcore, 1.375 vtt passes 8 hours Intelburntest2.5.)

original.jpg

original.jpg


original.jpg


I am not convinced there is much actually going on here - I have got success from between -78 and -85, but hey, it works.
 
Last edited:
This is great M.I.B.W.

I'll try this tonight, and thanks for your efforts, they're hugely appreciated.

I did stumble upon a little tweak which does not affect OC/stability and removed just under 2 minutes on a P2684 unit's TPF.

This takes PPD from 78,000 to 92,000, on a 2684 of all things.

You have to turn OFF NUMA, and turn ON UMA (not Thurman, she's not in the BIOS unfortunately)

Now, for the life of me, since I was gunning through the BIOS this morning, I can't remember where the heck it is, but I'll find it when I go home at lunch and update with a pic, unless someone else posts first.

I think it was under the ACPI configuration menus.

I remembered this from my Magny-Cours box, where there was nearly a 5 minute change in TPF when I turned OFF NUMA.

NUMA allows one CPU to access another's memory bank, but unfortunately, since F@H is not NUMA aware, it takes a great feature and unwittingly uses it to actually lower performance.

UMA only allows a CPU to access its own dedicated memory bank which is the fastest path to RAM.

Try it out. Hopefully it helps.
 
Oh you bastard :D

It is 2:20 am here and I am rebooting both to play!

OF COURSE! It all went click when you wrote that - the benchmarks that vary so much... it is NUMA. Time to learn all about it. :p
 
may i be worthy said:
Oh you bastard :D

It is 2:20 am here and I am rebooting both to play!

OF COURSE! It all went click when you wrote that - the benchmarks that vary so much... it is NUMA. Time to learn all about it. :p
Click to expand...

Yeah, I saw this in LinX. With NUMA I was getting 55 GFlops rate and with UMA it went up to 77 GFlops.

This is how I thank you for all your efforts! More work :p

jebo_4jc said:
I thought NUMA was an AMD-only feature?
Click to expand...

Ninja'd by Tobit!!!!!!!!!!!!!!

It's prevalent in SMP systems. I had to share it with the team. Imagine every SR-2 on the team getting a 10 - 15K PPD upgrade just from this.

After this I'm going to try and tweak it to get to a 14:00 TPF, so it will drop a BigAdv unit at least once a day, even 2684s!

Go [H]0rde!
 
Ran one frame of my 2686 bench, and... 11 mins 20
161,400 PPD
:D:D:D:):cool:;):D:eek::eek::):):cool::):D

EDIT 1 hour later: - that was with nocpulock=0 - the next frame was 11:43 - drifts.But with nocpulock=1 - first 2 frames were each exactly 11mins 16:
162,850 PPD
:D:D:D:):cool:;):D:eek::eek::):):cool::):D:D:D:D:):cool:;):D:eek::eek::):)::):D:D:D:D:):cool:;):D:eek::eek::):):cool::)

Farrrrrk meeeeeeeeeee!

Kicking myself, I thought of this ages ago, but dismissed it as I thought it halved your effective RAM.

Took my 6060 standard from 81,000ppd to 88,000ppd on SR2#1 too.

I HAVE to go to bed, but I am naming my next born 10e Jr....;)
 
Last edited:
Any answers on this, cuz I'm gonna upgrade skynet in 50 minutes..

ninja'd.. so where is it in bios, this is amazing!
 
Can someone explain to me what this uncore stuff is, and why it's not on my bios? I thought the only bios update fixed some issue with using all the mem banks and wasnt needed?
 
162,850 ppd :eek: |V| G!
Can't wait to see how much of a gain with L5640s.

Btw, what ram kits are you guys using with sr-2? I'm shopping around for kits, 6x1gb would be enough, unless 2gbs are near that price.
 
404 - File not found said:
162,850 ppd :eek: |V| G!
Can't wait to see how much of a gain with L5640s.

Btw, what ram kits are you guys using with sr-2? I'm shopping around for kits, 6x1gb would be enough, unless 2gbs are near that price.
Click to expand...
corsair xms3 1600mhz

flawless memory.
 
Disabling NUMA added 7 seconds to my frame times for a P6073. I do not see an UMA setting.

FYI, these are under the Power Management main menu.
 
You must log in or register to reply here.
Back
Top