1366 BCLK-binning

[ground]

A couple months I figured out that my Rampage II is a beast regarding BCLK - with my example beating the previous known top BCLK by 15 MHz with ambient cooling. I then decided that I had a new goal regarding overclocking - reaching the highest BCLK on x58 (if my boards can handle it...)

32nm, '11:
E5606:
E5606 3128C492 273.6 (286 Dice, valid.x86.fr/b2qbur)
E5606 3152A744 264 (272 at -~30, SS valid.x86.fr/bb12hx)
E5606 3128C492 262
E5606 3152A744 261 (281 at -100, cascade: valid.x86.fr/vphh25)
E5606 3128C492 259
E5606 3128C492 258
E5606 3128C492 256
E5606 3128C492 255
E5606 3116B954 255
E5606 3128C492 254
E5606 3152A744 253
E5606 3128C492 253
E5606 3128C492 253
E5606 3128C492 252
E5606 3152A744 252
E5606 3152A744 250
E5606 3204B111 250
E5606 3128C492 249
E5606 3152A744 245
E5606 3214D005 235
E5606 3128C492 <250
E5606 3128C492 <250
E5606 3128C492 <250
E5606 3128C492 <250
E5606 3152A744 <250
E5606 3152A744 <250
E5606 3152A744 <250
E5606 3116B954 <250
E5606 3116B954 <250
E5606 3116B954 <250
E5606 3202B183 <250
E5606 3202B183 <250

E5649x1
E5649 3147C174 269.65 (4.76 R15 @1.4V) (283 @dice)




32nm, '10:
E5640x16
Sadly most of these were on a board which maxed out at 253 BCLK, so I couldn't do as extensive testing (and I already resold most of these). Best one did 253 BCLK (this was after I was mad at the chip being unstable in 4.9 Cinebench at 1.45V). Binning Data incomplete.

E5640 3104C403 270 (4.74GHz R15 at 1.4V) (287.3 Dice valid.x86.fr/7meuwa)
E5640 3104C403 265 (4.78GHz R15 at 1.4V)
E5640 3104C403 255
E5640 3116B925 253
E5640 3104C403 243
E5640 3003B404 241
E5640 3038A995 230
E5640 L038A827 225



E5620x4 + x8 Tapakah
E5620 3201B759 256 (4.7/1.3V 32M)
E5620 3201B759 252
E5620 Unknown   238
E5620 3038B045 230

E5620 3019A816 243
E5620 3019A816 <240
E5620 3019A816 <240
E5620 3019A816 <240 
E5620 3019A816 <240
E5620 3019A816 <240
E5620 3019A816 <240
E5620 3041B447 <240

X5650x4
X5650 3038A141 244 (266 SS)
X5650 unknown   238
X5650 unknown   235
X5650 unknown   222





45nm, D0:
W3520
W3520 3849A734 238 (4.5/1.3V 32m)
W3520 3847A700 233
W3520 3949A608 231
W3520 3847A703 231
W3520 3946A269 228
W3520 3946A269 225
W3520 3946A269 225
W3520 3946A269 225
W3520 3013A424 224
W3520 3913B220 222
W3520 3946A269 221
W3520 3941B133 221
W3520 3946A269 221
W3520 3002A679 221
W3520 3003B139 221
W3520 3919A569 221
W3520 3941B113 219
W3520 3951A169 210
W3520 3844A733 150

W3530
W3530 3020A571 232 (4.5/1.4V 32m)
W3530 3101A273 225
W3530 3140B357 221
W3530 3102C311 220
W3530 3052B884 220
W3530 3043A272 220
W3530 3006B132 220
W3530 3110C700 215
W3530 3041B239 215
W3530 3109C356 210
W3530 3101A273 210
W3530 3111B660 205

i7 920
i7 920 3849A823 230 (4.5 1.4V 32m)
i7 920 3926A286 218 (4.5 1.4V 32m)
i7 920 3913B268 212

