Asrock G965M-S
Intel E-1200 (think E-2140 with half the cache)
2x2G Corsair XMS2 DDR2-800
Sparkle/FSP 400 watt psu
using integrated video for now
This hardware is not installed, it's just sitting loose for testing with a cdrw connected so I can use the UBCD.
The ultimate purpose of this is not definite yet. It may go into a htpc case with a low end ATI card. Or not. I want to see where the results point me.
I had this cpu at 3G outside a case on reasonably low voltage in a Gigabyte P35 board a couple of years ago, and my son ran it at 2.8G, I think, in a gaming rig. I might have to run it close to stock and undervolt but hubris is pushing me to see what I can get out of this. The bios on this cheap ($45Cdn at NCIX) Asrock board is surprising, and much of the reason for this post. It's all there, all of it: voltage control (yes, undervolting), ram ratio and timings. There is so much it's intimidating and confusing.
I jumped in and tried a 240 fsb resulting in 1928MHz, but 260 and over produced no boot until I changed more bios settings. It's now at 250 fsb and 1968Mhz according to the bios and 1959.2 at 8x244.90 according to a utility from UBCD. It ran 28 min of cpu stress from UBCD with only a 1C increase in cpu and NB temps. I suspect there is more to be had from this board, though never close to what the P35 could allow, and it must be some bios settings I don't understand. Here's where it sits now:
cpu/PCIe Async
cpu fsb: 260
cpu vcore: auto, but reads 1.384 in HW monitor in bios
ram: 346
timings: 5 5 5 15 the rest left on auto (SPD is 4 4 4 12)
ram flexibility: enabled
boot failure: enabled
Speedstep: disabled
ram voltage: on auto, 2.201, reset to 1.908 after boot at 250fsb
halt (C1E): disabled
NB voltage: auto, reads 1.319V if switched to manual
There must be something I'm missing, and I suspect it could be northbridge or graphics chipset related. And I thought SPCR was the best place to ask. Most forums would tell me to get another board. I wanted to see what I could get out of a cheap one, and I already had the cpu. And given my SPCR tendencies, as mentioned, I can always use the integrated video and undervolt everything if I have to run it close to stock.
I searched the forums and found indications that someone had reached 280 on a G965, but no definitive info on all the bios settings used to get there.
Overclocking on G965
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
well, i got the same board.
It's got a Cel 420 and 1g ram, and is currently running at 8*[email protected]
I can't get it over 333 FSB, though....
IDE and SATA drives become crazy if I set more than 340 FSB...
I think it can't actually lock the PCIE's speed propertly...
Tried to lock it to 100, but will not boot once the FSB is set more than mhz xD
It's got a Cel 420 and 1g ram, and is currently running at 8*[email protected]
I can't get it over 333 FSB, though....
IDE and SATA drives become crazy if I set more than 340 FSB...
I think it can't actually lock the PCIE's speed propertly...
Tried to lock it to 100, but will not boot once the FSB is set more than mhz xD
Proves I've got something set wrong in the bios, but I can't tell which one. If anything in my settings above seems off, please comment. Maybe you could take a look in your bios if you have a chance?caragua wrote:I think it can't actually lock the PCIE's speed propertly...
Tried to lock it to 100, but will not boot once the FSB is set more than mhz xD
Hi there, people (and especially Danceman, got your PM and I think I can help you somewhat).
First of all, from my experience with ASRock motherboards, especially older models, I can tell you they are EXTREMELY picky when overclocking. I have two identical ConRoe945G-DVIs (BIOS-modded to the ConRoe1333 version), one with an E1200, other with an E4300, and while the E1200 can easily bump to 266MHz FSB (without a dedicated VGA card and with a lousy OEM PSU), the E4300 (with a dedicated GPU and branded PSU) just won't budge 1MHz without failing the overclock settings... Oh, well...
So, you should keep that in mind in the first place when fiddling with an older model (newer ones seem more OC friendly, though I don't really have any hands-on approach).
Now, for the REALLY bad news.
