Odd speed reading and Prime95 lockups & reboot
GA-X58A-UD5 (BIOS v 5), i7-930, Win7 ultimate 64 bit (not yet updated cuz I don't have this online yet). Not running anything unusual like overclocking. But POST is telling me the CPU is running at 2.93 (x22) instead of 2.8 (x21). The BIOS is set at 21 and I disabled the turboboost. That seems strange. And when I run Prime95 for 10 or 15 minutes, it still runs but everything else is locked up. Core Temp won't respond, task manager won't respond, and can't even get to the start button. I've never had that happen on my WinXP Q6600 system. I can run Prime95 all day and still do other things. Any ideas?
I ran memtest86+ for 4 passes which took 4 hours. No problems. But Prime95 (blend) crashed the machine after about hour and a half with blend test. Just completely restarted. No blue screen that I noticed, just complete shut down and restart. How can I find out why? Also lockups previously. This machine is barely overclocked. It's mostly out of the box settings. CPU clock x21 (default setting); Performance Enhance=Turbo (default setting); Extreme Memory Profile(X.M.P.)=Profile1 (not default but set this way so my DDR1333 will run at 1333 instead of 1066); Profile DDR Voltage 1.65 (automatically set); Profile QPI voltage 1.25 (automatically set); CPU Vcore 1.257 (auto); QPI/Vtt voltage 1.175 (auto); IOH Core 1.100 volts (auto); DRAM voltage 1.500 (auto).
Like I said, most everything is set at defaults. I have not changed any voltages or clock settings. Just the one for the RAM to be 1333 instead of 1066. And because of the Turbo setting above, CPUZ shows the CPU to be running at x22 which gives me 2.93 or so. But I just noticed that it changes randomly. Right now it shows Core #0 at 1619.3 multiplier x12 bus speed 134.9; QPI link 2428.8. When I run Prime95 it goes up to 21 or 22 (mostly 21) but when idle it goes down to 12. I didn't know it could do that. Is this a problem? Is this what's causing Prime95 to fail?
Here's my equipment:
Motherboard = GIGABYTE GA-X58A-UD5 (BIOS v 5)
CPU = Intel Core i7-930 Bloomfield 2.8GHz LGA 1366
RAM = (12GB total) CORSAIR XMS3 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10666) 7-7-7-20 Model TR3X6G1333C7 G
Cooler = Noctua/XION NH-U12P SE2 Universal CPU Cooler
Video = EVGA 512-P3-N871-AR GeForce 9800 GTX+
Sound = HT | OMEGA CLARO Plus+
Case = LIAN LI PC-P80 or PC-A77F
PSU = CORSAIR CMPSU-1000HX 1000W
Storage = (4) WD6402AAEX 640GB SATA 6.0Gb/s & (4) WD1001FALS 1TB SATA 3.0Gb/s
Optical = (2) LITE-ON DVD Writer - Bulk - Black SATA Model iHAS224-06
OS = Windows 7 Ultimate 64 bit
EDIT: Just ran it again. Was installing Adobe CS3 Master Collection at the same time so I could run some CPU/RAM intensive Photoshop things to load it up even more. But after an hour and 12 minutes, crash. I was watching closely this time. No BSD. Just black and then reboot.
EDIT 2: Speedfan shows my +3.3v at 3.3; +5v at 5; +12V at 9.54; -12v at -1; -5V at -1.5; +5V at 3.63; Vbat at 3.12. I'm not well versed at this part. This is the first time I've used a PSU with >1 rail. I've always avoided that. My fans and HDD's are on the top row of black connectors. I'm not using any of the blue PCIe outlets. There are 2 PCIe cables hardwired into the PSU. I'm using one of those for my video card. Also the first time using a modular PSU. Always avoided that too.
Quote:I just noticed that it changes randomly. Right now it shows Core #0 at 1619.3 multiplier x12 bus speed 134.9; QPI link 2428.8. When I run Prime95 it goes up to 21 or 22 (mostly 21) but when idle it goes down to 12. I didn't know it could do that.
Likely 'core throttling' - but seems really odd that it would do it during Prime (unless a core has already failed & its thread 'shut down'...Quote:Is this a problem?
In theory - no - it's a 'feature'; in practice - often! This is an energy saving thingie - great if you're running a server farm, or a notebook; not so great if you're running a lot of RAM fast...
