I have to say this is easier than expected! Took me a little while before I got used to the AMD terms and variables used now. Not 100% on the various power saving and dynamic clocking options though. I haven’t found how to disable the idle power saving yet.
Currently in process of seeing how far the CPU itself overclocks. I’ve been using other lower multipliers so that nothing else is overclocked. At this moment I’m stability testing a mild overclock to 3360 MHz (+20%). The thing is, however the voltages work, it turned out I’m undervolting it with a setting of 1.25, where I think the nominal is 1.4 (actual 1.312). So the system power consumption is actually down a lot from stock!
3360 MHz 1.25V 176W
2800 MHz 1.31V 194W
Is this a crunching dream or what? Much more tinkering needed to see how far this goes!
I was reading up on cpu’s the other day as your crunching itch thread got me looking, 1055t generally will push very close to 4ghz stable! from what i saw this was pretty much the norm but very soon after 4 like 4.1 was unstable in a lot of cases. I have also been reading up on the motherboard you got, i know it’s difficult to judge how good is good but is the onboard graphics any good? (i’ve read that it’s overclockable too?). It’s not got a lot to beat in my case as i’m upgrading from an old radeon x1950. i was looking at putting the motherboard with either a 2 core 555 (supposedly able to unlock 1 or 2 more cores based on luck) or the 635 processor (4 core).
Whats your verdict so far on bios settings etc, generlal speed of things, advice on ram (guessing as much of the best quality you can afford?)
Oh and the important bit whats its dc output like (i’m guessing your still better of crunching using a cuda based graphics card)
Running at 4060 MHz now Needed 1.4V and temps are rising, so I’m not going to push further than this. 1.35V was ok for 3920 MHz though.
It might be easier if I do this:
2800 200x14 1.32V HTx10 194W 37C
3080 220x14 1.25V HTx9 169W 34C
3220 230x14 1.25V HTx8 172W 35C
3360 240x14 1.25V HTx8 176W 35C
3500 250x14 1.25V HTx8 184W 36C
3640 260x14 1.35V HTx7 217W 40C
3780 270x14 1.35V HTx7 224W 41C
3920 280x14 1.35V HTx7 233W 41C
4060 290x14 1.40V HTx7 267W 44C
Temps are under prime95 large FFT stress test, and stable for the short times I run at each step. Power measured at mains. I think I’ll go back to 3500 at 1.25V for normal running if it passes every stress test I can throw at it. The above can be taken as indication, as I’ve not optimised other settings like ram etc. yet.
The mobo seems fine. I haven’t tested the integrated gfx yet, but can put it through the usual 3D benchmarks later.
Really useful as I’m planning on getting either one of these or an i5 760 next month
Silly question - what does HT stand for ?
HyperTransport is a link used between CPU and chipset. As I’m fsb overclocking I turn it down to keep it roughly in spec. I should add that the northbridge clock is set to the same multiplier in my testing so far. Some suggest there is a gain from running it higher.
While I’m here, I have some 3D benchmarks. 256MB ram allocated, default GPU clock I think is 500 MHz.
There’s more to this overclocking lark than I first thought :s Not just about the CPU these days :s
It’s not that bad… (assuming you don’t have multiplier adjustment)
Old school overclocking:
Adjust ram frequency ratio as needed
Increase some bus clock or other (can’t remember name)
Adjust ram frequency ratio as needed
Adjust Northbridge multiplier as needed
Adjust HT multiplier as needed
I keep it simple by just turning the last two down to around nominal levels with everything else clocked.
Ooh, have some more 3DMarks…
Should point out all benchmarks are with CPU at 3.5 GHz.
I’m settling on a configuration for long term running now… to summarise the changes from normal:
Turbo core disabled
C1E support disabled
HT at 8x
NB at 10x - seems stable, doing quick benchmark now to see if it has any significant impact. Note earlier benchmarks were all obtained with this at lower multiplier.
Bus 250 MHz
Ram at 533 multiplier, for 666 actual (1333 effective). Timings are 7-7-7-20.
Voltages at mobo defaults, which sets the CPU to 1.25V.
Think I’ll do screenshots once done.
The CPU itself. Marking in case it means anything to anyone looking to get a similar one. I have to wonder if all modern X6s might be “95W” like even if not sold as one given how easy it was to run this at high clocks and low voltage. Or would an actual 95W model be even better?
The settings I’ve picked for normal running.
Mobo details. Not shown here, the PCB of the mobo I got is rev 1.1.
The ram… I’m using kingston rated at 1333 7-7-7-21 at some increased volts. It seems to run at that ok even at 1.5V. Unganged operation as mobo default. There doesn’t seem to be much definitive testing out there, but it should be better for multi-thread operation. There’s also some weirdness on the reported memory speed. The BIOS POST shows 1066, as does memtest86+, but inside bios itself and CPU-Z here show 1333. I can only guess the earlier ones are reporting the nominal setting, but doesn’t take account of the increased speed from overclocking.
Not a lot to see here… I think the BIOS reported the core as 500 MHz though, so is it underclocking itself for desktop use?
Win7 score. The CPU score at stock was 7.4. Note this was run with the gfx ram set to 128MB as I wanted to maximise the ram for tasks.
I think the 95W versions of the CPU are only allowed to be sold in a pre-built system, or possibly a bundle. I believe AMD are stopping vendors selling the chips as OEM
Yup, the 95W availability was covered in the other thread. But the thing is, I’m wondering if they’re really much different from the 125W version out now, as opposed to any earlier 125W version. As in, they might be so good now they’re almost the same as 95W even if not sold as such. The headroom in it is Phenominal (bad pun intended) seeing as I’m running it at 3500 at lower voltage than AMD wants for 2800.
As part of research, I’ve been looking at reviews of the X6 written when it came out. With those early samples, reviewers seemed to need a fair voltage bump up and a bit of work to get 4 GHz out of it. I did that on this sample hardly breaking a sweat. So either I have an exceptional sample, or they’ve improved. It’s never easy to draw good conclusions from 1 sample…
Trying a mild overclock on my 945 (3GHz), just upped the FSB to 213 and haven’t altered anything else and it’s now running at 3.196GHz quite nicely!
Sure I wrote it somewhere but this box became unstable, that even a return to stock settings didn’t help. Well, I might finally have found the problem. Hard disk. Or did I try replacing that before? The disk in it has died. Maybe previous crashes were from occasional access attempts to bad bits. It is click-of-death bad now.
Now to prove it, I’ll need to find a spare HD, and put a new OS on it…
One of my systems had a habit of crashing… tried everything, even went as far as replacing the boot SSD, which improved it slightly but didn’t cure it… ended up doing a fresh install on the new SSD and it has worked perfect since. Found the original SSD was faulty after running some diagnostics on it.
I’m at the wait and see stage now. I did OS reinstalls which didn’t help, but if I used the same disk, that could be the problem. I’m really digging deep now. Using an 8 year old WD SE16 320GB hard disk. No idea where any of my old install disks are so put on Windows 10 preview. Sadly amusing watching task manager show the hard disk is usually the limiting factor.
I have some spare HD that have not been used much if at all if you wanted one?
I think I’m ok to keep rolling as is, since this is a cruncher only. The slow hard disk is only noticeable during install or other tasks so not a big deal.
Nice to see someone else running an old Phenom II X6… down to my last 1090T, completely stable @ 3.9 (19x205) on an old ASUS Crosshair IV (890FX)
I’m running at stock for now, seems ok. This weekend I’ll try putting it back to the old overclock of 3.5 GHz. I did quickly try but it crashed on boot. I think I got the voltages off but the bios settings aren’t terribly helpful.