The G965 chipset, much like its older sibling, the 945G, has a LOUSY PCIe lock. Which means overclocking will always be a hit-or-miss thing, depending on the PCIe clock generator straps. You might even have FSB holes, like from ~290MHz to 332MHz, where no FSB speed works, just to have 333MHz works out of the blue (the PCIe bus resets to 100MHz at this strap, or it should reset). I, for one, have experienced this kind of behaviour on my 945G-based boards (though only on the ~233-265MHz range, the board doesn't seem to like getting a 200MHz FSB CPU to 333MHz FSB... lol).
Also, if I remember correctly, the 965G IGP (1st-gen DX10-capable Intel IGP) doesn't seem to handle high FSBs very well (above ~340MHz, if memory serves me right), so that should be a hard FSB wall...
If direct FSB input works, try going straight to 333MHz to see if it works. If not, you can still try FSB modding your CPU, to try and circumvent the lack of a proper PCIe lock.
Hope this helps.
Cheers.
Miguel
First of all, from my experience with ASRock motherboards, especially older models, I can tell you they are EXTREMELY picky when overclocking. I have two identical ConRoe945G-DVIs (BIOS-modded to the ConRoe1333 version), one with an E1200, other with an E4300, and while the E1200 can easily bump to 266MHz FSB (without a dedicated VGA card and with a lousy OEM PSU), the E4300 (with a dedicated GPU and branded PSU) just won't budge 1MHz without failing the overclock settings... Oh, well...
So, you should keep that in mind in the first place when fiddling with an older model (newer ones seem more OC friendly, though I don't really have any hands-on approach).
Now, for the REALLY bad news.
The G965 chipset, much like its older sibling, the 945G, has a LOUSY PCIe lock. Which means overclocking will always be a hit-or-miss thing, depending on the PCIe clock generator straps. You might even have FSB holes, like from ~290MHz to 332MHz, where no FSB speed works, just to have 333MHz works out of the blue (the PCIe bus resets to 100MHz at this strap, or it should reset). I, for one, have experienced this kind of behaviour on my 945G-based boards (though only on the ~233-265MHz range, the board doesn't seem to like getting a 200MHz FSB CPU to 333MHz FSB... lol).
Also, if I remember correctly, the 965G IGP (1st-gen DX10-capable Intel IGP) doesn't seem to handle high FSBs very well (above ~340MHz, if memory serves me right), so that should be a hard FSB wall...
If direct FSB input works, try going straight to 333MHz to see if it works. If not, you can still try FSB modding your CPU, to try and circumvent the lack of a proper PCIe lock.
Hope this helps.
Cheers.
Miguel
Many thanks for the quick reply, Miguel.
While the chipset is older, the board itself is fairly new; first bios release is dated 4/15/2009. It's only on a second bios release, and that was only to change two seemingly minor default settings.
I'll try 266, and then try 333. I looked over the other settings and will leave them where they are, except to change the ram voltage back to auto when I try 333.
While the chipset is older, the board itself is fairly new; first bios release is dated 4/15/2009. It's only on a second bios release, and that was only to change two seemingly minor default settings.
I'll try 266, and then try 333. I looked over the other settings and will leave them where they are, except to change the ram voltage back to auto when I try 333.
Tried 266 -- no boot. So I tried again with the Overclock Mode set to [CPU, PCIE, Sync.] and it booted. I'm positive I tried this option before, at something like 260 with a no boot result. By no boot in these cases I mean no display. Fans run and I can hear the cdrw cycling as well. Cpu is at 2169 according to the bios display.
I was able to get it to 290fsb with loosened ram timings and voltages on auto, and it ran the cpu stress test from UBCD for 12 minutes. But attempting to fine tune the memory timings and voltages returned it to no boot. Increasing the ram voltage all the way above 2.0V didn't work; loosening the timings again did. But reducing the ram voltage and running Memtest produced a screen of red. Increasing ram voltage up to 2.1V didn't change it. Afraid for my ram I set everything back to defaults, cpu etc., and ran Memtest to make sure I hadn't damaged it.