I'd give this a try -
On the "Advanced CPU Features" sub-page of the "MB Intelligent Tweaker(M.I.T.)" page of the BIOS:
"CPU Enhanced Halt (C1E)" to "Disabled"
"C3/C6/C7 State Support"
"CPU EIST Function"
On the "Advanced DRAM Features" sub-page:
"Performance Enhance" to "Standard"
...'bump' your DDR voltage by half a tenth to 1.55
...set your tRFC about 10%-15% higher than the SPD/XMP sets it - i.e., if it's at '60', bump it up to '66' or even '69'
Quote:"CPU Enhanced Halt (C1E)" to "Disabled"
"C3/C6/C7 State Support"
"CPU EIST Function"
ALL of the above 3 to disabled?Quote:...'bump' your DDR voltage by half a tenth to 1.55
DDR Voltage reads 1.65 in the BIOS on that page and is dimmed. Are you talking about the DRAM Voltage near the bottom of the main "MB Intelligent Tweaker(M.I.T.)" page? If so, the options are 1.54, 1.56, 1.58, etc. I made it 1.54 and Windows wouldn't start.Quote:...set your tRFC about 10%-15% higher than the SPD/XMP sets it - i.e., if it's at '60', bump it up to '66' or even '69'
If I'm in the right place, tRFC is not editable because DRAM Timing Selectable is set to Auto but the tRFC for channels A, B & C are all at 74
This is the hardest build I've ever done. I usually keep things pretty vanilla because most of the time I canabalize my current machine for the new build and can't afford much down time. THis is a whole new box with all new gear. I wanted to keep it simple though (for example not overclocking) initially so that I could at least get it up and running and then learn about OC afterward. So much about that that I don't know. I'm a little perplexed that the default settings are giving me trouble.
Sorry misread something - I thought I saw your RAM set to 1.5V - also, just assumed the 'DRAM timing selectable' change - will add it to my 'canned' stuffQuote:ALL of the above 3 to disabled?
Yup - all of 'em - makes your processor/MOBO quit 'wobbling aroung' your voltages and frequencies in the attempt to save twelve cents a week in power cost!Quote:I'm a little perplexed that the default settings are giving me trouble.
Your problem here is two DIMMs per channel; auto settings are made to accommodate one DIMM per channel - has ever been thus: more RAM = more 'tweaking'! What confuses me is that you seem to need XMP enabled to get 1333; I'll have to check the datasheet for the 930, but I'm pretty sure it's one of the ones that, 'out-of-the-box' supports 800, 1066 and 1333 (which, by the way, are the only speeds supported on any 1156/1366!); the 'regular' JEDEC SPD on the RAM should have 1333 tables (I think?). Why don't you download CPU-Tweaker 1.5 beta 4, click on the SPD button, and post the resultant screen here, as well as the main screen...
If you've never posted an image, I have a tutorial 'how-to'...
Here's the CPU Tweaker data:
I called Gigabyte last week. I had originally ordered 12 GB of 1600 RAM. The tech told me that the 930 would only support 1600 if I used 3 sticks. If I was to use 6 then I have to go with 1333. Not sure if that's entirely true or not but it makes sense based on the 930 Intel page. It says up to 24GB and 800 or 1066. The Gigabyte tech said that the 1066 limit is if you go to 24GB. If you go to 12 then you can use 1333. I have to say I spent a ton of time looking into what I could use and found very little that made anything clear enough. THat's why I ended up ordering 1600 initially because nothing was clear and it seems like many people out there are getting that or higher. But after my conversation with Gigabyte I refused shipment on the 1600 and ordered the 1333. Then after I started runnign things I noticed it was running at 1066. Another tech at Gigabyte told me I had to set the XMP so that it would see the 1333. So that's what I did. Here's the page on the 930: http://ark.intel.com/Product.aspx?id=41447
Anyway, I had no idea using 6 sticks would cause me so much grief! LOL. At least I'm not in a rush to get this running properly. Well, sort of. I am under the pressure of 30 day RMA issues of course. I ordered quite a bit of the gear on April 13th. So I still have a bit of time. I want to be sure that if anything is bad I catch it in the next week or two so that's the only pressure I'm under right now.
Take me a bit to get a look and try to see what's up... In the meantime, something to digest while I change a water filter and finish reassembling a water softener I'm servicing:
"Intel® Core™ i7 Processor Extreme Edition and Intel® Core™ i7 Processor Datasheet, Volume 2"
2.14 Integrated Memory Controller Miscellaneous Registers
2.14.1 MC_DIMM_CLK_RATIO_STATUS This register contains status information about DIMM clock ratio
Device:3 Function:4 Offset:50h Access as Dword
Bit 28:24 MAX_RATIO. Maximum ratio allowed by the part.
Value = Qclk
00000 = RSVD
00110 = 800MHz
01000 = 1066MHz
01010 = 1333MHz
QCLK_RATIO. Current ratio of Qclk
Value = Qclk.