E5506
E5506 3122D673 230





I had a couple more that I resold before doing more in depth testing (W3680 that didn't manage 4.0 at 1.4V and barely managed 250 BCLK, L5640 that maxed out somewhere around 238 BCLK)



Most testing was done with the following settings:
MB - Asus Rampage 2 Extreme
RAM - 2 or 6 GB Elpida BDSE (manages 2000 MHz, shouldn't hold me back. I've now managed to find a set of Hypers rated 8-8-8 @2000, but they seem to manage 6-6-6 @1900 with relative ease. Need more testing for 6-6-6 @2000)
CPU Cooler - Alpenföhn Brocken Eco (usually weighed down by a pile of books)
Fan - fans pointed at VRM and NB
VGA - 8600 GT, maxes out at 138 PCIe
PSU - BeQuiet! L8 530W
HDD - Maxtor 61 GB IDE
----------------------------------------------
AI overclock - manual 
OC from CPU level up - auto
OC from Memory level up - auto
CPU Ratio setting - 13

CPU configuration
CPU Ratio Setting - 13
C1E Support - disable 
Hardware prefetcher - enable 
Adjacent Cache line prefetcher - disable
Intel Virtualization tech - disabled 
CPU TM Function -  disable
Execute Disabled bit - disabled
Intel HT Technology - disable (though this made little difference)
Active Processor Cores - all (made little to no difference)
A20M - disabled 
Intel Speedstep tech - disable
Intel C-STATE tech - disabled 

BCLK frequency - 250, max the board can boot at is 261. Oddly enough the shown clock speed freaks out above 255 (I suspect an overflow)
PCIE frequency - 118 (max the board can boot)
DRAM frequency - DDR3 1504MHz
UCLK frequency - 3510MHz
QPI frequency - Slow Mode (this starts being necessary above ~225-240 BCLK)

Dram timing control
Fairly lose timings for BCLK testing

EPU II phase control - full phase
Load-line calibration - disable
CPU differential amplitude - 1000mV
Extreme OV - disabled

Current voltage 1.45v - 1.812v - 1.375v (May be different)
Cpu voltage - 1.45V
CPU PLL voltage - 1.812V
QPI-Dram voltage - 1.375V (haven't had a CPU die with these settings, basically no scaling even if upped to 1.5V. I have a couple basically useless E5606s that I want to test with up to 1.8V and see if that makes any difference...)

Current voltage 1.139v - 1.508v - 1.26v - 1.508v
IOH voltage - 1.39166v (without slow mode this is extremely important, can bring 10-20 MHz)
IOH PCIE voltage - 1.51106v
ICH voltage- 1.25916v (this can help booting more PCIe, but it requires about 0.2V/MHz and isn't really worth touching)
ICH PCIE voltage - 1.51106v

Current voltage 1.8v
DRAM Bus voltage - 1.64356v
DRAM REF voltage - auto

Debug mode - string
Keyboard TweakIt control - disabled

CPU spread spectrum - disabled
PCIE spectrum - disabled
CPU clock skew - auto (for binning these remain on auto, manually tweaking them can yield 2-3 MHz and isn't really worth the time while binning)
IOH clock skew - auto


After that, I booted into windows, started SetFSB, selected ICS9LPRS916JGLF, upped PCIe to close to the max the current setup can handle (135-138 MHz) and then upped BCLK in small steps (1-3 MHz at a time). 


PCIe formula:
Rough formula for PCIe scaling: (BCLK-221)*0.445+100


(I also have a couple other boards, a DFI x58 DK-T3eH6 with a dead BIOS and a Gigabyte x58a-oc that maxes out at a usable 266 BCLK/268 very unstable)

If anyone has any questions, more ideas or experience feel free to share 

Temperature scaling:
Going with very limited information (2 CPUs, air, SS and LN2):
Scaling on x5650:

