SR-2 Optimization Thread

[may i be worthy]

hmmm - Chilly - you missing 2GB of ram?

 

[Chilly]

hmmm - Chilly - you missing 2GB of ram?

No

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

10GB Total, yea I run a messed up ram configiration... I ran out of money

 

[may i be worthy]

lol - righto, good for you for making it work though! I just thought you might be having the tight timings/disappearing memory issue I had...

 

[Aix.]

Great job jamming it into the cosmos 1000. I have the same case that house a x8dah+f but plan on ordering the sr2 soon. How many of the stand off holes lined up for you?

...not many. All of the standoffs (four IIRC) along the left side of the board lined up, but many of the others were hit and miss. There's enough that board is secured and the SR-2 also comes with some plastic standoffs (with adhesive mounting) that you can use for the holes that don't line up.

Heads up - hacked bios that unlocks uncore.

I have put it on my x5650, and it works, currently testing just what it buys me.

Proceed at own risk.

So...what exactly will/could this help with? After setting up my 980X with a 1.5x uncore I was surprised when the requirement for my L5640's was 2.0x. Will a lower uncore be a good thing? Will it even work?

 

[may i be worthy]

So...what exactly will/could this help with? After setting up my 980X with a 1.5x uncore I was surprised when the requirement for my L5640's was 2.0x. Will a lower uncore be a good thing? Will it even work?

Basically with the current bios uncore is locked at 20x. Loosening this to 19x or 18x reduces the vtt needed to be stable - but also loses a fraction of performance.

I have flashed my x5650 rig and have been experimenting - it does seem to allow slightly higher clocks, but it will take time to evaluate the benefits. So far it is not the second coming...

For me it is a subtle difference - I would be very motivated to start over if I was running 1.4v or higher vtt - as it is I am at 1.35 and 1.375 or each rig, so I wasn't being held back too much.

Still, it is a new thing to learn. But now I have another variable to test for

 

[Aix.]

Basically with the current bios uncore is locked at 20x. Loosening this to 19x or 18x reduces the vtt needed to be stable - but also loses a fraction of performance.

I have flashed my x5650 rig and have been experimenting - it does seem to allow slightly higher clocks, but it will take time to evaluate the benefits. So far it is not the second coming...

For me it is a subtle difference - I would be very motivated to start over if I was running 1.4v or higher vtt - as it is I am at 1.35 and 1.375 or each rig, so I wasn't being held back too much.

Still, it is a new thing to learn. But now I have another variable to test for

I look forward to your results! I've been reading over the entire thread since I'm looking for every bit of advice, but it's a little tricky when many of us are using slightly different processors and the BIOS seems to have come a long way in only a couple of months.

I was able to clock my L5640's at 3.6GHz (9-9-9-24-1T 1600MHz) after about half of a day of stress testing, but I haven't made any effort to fine-tune anything, or see what I can get past 200 BCLK (if anything).

Now that I'm watching the points from -bigadv projects roll in I'm curious what kind of performance I'm still leaving on the table (from either CPU or RAM MHz).

So far I've seen average -bigadv TPFs of:
2684: 18:57
2685: 14:14
2686: 13:29

So many settings to tweak...

 

[musky]

I look forward to your results! I've been reading over the entire thread since I'm looking for every bit of advice, but it's a little tricky when many of us are using slightly different processors and the BIOS seems to have come a long way in only a couple of months.

I was able to clock my L5640's at 3.6GHz (9-9-9-24-1T 1600MHz) after about half of a day of stress testing, but I haven't made any effort to fine-tune anything, or see what I can get past 200 BCLK (if anything).

Now that I'm watching the points from -bigadv projects roll in I'm curious what kind of performance I'm still leaving on the table (from either CPU or RAM MHz).

So far I've seen average -bigadv TPFs of:
2684: 18:57
2685: 14:14
2686: 13:29

So many settings to tweak...

Have you turned off NUMA? You are about 1 minute/frame too high across the board compared to mine at 3.6.