00000 = RSVD
00110 = 800MHz
01000 = 1066MHz
01010 = 1333MHz
2.14.2 MC_DIMM_CLK_RATIO This register is Requested DIMM clock ratio (Qclk), the data rate going to the DIMM. The clock sent to the DIMM is 1/2 of QCLK rate
Device:3 Function:4 Offset:54h Access as Dword
QCLK_RATIO. Requested ratio of Qclk/Bclk.
00000 = RSVD
00110 = 800MHz
01000 = 1066MHz
01010 = 1333MHz
As Elmer Fudd used to say, at the end of every cartoon, "Th-Th-Th-That's all, folks!" Everything else falls under the broad label of 'undocumented' - like fifteenth century maps marked "here be dragons!" I'm not saying it can't work; it obviously does work, sometimes... Somehow, the BIOS and the board hardware are being manipulated to 'fool' the CPU into clocking the memory faster than spec - but it's one of those "pay no attention to that little man behind the curtain" things... As I posted to someone yesterday about the same issue: If you 'rob Peter to pay Paul' long enough, you wind up with a sore peter!
Clarke's third law is: "any sufficiently advanced technology is indistinguishable from magic" - my problem is, I don't like 'magic', and I don't like 'dragons'! Though people often look at the little 'widgets' dangling around my house, ask 'how's that work", and I tell 'em "FM" , it's not! The actual 'effing magic' is, say, an IR phototransistor going into saturation, triggering the first of a pair of cascaded 555's... If I've got a handful of specs and datasheets, I can look and say, well, this address line has an eight picosecond capture window, but this controller has a ten picosecond rise time - ain't working unless we can 'fiddle' one or the other; barring that, what have we? "It should work?"
And, I still confidently maintain, that for the vast majority of users, with the vast majority of applications, they'll never 'see' the difference! With a very few exceptions, the average memory access will statistically almost always be to 'correct' a cache problem'; either a prediction 'miss' ("I haven't got the 'chunk' requested by this branching..."), or a 'dirty' page/line ("I gotta write this out before someone else screws with it..."); these accesses are almost always waiting on one or the other of the latencies to pass, and are almost never dependent on raw throughput speed.
More info. Here's an HWMonitor output:
Voltage 0 1.18 Volts [0x4A] (CPU VCORE)
Voltage 1 1.63 Volts [0x66] (VIN1)
Voltage 2 3.34 Volts [0xD1] (+3.3V)
Voltage 3 4.97 Volts [0xB9] (+5V)
Voltage 4 9.79 Volts [0x99] (+12V)
Voltage 5 -12.48 Volts [0xC3] (-12V)
Voltage 6 -9.15 Volts [0x8F] (-5V)
Voltage 7 3.63 Volts [0x87] (+5V VCCH)
Voltage 8 3.12 Volts [0xC3] (VBAT)
Temperature 0 35°C (95°F) [0x23] (TMPIN0)
Temperature 1 50°C (122°F) [0x32] (TMPIN1)
Temperature 2 49°C (120°F) [0x31] (TMPIN2)
Fan 0 1361 RPM [0x1F0] (FANIN0)
Power 0 130.00 W (Processor)
Temperature 0 61°C (141°F) [0x27] (Core #0)
Temperature 1 58°C (136°F) [0x2A] (Core #1)
Temperature 2 58°C (136°F) [0x2A] (Core #2)
Temperature 3 53°C (127°F) [0x2F] (Core #3)
I'm concerned about the -5V and the -12V readings. I don't think I'm doing anything that should cause those to read so high. Is this a problem?
OK - finally got back from parts run (water softener, RadioShluck. hardware store, & on & on... ), and got a look at your parameters - here's what I'd do:
First, I don't know what the limit on or 'granularity' of your RAM voltage is, but you're showing a couple hundredths low (which I would expect with 6 DIMMs) - bump it up either that couple hudredths (if you can - might be limited to 1.65, don't worry if it 'turns red'; go for a half-tenth (.05) if that's all the 'granularity' allows...
Change your "Command Rate" to "2T", and she'll likely run 'as-is'... 1T with multiple DIMMs per channel is almost impossible to get...
If still instability, try the previous note re tRFC; you're showing 74, might wanna try 82, and then 'walk it down' two at a time 'till instability returns...
Note that the tRAS, which is listed in the XMP as 20, is more normally set, as shown, to CL + tRCD + tRP; some fast RAM can 'squeak by' lower, but, again, with more than a DIMM per channel, it's 'risky business'...
So the Voltage 6 -9.15 Volts [0x8F] (-5V) didn't bother you at all? That's way out of spec.