Very inaccurate representation, but enough to see a general trend with just the 4 given points. It is possible there are settings we missed (was just helping out a teammate maxing out his x5650 - top 3 for an entire platform seems fair for a random CPU). Peak BCLK on this CPU is somewhere between -35°C and -50°C, while peak benchable clocks are closer to the cold bug. Will do more testing once I get my own source of cold - currently only backseat benching with a member of my team. Btw, if anyone wants me to join them for for their own session with any 1366 CPU under cold I would be up for that for sure, I need more data 


Information found elsewhere:
high vtt can change the coldbug, usually in an unwanted way
www.techpowerup.com/forums/threads/picking-cpus-by-batch-cherry-picking.157547/
www.overclockers.com/forums/showthread.php/592186-i7-920-max-bclk-on-AIR-230Bclk
www.xtremesystems.org/forums/showthread.php?214381-DFI-X58-i7-920-max-bclk-on-AIR-230Bclk
extreme.pcgameshardware.de/overclocking-prozessoren/359269-how-uebertakten-mit-rw-everything-auf-non-oc-boards-alternative-zu-setfsb.html
forums.tweaktown.com/gigabyte/39229-gigabyte-award-x58-bios-led-debug-codes.html
www.xtremesystems.org/forums/showthread.php?223694-222-Bclk-quot-wall-quot-help-me-out-here&p=3769982&viewfull=1#post3769982 <--- reason for PCIe requirement and some extra data
www.xtremesystems.org/forums/showthread.php?209143-Team-AU-Gigabyte-X58-UD5-3GB-Corsair-1866MHz-CAS8-Core-i7-920-Testing-thread/page4&p=3794415#post3794415 more pcie stuff
www.xtremesystems.org/forums/showthread.php?261643-I7-uncore-dram-rule
www.madshrimps.be/vbulletin/f10/mass-msi-x58-eclipse-sli-64610/ Slowmode explained
www.madshrimps.be/vbulletin/f10/core-i7-thoughts-findings-one-thread-combine-them-65704/
i4memory.com/showthread.php?t=15908 x58 + Hyper tricks

Stuff to test on a CPU basis after initial binning:
45nm VTT voltage scaling (impact almost not noticeable for 32nm, heard there are chips that scale up to 1.65V VTT at ambient which needs more testing. 
45nm DFI vs Asus, have a semi-dead Lanparty board that needs a resurrection...
High DRAM volts (1.8+)



Board Limitation:
Given that on other BCLK platforms the board is often the limiting factor, I've been trying to find more data on this; currently I only have 4 boards worth noting on the list:
1: x58a-oc: 268; limited by board (seems like I've hit a dud there, the CPU did, at the time, 269.65)
2: Rampage II Extreme: 271; likely limited by CPU
3: Rampage II Extreme: 276.5 COLD; either limited by board, CPU, or just by cooling (SS wasn't as cold as it could've been)
4: Rampage III Formula: 277.59 COLD; likely limited by CPU/temp precision



Likely best boards for the platform/bclk (only listing boards I’ve heard reports/had myself/can extrapolate from experience with very similar boards)
Top tier: Rampage III Black Edition, x58a-oc
High mid tier: x58a-ud9, x58a-ud7, Rampage III Extreme/Formula, Foxconn Bloodrage(needs hardmod!), MSI Big Bang, Evga Classified,
Mid Tier: Rampage II Extreme, x58a-ud3r, x58a-ud5, P6T WS variants, Rampage III/II Gene
Lower tier: P6T SE, Asus Sabertooth, gigabyte EX58 (need hardmods), Evga non-classified, MSI x58 Pro-E


-Ground

[Vinster]

Great info, when I got to my 229, BLK wasn't the goal, I was benching at the time. it was being cooled by a water chiller (Home-Brew AC) and at idle would hold -25C and loaded would be between -10C to -15C. I still have that board, but in the process of seeling the CPU on an old MB.

Vin

[Aleslammer]

ground, THANK YOU!

[ground]

Now, do I buy 1 X5675 so I have a new CPU for my daily or do I get 72 E5606s for about the same price?