 

[Axdrenalin]

Have you turned off NUMA? You are about 1 minute/frame too high across the board compared to mine at 3.6.

I'm into a 2685 right now seeing about 13:17 per frame, and my 2684's finished out at 17:53 as well, so I agree with Musky here. Double check that NUMA option and disable it if you haven't already.

 

[may i be worthy]

I look forward to your results! I've been reading over the entire thread since I'm looking for every bit of advice, but it's a little tricky when many of us are using slightly different processors and the BIOS seems to have come a long way in only a couple of months. {SNIP}
So many settings to tweak...

Yes, it looks like there is quite a different experience with overclocking the x5650/60 and L5640 - we all share the same dramas with memory, but the lower multis really are hitting the SR2 a lot harder for baseclock, and need more vtt, but less vcore. Which implies that this new unlocked bios would be of even greater value to those with lower multiplier CPUs.

I had hoped that all my research would be of greater benefit to [H], but then you all had to find that goldmine of special cpus! Locally x5650s are still the cheapest hex you can get here, but it turns out I was going the gold plated route

My x5650 hacked uncore unlocked bios update - working fine still. Uncore does work - the worst is that when changing uncore you will often get a FF failure to post, but it is no drama - just hit reset and things boot up fine, no need to reset cmos.

I have done one successful overclock validation, where I reduce my vtt from 1.375 to 1.35, but as I tweaked a few other things at the same time (going for a lower power/more efficient summer overclock) I can't say for sure that this uncore was the sole benefit, but it is very likely to have helped. But one thing that is proved to my satisfaction is that 19x uncore and this hacked BIOS can be completely stable.

 

[musky]

Shouldn't uncore on Westmeres be at 1.5 times memory, so either 12x (8x memory) or 15x (10x memory)? It seems like you are still way high at 19x even.

 

[may i be worthy]

Supposed to be 2x I thought. The official bios which is locked always sets it to 20x, so there is the answer.

 

[Punchy]

There was a comment somewhere in the EVGA forum regarding uncore. I vaguely recall what it said was that the minimum was 2x for Nehalem and 1.5x for Westmere.

 

[nitrobass24]

Bloomfield is 2x
Westmere is 1.5x

http://techreport.com/articles.x/18581

 

[may i be worthy]

I am trying 19x, 18x, 17x etc and benching them on my captured 2686 unit.. more to follow in a full writeup later. The performance loss on folding looks miniscule...

14x uncore on my 2:10 x5660 does boot and show the expected amount of memory - for what that is worth.

 

[10e]

There was a comment somewhere in the EVGA forum regarding uncore. I vaguely recall what it said was that the minimum was 2x for Nehalem and 1.5x for Westmere.

I believe that's true, but for the life of me I can't remember where else I read it.

I am trying 19x, 18x, 17x etc and benching them on my captured 2686 unit.. more to follow in a full writeup later. The performance loss on folding looks miniscule...

14x uncore on my 2:10 x5660 does boot and show the expected amount of memory - for what that is worth.

That's good stuff. I always wondered why CPU-Z showed the "northbridge" at 20x even when I changed uncore in the BIOS. "Phantom setting" or am I mixing things up?

 

[nitrobass24]

I believe that's true, but for the life of me I can't remember where else I read it.

look 2 post up

 

[SazanEyes]

I had hoped that all my research would be of greater benefit to [H], but then you all had to find that goldmine of special cpus!

I for one appreciate all the work you're doing with the X56x0 chips. My X5650s should be back up and running soon, and I'd really like to be able to hit 4GHz.

 

[10e]

look 2 post up

No that wasn't it but thanks, I saw that as well

I never said my posts make sense heh.

I for one appreciate all the work you're doing with the X56x0 chips. My X5650s should be back up and running soon, and I'd really like to be able to hit 4GHz.

Me too. 3.96 (180x22) is the highest I can get without going into the 1.4x vcore range, and 3.9Ghz (195x20) requires 1.425 VTT for each chip. So far neither is showing much advantage in folding performance, as each is within 2 seconds of each other on a P2685.