I have a lot of work for nearly a week and won't be able to try anything more during that time. Not being an experienced overclocker I'm wondering where to go from here. I don't want to run really high voltages on either the ram or the cpu. Syncing the fsb and PCIe allowed the overclock, but the limit comes when the PCIe climbs too high. At 290fsb it's at 109. I'm wondering if the ram troubles I'm finding are due to the high PCIe, and if dropping the fsb to 266 or not too far above, which drops the PCIe closer to 100, would eliminate the problems with the ram.
Any suggestions from the experienced overclockers gratefully accepted.
I have a lot of work for nearly a week and won't be able to try anything more during that time. Not being an experienced overclocker I'm wondering where to go from here. I don't want to run really high voltages on either the ram or the cpu. Syncing the fsb and PCIe allowed the overclock, but the limit comes when the PCIe climbs too high. At 290fsb it's at 109. I'm wondering if the ram troubles I'm finding are due to the high PCIe, and if dropping the fsb to 266 or not too far above, which drops the PCIe closer to 100, would eliminate the problems with the ram.
Any suggestions from the experienced overclockers gratefully accepted.
Ups, I forgot to mention that little piece of information. I was replying late last night, after a very long day (read: week), so something slipped.DanceMan wrote:Tried 266 -- no boot. So I tried again with the Overclock Mode set to [CPU, PCIE, Sync.] and it booted. I'm positive I tried this option before, at something like 260 with a no boot result. By no boot in these cases I mean no display. Fans run and I can hear the cdrw cycling as well. Cpu is at 2169 according to the bios display.
It does seem that overclocking with PCIe lock-limited chipsets on ASRock boards should be attempted with Synced FSB and PCIe speeds. That way the PCIe bus flows freely, and drops back to 100MHz more easily (like when you hit 266MHz on the FSB side). If you Async the buses, the PCIe bus will be all over the place because of the faulty PCIe lock, and more OC failures are bound to happen.
However, 100MHz on the PCIe lock is rather difficult. Especially on higher FSB speeds. The 945G-based motherboards only seem to be able to hit 333MHz with the PCIe bus set at 115MHz (which is around 5~10MHz short of potential data corruption issues on the SATA ports). If you try 333MHz, try a manual PCIe speed.
Also, if there is a "PCI fix" option, try enabling it, since ASRock tried to implement some sort of behavioral control through BIOS on the secondary buses.
Hmm, you're right, this is a rather new motherboard. Though I don't really know why ASRock decided to release it, honestly... Unless it's because of the 533MHz FSB compatibility (which at least some 3x and 4x series chipsets can handle unofficially), I'd still prefer a G31-based board. Even with the old ICH7 southbridge...
Cheers.
Miguel
You have to be obsessive to do this. 
I finally got back to this and did another round of testing:
Bios to Performance
CPU PCIE Sync
CPU Freq to 267
DRAM freq to 355/711
DRAM timings Auto (4-4-4 etc)
Then I played with voltage, lowering CPU below the default 1.325V and ram to 1.794V. No boot. I left ram voltage on Auto and chased the CPU voltage up to 1.3500V, all without a boot. Left it on Auto and still no boot.
For some reason on the first time through the bios the Auto ram timings were a looser 5-5-5 etc. I'm going to try loosening the ram timings below defaults even though it's clocked slower at 711 than its stock 800 and see if that helps. I'll use 266 instead of 267 just in case that makes any difference. Current objective is to use a 266 fsb because of the PCIE lock issue and manually adjust the cpu and ram voltages to the minimums that pass stress testing, then see if enabling Speedstep works.
I finally got back to this and did another round of testing:
Bios to Performance
CPU PCIE Sync
CPU Freq to 267
DRAM freq to 355/711
DRAM timings Auto (4-4-4 etc)
Then I played with voltage, lowering CPU below the default 1.325V and ram to 1.794V. No boot. I left ram voltage on Auto and chased the CPU voltage up to 1.3500V, all without a boot. Left it on Auto and still no boot.
For some reason on the first time through the bios the Auto ram timings were a looser 5-5-5 etc. I'm going to try loosening the ram timings below defaults even though it's clocked slower at 711 than its stock 800 and see if that helps. I'll use 266 instead of 267 just in case that makes any difference. Current objective is to use a 266 fsb because of the PCIE lock issue and manually adjust the cpu and ram voltages to the minimums that pass stress testing, then see if enabling Speedstep works.