[Vinster]

for a lot of that stuff I go to ServerMonkey.com

Here's one for $70USD; www.servermonkey.com/3-06-ghz-six-coreintel-xeon-x5675-six-core-12-mb-cache.html

they call it refurb, but it's from a server pull. they deal with off-lease. so you know it's not battered gear.

Vin

EDIT: because of a sale, for $5 more you can get an x5680; www.servermonkey.com/3-33-ghz-six-coreintel-xeon-x5680-six-core-12-mb-cache.html

[ground]

Jun 24, 2018 20:03:41 GMT -5 Vinster said:

for a lot of that stuff I go to ServerMonkey.com

Here's one for $70USD; www.servermonkey.com/3-06-ghz-six-coreintel-xeon-x5675-six-core-12-mb-cache.html

they call it refurb, but it's from a server pull. they deal with off-lease. so you know it's not battered gear.

Vin

EDIT: because of a sale, for $5 more you can get an x5680; www.servermonkey.com/3-33-ghz-six-coreintel-xeon-x5680-six-core-12-mb-cache.html

Vinster Avatar
Jun 24, 2018 20:03:41 GMT -5 Vinster said:
for a lot of that stuff I go to ServerMonkey.com

Here's one for $70USD; www.servermonkey.com/3-06-ghz-six-coreintel-xeon-x5675-six-core-12-mb-cache.html

they call it refurb, but it's from a server pull. they deal with off-lease. so you know it's not battered gear.

Vin

EDIT: because of a sale, for $5 more you can get an x5680; www.servermonkey.com/3-33-ghz-six-coreintel-xeon-x5680-six-core-12-mb-cache.html
I wouldn't expect chinese sellers on ebay to be selling binning rejects, not too worried. Kinda broke anyways, bought 24 W3530s and 12 W3520s (cheap through auctions) so I'll delay that a bit. X5680 is likely to be a worse bin at least according to TDP and being an earlier chip

Given how impractical overclock.net is to use now, anyone wanting to add some more x58 clockgens?
Rampage II extreme: ICS9LPRS916JGLF
x58a-oc ICS9LPRS918BKL
DFI DK-x58-T3eH6 ICS9LPRS918JKLF
Rampage III Formula: ICS9LPRS918JKLF

Aleslammer : Are you willing to take on my E5649 challenge?

[Bones]

Great opportunity to get in on this:
warp9-systems.proboards.com/thread/677/warp9-hold-line-bench-off

Since it's about ref frequency you guys could go after each other and participate at the same time.

[ground]

Jul 8, 2018 18:42:00 GMT -5 Bones said:

Great opportunity to get in on this:
warp9-systems.proboards.com/thread/677/warp9-hold-line-bench-off

Since it's about ref frequency you guys could go after each other and participate at the same time.

I'm on the wrong team and don't have anything that could compete with 775  
Done some first testing with 45nm D0s:
CPU     Batch        Max BCLK under Air
W3520 3849A734 238
W3520 3847A700 233
W3520 3949A608 231
W3520 3847A703 231
W3520 3946A269 225
W3520 3946A269 225
W3520 3946A269 225
W3520 3013A424 224
W3520 3913B220 222
W3520 3918A768 222
W3520 3946A269 221
W3520 3941B133 221

While I can't confirm anything yet, given HWbot rankings for this CPU and me doubting that all those binned 10+ CPUs to just find one that does well, I suspect that BCLK actually scales with cold. If anyone would be willing to try I would give away the W3520 3847A700 233 for shipping costs (as well as a 258 E5606 to confirm that the board can do high BCLK beforehand)


Oh, and Aleslammer - I haven't forgotten that you've taken my E5640 and E5649 scores, I now have everything I need to beat most of those again...

[Bones]

Don't worry about that - 775 boards are easy to come by, you just have to know what to look for in both boards and chips.

[Vinster]

Jul 12, 2018 4:51:53 GMT -5 ground said:

Jul 8, 2018 18:42:00 GMT -5 Bones said:

I'm on the wrong team

these internal competitions are open to all Warp9 members, regardless of the HWBot team you are on

Wolfdales seems to be kings atm.

and Ales would be the member to push those chips properly. I have a few 1366 boards, but my chiller is not setup atm. My chiller does about -30c unloaded and about -10c to -15c loaded on those board..