I'm sure with the uncore running slower I can either lower VTT for higher BCLKS, or use these BCLKs without the magic disappearing DIMM.

 

[may i be worthy]

Thanks guys, appreciate the appreciation Your reward:

Effect of Uncore on Folding Times

The purpose of lowering Uncore from default is to relieve stress, that can result in lowering the potentially harmful vtt voltage. If that was also your bottleneck, slightly higher clocks.

Uncore is locked in the current A49 bios, but has been unlocked thanks to the folks here:

Prelim conclusions:

• Lowering uncore has a tiny performance impact, but has the potential to gain back more than you lose, particularly on stability.
• 18x to 19x (tested on a 2:10 / "1333" divider) looks like the sweet spot.
• The major gain is in maybe lowering VTT and perhaps vcore, not boosting your OC dramatically.
• Funky behavior has resulted in one of my systems - under test.

As the difference is a fraction of a percent, I had to run a TEDIOUS series of hour long tests. Not helped by effing windows 7, that is very hard to run benchmarks 5 mins after boot without misc background processes running here and there, which can add 15 sec to frame times (the dud times deleted in table) So [H], for the love of the game, I test so you don't have to

Win some/lose some:

My x5660 rig is now 9 hour stable at 4.3Ghz, (up from 4.255) using uncore 18x This is at a higher voltage than my everyday 4.255Ghz, but the point is that it is stable now at voltages is was not before:

• 1.425v vcore and 1.350vtt, 18x uncore hacked A49 = now passes 9 hours Intelburntest2.5/linpack v.high 32 thread
• 1.431v vcore and 1.375vtt, 20x uncore locked A49 bios = would fail in 1.5 hours

I will try to lower vcore more, or try 19x.

So what did all this work on uncore gain, by going up to 4.3Ghz?

The dropping of vtt is the main benefit. The 2686 bench file above testing at 11:12 across 4 frames = 164,300PPD, a net gain of 1450PPD!! (I can tell you are underwhelmed - I prefer to think of it as half a million points a year extra )

But as this is currently the fastest (and ugliest) rig in the [H]orde, I would have to hand in my [H] badge and gun unless I did another max performance 2692 bench on the flagship: 11:08, 11:10, 11:09, 11:09 = 11:09 average = 165,400 PPD.

BUT.....

My x5650 rig, while Linpack stable, now has a problem. It passes tests, and folded SMP overnight successfully, but since I have been fiddling and testing since, it has developed a problem. It now hangs/double frame times on the second or third frame of my bigadv test, at 3 known stable OCs. Also at stock optimised defaults. Also has long frame times on another captured 2685 Now this machine has been prone to this in the past, with the odd frame running long, but nothing like this. I have flicked the switch and loaded the official A49 bios, and so far it seems fine and consistent. I will reflash and try again. EDIT: reflashed and now working as expected, back to testing to see if I can get same OC for less volts.

So there are two canarys in this coal mine - one singing, one having a little lie-down.

PS; 4.3 stable screenie:

 

[nimbyfaygo]

Well after spending almost 3 hours reading this thread, I have came to the conclusion that everyone is crazy and needs to hand over the keys.

I look forward to getting mine in and seeing what I can do with it. Before I can do any of that I have to spend the effort to get it physically mounted in the case.

I must give props to everyone who has posted in here. The sheer amount of knowledge is amazing, and the amount of time documenting all the trial and error is very helpful.

 

[Vaulter98c]

I was able to get 75k at 3.95 on a good unit from those chips bro, you should be able to squeak near 80 I'm hoping with the Numa trick and now unlocked uncore's

good luck, your mileage may vary but I'm hoping you can get 4 stable, I don't think any of us could do that on 5530's on the stock bios, but now you've got a fighting chance!
Posted via [H] Mobile Device

 

[SazanEyes]

Well, installing my board in a case seems to have cooked the OC on the L5640s. I think it's a temp issue, even with the open case. (I saw temps go a few degrees higher just by installing the SR-2 on the Mountain Mods tray but otherwise naked.) I've dropped back to my old standby voltages of 1.3V core, 1.35V VTT/IOH and I'm trying BCLK at 185 (= 3.33GHz). I've made it through a few LinX passes so far and temps are down more than 10 degrees. A few more LinX passes and I'll go back to F@H, and experiment some more later. I lost a few WUs already tonight, so I want to get a few successes.