I disconnected the PSU and put a VOM to all the connections and they all came to within tolerances (using info from here http://pcsupport.about.com/od/insidethepc/a/atx-pinout-24-pin-12v-psu.htm). The only ones I didn't test (because they're not color coded and I didn't have a mapping of the pins when I tested it) were the SATA connectors. I did test all the modular outputs with a molex so I know 4 of them on each output are okay. Still need to test the other 2.
I'll do what you suggested above and report back tomorrow. I switched some drives around so I have to reinstall Windows again. And then again later in the week when I get the SSD's I ordered today. Kind of excited about that. Thanks for all your feedback on this. Very much appreciated.
Always !Quote:So the Voltage 6 -9.15 Volts [0x8F] (-5V) didn't bother you at all? That's way out of spec.
Nah - the measurement technique itself is obviously FUBAR'd; I couldn't even tell you offhand how this stuff is sensed - I think the LPCIO does some of it, god only knows about the rest That's why I posted mine - they never have been able to even come close!
Trust me - if any actually used rail voltage was actually that far off, your computer wouldn't have a chance in hell of ever starting up, much less running, and the 'magic smoke' would probably be leaking out the back, somewhere!
Quote:Change your "Command Rate" to "2T", and she'll likely run 'as-is'... 1T with multiple DIMMs per channel is almost impossible to get...
I see where to set the Command Rate for all 3 channels but there's no "T". I can set it to 0, 1, 2 or 3. Is that it?Quote:First, I don't know what the limit on or 'granularity' of your RAM voltage is, but you're showing a couple hundredths low (which I would expect with 6 DIMMs) - bump it up either that couple hudredths (if you can - might be limited to 1.65, don't worry if it 'turns red'; go for a half-tenth (.05) if that's all the 'granularity' allows...
I think I cannot change it. No matter what I set, the value of 1.65 remains dimmed. Does that sound surprising?
EDIT 1: I took out the 2nd bank of RAM, reset the BIOS and made a couple of small changes. Prime's been running for over 9 hours in blend and is just fine. Temps are fine. Going to run it a while longer then swap the RAM for the other set and do the same. At least then i know most everything hardware wise is okay except for the slots for the 2nd set of RAM. Will test all 6 after all that. Wondering about "Orthos". Just now heard about it. Should I run that too? The page I found for it looks old and not updated and doesn't mention Windows XP, Vista or 7. Any other stress testers other than Memtest86 that could be run?
Quote:I see where to set the Command Rate for all 3 channels but there's no "T". I can set it to 0, 1, 2 or 3. Is that it?
Yup - that's the spot; they normally are listed/expressed as '1T', '2T', and 3 is interesting - never seen/used three, have to look in some BIOS and see if it's always been there, & I never knew!Quote:EDIT 1: I took out the 2nd bank of RAM, reset the BIOS and made a couple of small changes. Prime's been running for over 9 hours in blend and is just fine.
I'd expect that - the RAM appears to be set up correctly, just needs to be 'twiddled' a bit for the extra DIMMs; even with 775's, using four DIMMs (again, two per channel) has always taken a bit of adjustment...Quote:I think I cannot change it. No matter what I set, the value of 1.65 remains dimmed. Does that sound surprising?
You may need to disable XMP for this to be available, or you may be looking at the item displayed on the "Advanced DRAM Features" sub-page, rather than the " Advanced Voltage Control" sub-page, where you need to be to adjust it. Let me know - if we need to disable XMP, it may take a bit more fiddling - might set itself to 9-9-9-24, and need to be manually adjusted back to 7-7-7-21...
Well I got 11.5 hours out of this Prime95 run. It just crashed and restarted. Is there a log for Prime that shows what happened? I found results.txt but that just shows all the things that passed. Not very useful IMO. I can't find any sort of log file. I tried to find txt and log files but the search capabilities in 7 are not familiar to me. I LOVE the search feature in W2k and had to turn off indexing in XP to get it to work right. I hate all the dumbing down of these things. I VERY often need to search for specific file types containing specific strings of text (for example, if I'm programming in Delphi, I want to find all the dfm files containing "active = true". Sorry for the digression here but how is this done in 7? In XP you have to completely turn off the very slow indexing feature to allow XP to search inside of any file type you like otherwise it ignores many file types. Sorry for the rant.
Quote:Is there a log for Prime that shows what happened?
Not sure, but I'll see if I can find out... Never had the need - usually, if I'm getting failures, I have a pretty good idea who the 'culprit' is
Seven's search 'just works' for me - I leave indexing on, and I have a number of 'canned' overclocks that contain "Balls2Wall" - just typed it into the search line, and it 'popped' the two dozen or so files... When I'm doing 'fussy' work, for programming, I tend to use GREP...