Vin

[Bones]

That's right - Join in anytime you'd like, it's all good.

[ground]

Quick testing with a couple more xeons:
E5620:
1: 256 BCLK max
2: 252 BCLK max
X5650:
1: 222 BCLK max
2: 238 BCLK max
3: 235 BCLK max

I also may have found a source of as many xeons as I want for future binning projects, though thats gonna require some more time to stabilise...

[ground]

Quick update, had one of my team members test an E5606 (261 BCLK under air) under cold (weak SS). At about -30 he got the CPU to 277.6 BCLK. CPU will soon go under LN2... And this wasn't even the best one from the batch   Gonna add the information to the first post once everything is figured out.
Now to figure out how it scales. Is it gonna be +16 MHz for any CPU (given 50C temperature difference) or is it gonna be different on any chip? Or do different between generations (Nehalem, Westmere '10 and Westmere '11 are available for testing with prebinned BCLK) scale differently?

And lastly, did I manage to find a 300 BCLK CPU? Gonna find out in a while... (if the boards can handle it)

[Aleslammer]

Ground, will be setting up S1366 for TC, although only have two working boards one untested and a RIIIE that was starting to get a little ragged in the socket so willing to do a few Xeons just to see if I can get Bclk up

[ground]

Aug 3, 2018 9:29:52 GMT -5 Aleslammer said:

Ground, will be setting up S1366 for TC, although only have two working boards one untested and a RIIIE that was starting to get a little ragged in the socket so willing to do a few Xeons just to see if I can get Bclk up

Great! Looking forward what you can get out of them   Also please take on my E5649, though if I actually go cold I doubt you'll get anywhere close - it's benchable at 4.8 under air after all... 


Pretty sure the current 1366 BCLK record will fall before this Team Cup ends no matter what, the 277 MHz was under a weak SS and on a chip "only" capable of 261 BCLK under air. I still have 3 chips that can do more (262, 269.65 and 271.3 under air), and the better one will go under LN2 in the hands of a teammate of mine 

I don't see 300 BCLK as impossible given the right chip and the right cooling method...

[Aleslammer]

Gas in the chiller only good to -25 +/-, VGA pot but no CPU pot so will do my best!!!

[ground]

Added some more information I got about cold scaling to the first post, its not how I expected it to be. If anyone wants to run one of these on cold and wants to share data/maybe want someone tagging along, I would like to see if I can be of any help.

[ground]

Went through my remaining W3520s and found a chip not able to boot windows at 160 BCLK, worst Nehalem CPU I've ever heard of. Nothing else interesting from that batch. Still have 12 W3530s and a 920 to test and then I need to find a new source of CPUs...

[Bones]

Keep at it Ground, I do wish you luck with what's left to check and maybe you'll find more chips cheap too.

[ground]

Went through 6 more E5640s today, I think I got pretty lucky with these.
E5640 270 (4740 R15 at 1.4V, Uncore 3800 at 1.4V QPI, DRAM 1940 at 1.4V QPI
E5640 265 (4780 R15 at 1.4V, Uncore 4100 at 1.4V QPI, DRAM 1940 at 1.4V QPI
E5640 255
E5640 243
E5640 230
E5640 225
Gonna run the top two on dice in a couple weeks, got a pot now to take care of that. I have 4 candidates for max bclk (current record for the platform is 290), I hope one of them is good enough for it (271 E5606, 270 E5640, 269.65 E5659, 265 E5640)

[ground]

Getting a known 277 ambient CPU (a badly tortured E5620) for the 265 E5640, thats gonna be fun if my board likes it...

[ground]

...277 ambient has to be taken with a big grain of salt, the previous owner was tuning a lot more skews then I have access to. Not disbelieving it, but as long as I don't have access to those settings its a 240 BCLK CPU for me.
However, I played a bit more with clock skews and managed 273.6 on my best E5606 at ambient, still looking forward to the dice session next week valid.x86.fr/nx11ae

[Bones]

Looks good!