 

[10e]

Well, installing my board in a case seems to have cooked the OC on the L5640s. I think it's a temp issue, even with the open case. (I saw temps go a few degrees higher just by installing the SR-2 on the Mountain Mods tray but otherwise naked.) I've dropped back to my old standby voltages of 1.3V core, 1.35V VTT/IOH and I'm trying BCLK at 185 (= 3.33GHz). I've made it through a few LinX passes so far and temps are down more than 10 degrees. A few more LinX passes and I'll go back to F@H, and experiment some more later. I lost a few WUs already tonight, so I want to get a few successes.

It seems those L5640s are temperature sensitive moreso than x56xx chips. But I have had the same issue with my box. Since closing the PC-P80, I had to fiddle with volts as CPU0 (primary) shot up by something like 4 to 5 degrees at the socket.

This is the one closest to the fans. WTF. I have since lowered the front fan's speed and all is well (??).

MIBWorthy: You rock. I'm going to try that this weekend. Always remember the BIOS switch. I usually leave 2 at original BIOS, and 1 and 3 at two later ones for those times when things just aren't going my way.

 

[SazanEyes]

It seems those L5640s are temperature sensitive moreso than x56xx chips. But I have had the same issue with my box. Since closing the PC-P80, I had to fiddle with volts as CPU0 (primary) shot up by something like 4 to 5 degrees at the socket.

It's always nice to know I'm not the only person dealing with something. The L5640s definitely run cooler than my X5650s at the same voltages (and higher BCLK but lower clock speeds). I can probably get back to 3.6GHz with some adjustments. As it is, I only lost about 10K PPD on bigadv judging by the P2685 I got this morning.

 

[ebolamonkey3]

Hey guys, hopefully I'll be getting a SR-2 soon and a pair of Xeons and start folding. Just a question, where do you guys get your processors? Ebay?

Just trying to cut down my costs as much as possible

 

[jebo_4jc]

I bought mine from a trader here on the forums. I think he has more, so I can give you his name.

It seems many others bought theirs on ebay.

 

[SazanEyes]

eBay is the best place to find Xeon deals. I think all the L5640 "extra special" chips here came from eBay, but that well has run dry. I got my X5650s from TigerDirect when Bing was offering 15% cash back.

BTW, I went with G.SKILL Ripjaws for my new SR-2 build, and the fins don't seem to interfere with the TRUE heatsinks. The fins don't seem that much shorter than the Corsair Dominator, but I guess it was just enough. I'll know for sure when I actually mount the heatsinks.

 

[10e]

Hey guys, hopefully I'll be getting a SR-2 soon and a pair of Xeons and start folding. Just a question, where do you guys get your processors? Ebay?

Just trying to cut down my costs as much as possible

Ebay's a good place to look for 'em. Too bad all the L5640s were snapped up by [h]ardcore folders!

It's always nice to know I'm not the only person dealing with something. The L5640s definitely run cooler than my X5650s at the same voltages (and higher BCLK but lower clock speeds). I can probably get back to 3.6GHz with some adjustments. As it is, I only lost about 10K PPD on bigadv judging by the P2685 I got this morning.

I did some digging and on EVGA's forums everyone is seeing the same thing: Primary CPU gets higher voltage and temperatures regardless of the CPU model. Right now I've gotten them up to 4.0Ghz but that's the absolute maximum I can get without losing a DIMM of memory due to IMC issues, and this is at 182x22.

Unfortunately during folding I have cores on the primary CPU hitting close to 80 celsius so I might dial it down back to 3.9 or 3.8Ghz. But 195BCLK is my max with a 20 multiplier and that's with 1.425 VTT on each CPU.