Loosening the ram timings from 4-4-4-12 to 5-5-5-15-40-5-3 worked. It's now at 276/2207 and passes Memtest and the cpu stress test on UBCD. After that I took the cpu voltage off Auto and worked it down, each time running the stress test. I don't think the PCIE lock is the only bios control that doesn't work. It's now at a 1.2000V setting but as the voltage went down from an initial 1.35V the temp shown in the bios has not changed. The Vcore shown in the bios has come down from 1.384V to 1.272V (set to 1.2125V at that point). I probably won't get more accurate readings until it's in a case with an OS so I can run better utilities.
LOLDanceMan wrote:I don't think the PCIE lock is the only bios control that doesn't work.
But seriously, though that might probably be true (except for the VIA PT880 and Xy8-based boards, ASRock isn't particularly known for OC capabilities), the biggest reason for that OC lackluster is probably the G965 northbridge. That thing is a 945G derivative (though no one will put a hand on fire for that expression), and the 945G can be defined as "quircky" at best. As I've said, I have two 945G-based boards at home (same model, same BIOS), and one of them simply refuses to OC over 1MHz, no matter what I do, while the other one can bump a Celeron E1200 to 1066MHz FSB... Go figure.
So, in short, congratulations on your achievements, but you do really need to be obsessive to OC on such a motherboard... lol
Cheers.
Miguel
Thanks, Miguel.
Still dropping cpu voltage and running a quick stress test from UBCD. Last one was set at 1.125V and read out at 1.184V in the bios. When I have the cpu voltage stabilized I want to take the ram voltage off auto.
I do wish there were better cpu testing tools available that ran outside Win. I doubt this one I'm using is rigorous and suspect it may only work one core. It's so much easier to do this type of testing outside a case to get at the battery and clear cmos pins.
Still dropping cpu voltage and running a quick stress test from UBCD. Last one was set at 1.125V and read out at 1.184V in the bios. When I have the cpu voltage stabilized I want to take the ram voltage off auto.
I do wish there were better cpu testing tools available that ran outside Win. I doubt this one I'm using is rigorous and suspect it may only work one core. It's so much easier to do this type of testing outside a case to get at the battery and clear cmos pins.
Hats off for the dedication. I'd have given up DAYS ago... lolDanceMan wrote:Still dropping cpu voltage and running a quick stress test from UBCD. Last one was set at 1.125V and read out at 1.184V in the bios. When I have the cpu voltage stabilized I want to take the ram voltage off auto.
Well, for memory you have Memtest; as for CPU, I do believe there is a command in Linux that uses the *whole* CPU to produce massive amounts of data, which is promptly outputted to NULL (not a Linux geek, so I can't tell you which command that would be, but I remember reading something about it a while back). That should allow you to test stability thoroughly outside Windows, no?DanceMan wrote:I do wish there were better cpu testing tools available that ran outside Win.
I do feel like the harshest test ANY CPU can be put through in terms of power output is this baby, available for both Windows and Linux. Check it out, if the OS is stable after dealing with that piece of software, you should be just fine.
Careful with that approach. Open test benches usually have a lower overall temperature than closed benches, which can ultimately end up causing instability. Don't forget to keep that in mind *before* putting the system back on its case.DanceMan wrote:It's so much easier to do this type of testing outside a case to get at the battery and clear cmos pins.
Cheers, and good luck.
Miguel
Thanks again. I'm aware that once inside a case I might need to bump voltage up a bit, but I usually pay a lot of attention to airflow through a case. I don't think any of my cases have escaped the cutters, saws and grinders in the basement.
Of course, the next test after my last post produced a segmentation error in the cpu stress test. I went back up to 1.125V to clear that, and tried the Mersenne Prime test from UBCD. That produced an immediate segmentation error, so it's back up to 1.13 and running the Mersenne test.
Of course, the next test after my last post produced a segmentation error in the cpu stress test. I went back up to 1.125V to clear that, and tried the Mersenne Prime test from UBCD. That produced an immediate segmentation error, so it's back up to 1.13 and running the Mersenne test.