[ground]

Dice session delayed as the meet up failed, currently working out a way to get it in my own town while paying as little as possible.

tested 2 more boards with my E5649 (baseline CPU for boards, because it works on basically all of them)

So, list of boards properly tested with that CPU remains short, but here you go:
Rampage II Extreme: 269.65 (very unstable after 268)
x58a-oc #1: 268.03 (very unstable beyond 266)
EX58-UD3R: 267.29 (very unstable beyond 266)
x58a-oc #2 266 (instant crashes beyond that, but boots it right up)

Either I'm unlucky with gigabyte boards or asus can reach higher BCLKs on average. Getting a UD7 later this month, after which I'm moving on to another MSI board and then continue with Asus most likely. Once I get dice I'm gonna run with multiple boards seeing if they handle differently there

[ground]

So, bad news and good news.

Finally got to the dice session, sadly it didn't go quite as planned. I figured out that the max bootable PCIe clock is a fairly relevant factor - my Rampage II, maxing out with 118 MHz PCIe thus was fairly problematic for reaching high BCLKs. I managed 279.99 ( www.hwbot.org/submission/3960133_ground1556_reference_frequency_rampage_ii_extreme_279.99_mhz ) on it, a nice score but not enough for me. My E5606 handled a bit better, but still caused some trouble - like not properly saving a 286 MHz validation. My best E5640 only managed around 281, just like the E5649.
Onto the second Board I had on hand; things worked a bit better here, until they stopped working altogether. I hope the board just got wet and isn't dead, because I currently can't find a good replacement for the x58a-oc I loaned out from a friend . Sadly I only started with the second best CPU I have on hand, my E5640. Had I started with the E5606 I might have gotten further, but the board stopped working before that sadly. So I'm left with a validation of 287.22 MHz ( www.hwbot.org/submission/3960131_ground1556_reference_frequency_x58a_oc_287.22_mhz ). If I get the time and the board working tomorrow, I'll try again with the E5606 hoping to finally beat Nanok. Else... I need to start binning boards for max bootable PCIe.




So, what new data do I get from this?

- max bootable BCLK and particularly PCIe clock are quite relevant while testing boards.
- scaling seems to differ CPU to CPU and board to board
- insulate better you fool. Any tricks to swapping CPU while on cold? Ran into wet sockets twice today because of that...

[tapakah]

When you're talking about 'bootable' PCIe clocks, you mean clocks bootable from a cold boot (=powerdown) or (re)bootable? What if you try raising the BCLK in OS via SetFSB?

Get some blue Scott's Shop Towels (that's the actual name) for insulation. Once I've switched over to these I stopped using any other kind of insulation whatsoever. These can easily take 10 CPU swaps on LN2 during a single session.

[Macsbeach98]

Cant do without the Blue shop towels.

[ground]

Oct 17, 2018 17:09:34 GMT -5 tapakah said:

When you're talking about 'bootable' PCIe clocks, you mean clocks bootable from a cold boot (=powerdown) or (re)bootable? What if you try raising the BCLK in OS via SetFSB?

Get some blue Scott's Shop Towels (that's the actual name) for insulation. Once I've switched over to these I stopped using any other kind of insulation whatsoever. These can easily take 10 CPU swaps on LN2 during a single session.

Set in BIOS and boot right up. There are some things the board doesn't allow changing in BIOS, so booting up higher BCLKs (with high required PCIe clocks) helps with maxing it out. In particular - booting 260 & 118 on the x58a-oc got me to ~282, booting 285 & 130 got me to 287 and booting 287 & 131 got me to 287.22 (so only a fairly insignificant increase there, but it is there). If the board remains in working order I'll do more testing tomorrow

[ground]

Gonna hunt down some Blue shop towels next time! Thanks for that hint

[ozz]

warp9-systems.proboards.com/thread/84/insulation-guide