But this is with my "FrankenKit" of 3 dual channel G.Skill DDR3 1600mhz RAM kits, so I ordered two sets of triple channel Corsair TR3X6G1600C9 RAM for about $110.00 each on sale to see if this helps with the issue. I think 1.425 VTT is just a bit too high for VTT, and I'd rather temps went down.

Alternatively I might try my Noctua NH-D14 on the first CPU instead of the Megahalems I have on it now and see if this gets temps on the primary CPU down a bit.

Lots of fun!

 

[ebolamonkey3]

eBay is the best place to find Xeon deals. I think all the L5640 "extra special" chips here came from eBay, but that well has run dry. I got my X5650s from TigerDirect when Bing was offering 15% cash back.

BTW, I went with G.SKILL Ripjaws for my new SR-2 build, and the fins don't seem to interfere with the TRUE heatsinks. The fins don't seem that much shorter than the Corsair Dominator, but I guess it was just enough. I'll know for sure when I actually mount the heatsinks.

I have ripjaws too and their heatspreaders are a little bit lower than most other rams (unlike Sector 7 or Pi or GEIL). I can fit them just under a thermalright silver arrow, but the 140mm fan won't fit. This is the difference b/t noctua nh-d14 and the silver arrow in terms of fitting, noctua's a bit better b/c it uses a 120mm fan on the outside.

 

[10e]

I have ripjaws too and their heatspreaders are a little bit lower than most other rams (unlike Sector 7 or Pi or GEIL). I can fit them just under a thermalright silver arrow, but the 140mm fan won't fit. This is the difference b/t noctua nh-d14 and the silver arrow in terms of fitting, noctua's a bit better b/c it uses a 120mm fan on the outside.

I've had the G.Skills currently in the SR-2 in an 1156 motherboard with the Noctua NH-D14 and clearance was not an issue. These are F3-12800CL9D-4GBRL kits with the standard RipJaws heat spreader fins.

The issue with the 140mm fan on the TR Silver Arrow is that if you can clip it in a way that's higher on the heatsink and thus, further away from the DIMMs, it may actually suffer clearance problems with most standard width cases.

Either way, good luck!

 

[ebolamonkey3]

Oh btw, what cases are you guys using for the SR-2?

I've got an 800D now, but dunno if I want to mod it to fit or just buy a new case, probably would end up spending about the same amount either way.

 

[nitrobass24]

I am seriously considering buying a 700D and modding it for my second SR-2

 

[jebo_4jc]

yeah, if you ask me the 700d appears to be one of the best options. I assume the 800d is capable of containing it with the same mods

 

[SazanEyes]

Oh btw, what cases are you guys using for the SR-2?

Most people here fold nekkid.

I have a Mountain Mods U2-UFO and a Lian Li PC-V2120. I like them both, but the Lian Li is definitely more case for the money. However, the MM can fit on top of my desk without looking like the monolith from 2001.

 

[Kendrak]

Most people here fold nekkid.

I have a Mountain Mods U2-UFO and a Lian Li PC-V2120. I like them both, but the Lian Li is definitely more case for the money. However, the MM can fit on top of my desk without looking like the monolith from 2001.

The twins are nekkid, laying on the desk.

Not sure the monolith is a bad thing.

 

[may i be worthy]

Nekkid is good. Messy, sprawling, but oh so cool.

Thinking of decanting my first SR2 - in a Lian Li PC-P80B, as the weather is warming up down here.

 

[musky]

Oh btw, what cases are you guys using for the SR-2?

Compaq 9000 series, 42U...

http://www.avid-edge.com/pc/SR2_3.JPG
http://www.avid-edge.com/pc/SR2_4.JPG

 

[may i be worthy]

Love the compaq rack musky - I am guessing the SR2 has more oomph than what used to live inside?

 

[jebo_4jc]

that rack looks like a holdover from the Netburst era, at the latest

Musky, I assume you've considered dual fans? Dropped temps by 5 ish degrees on my tower heatsinks.

 

[musky]

Once I get the second one and the dually Asus in there, I'll pretty much guarentee it will have more processing power in it that it ever had...