Nice work :up:
Printable View
Nice work :up:
I know we have all spent hours testing, but I have never seen 50%LLC with additional overvolting to meet the minimum voltage requirement do better than 100% LLC. Maybe it's a chip to chip thing... But I've tested so many chips that I find it hard to believe.Quote:
Originally Posted by CryptiK
This weekend I'm spending the holiday breaking down my rig to replace _all_ of the fans. Currently I'm running a mixture of Enermax Magma UC-MA12's and Yate Loon D12SH's. I'm replacing every fan in my case with a Gentle Typhoon AP-15's. I'm also going to clean my radiators etc.
Once I get it all back together, I will try to stabilize a 4.6GHz run with LLC set to 50%.
That's exactly what I was getting at. Just needed a engineer level explanation. Thanks :up:Quote:
Originally Posted by CryptiK
Don't get me wrong Zoson, when I first started overclocking I used 100% CPU LLC and QPI LLC with my 920. I pulled off some nice clocks with it. But when I tried to fine tune my overclocks, and try to lower voltages as much as possible, I saw that not using LLC always allowed me to drop voltage one notch, whether Vcore or QPI. I know the downside is higher voltage at idle, but it's worth it to achieve certain overclocks.
I think those who have not tried should give it a shot. Looking forward to seeing your guys' results.
P.S. Zoson - I wish I can convince everyone out there who buys fans to choose Gentle Typhoons. Have you ever used them before? Great choice I must say, as I use the same myself. I have to say they are by far THE BEST fans I've ever used. Before realizing my 920 was just a high heat leakage chip I went through so many different fans. GTs excel in sound, static pressure, and quality. When you see them in person you can tell they are 'real' fans, what fans are supposed to be. All other fans you have used up until now will seem like toy fans. You'll see ;).
Very nice indeed! Let's see some Vantage or SPI32M runs :) (or even LinX if you dare)Quote:
Originally Posted by paill
I'll take it thats the 4 Gb sticks of the same ram that I have (which you PMed me a link to buy it from, but the heatspreaders are too tall :().Quote:
Originally Posted by XII
I am envious, I want 4 gb low profile modules of this stuff. TBH I would assume that any new modules that they sell wont have the same modules, it will likely be the new current cheap modules advertised in a premium line that do 1800 Mhz cas 10 instead of cas 7 :(
No problem :up:Quote:
Originally Posted by spoolindsm127
I was close to buying GT's but I went with Scythe S-Flex 1600's which are a pretty nice fan, never seen/used a GT but have heard good things about them so long as you dont mount them horizontally (increases bearing noise/wear).
The Geil`s i have are: GUP312GB2000C9TCQuote:
Originally Posted by bhavv
http://www.geil.com.tw/products/showSpec/id/293
wk29/2010
The one`s I linked are: GET38GB2000C9ADC
(Old stock)
Can`t find them on the site, but they came out in the same period.
Same time, Gskill came out with their Tridents,
http://www.xtremesystems.org/forums/...-12GBTDD-3x4GB
I just finished a full cleaning of my system and replaced all my fans. I'm even better off now, I have push/pull on every one of my 3 rads! Before I only could fit push/pull on two of the three... Re-arranged my internals and it fits PERFECTLY. Literally no space to spare, at all!!!Quote:
Originally Posted by spoolindsm127
Before:
http://i144.photobucket.com/albums/r...h_5032dc6f.jpghttp://i144.photobucket.com/albums/r...h_52141184.jpg
After:
http://i144.photobucket.com/albums/r...h_5f293551.jpghttp://i144.photobucket.com/albums/r...h_c3d1a667.jpghttp://i144.photobucket.com/albums/r...h_3106257c.jpg
Gentle Typhoons are AMAZING. Take it from someone who has tried all the other good fans. GT's are hands down the best. Skinnee said that the YL D12SH were in the same class... But they are not even close. NOT EVEN CLOSE.
If you are looking for new fans... Don't go cheap... GET GT's AND YOU WON'T REGRET IT!!!! I'm so happy.
http://i144.photobucket.com/albums/r...h_43dce274.jpghttp://i144.photobucket.com/albums/r...h_0736362b.jpghttp://i144.photobucket.com/albums/r...h_4865651e.jpg
http://i144.photobucket.com/albums/r...h_da85e602.jpghttp://i144.photobucket.com/albums/r...h_85dd4cd1.jpghttp://i144.photobucket.com/albums/r...h_40db5a8e.jpg
http://i144.photobucket.com/albums/r...h_5ca27f27.jpghttp://i144.photobucket.com/albums/r...h_9ff8ccec.jpghttp://i144.photobucket.com/albums/r...h_03d7608e.jpg
http://i144.photobucket.com/albums/r...h_51297c18.jpg
These ones would have been perfect, but as usual they were never available in the UK. The only retail stock of Geil modules available here were the value range (mine were OEM sticks) :(Quote:
Originally Posted by XII
That`s some serious dust, :yepp:.Quote:
Originally Posted by zoson
How long did it take to collect?
Remembers me, of my smokin`days:smoke:
That is about 4 months of buildup. I last took my pc apart on September 8th.Quote:
Originally Posted by XII
4 months, must be a dusty environment.
yup, new york is a very dirty place :(
I'm wainting a R3E and X980 roxxx for the next week.
:welcome:Quote:
Originally Posted by Le Titi
obsolete 980X
I just got my RIIIE today :D and it has 902 bios installed (on both ROMs I think). I don't have time to OC at the moment but will see what I can get with this lovely 990X as soon as I update the BIOS. Any agreement as to which is the best?
I've forgotten how nice it is to see all these voltages on desktop. Even if they aren't 100% accurate, it's sure better than nothing LOL.
Put 1502 on BIOS1 and 0003 on BIOS2.Quote:
Originally Posted by LennyRhys
0003 has additional memory controls that is not in any other bios.
1502 has the updated intel orom.
Better late than never hey Lenny ;) I run 0003 24/7
Heh :D I'll have a look at 0003. I've got 1502 on my pen drive ready for flashing but seeing that the board came with 0901 on #1 and 1102 on #2 I thought I'd have a play and get familiar with the BIOS without going in too deep. 0901 sucked... couldn't even POST at 188x25; but 1101 was perfect and I've been able to do some amazing benching with it.
Managed to get 18126 on Geekbench with 5GHz (151x33) using 4240 uncore and 2120 memory (8-9-8-24-88-2T), with 1.425 VTT and 1.525 vcore. 1.4125 VTT was just a little too low and I got a blue screen about halfway into the test...
Compared to the P6X58D-E the voltage behaviour of the R3E is VERY different - it really jumps about under load with LLC at 100% and overshoots the BIOS value by a considerable margin, which I didn't expect; and the fluctuation under load is really bad, going from 1.434-1.456 in LinX - my other board was rock solid under load, never changing in CPU-Z. I'm not quite sure how to access certain BIOS options, eg QPI LLC, which is greyed out near the top of the tweaker menu... but that will come later.
In any case I'm very happy with the board but, unfortunately, my GFlops problem (peculiar to LinX, interestingly) has returned - getting a good 10 GFlops lower than I should be and it's REALLY annoying, because I just managed to sort it out on the other board. I've already got a 4.7GHz HT notch on my LinX belt but the memory and uncore were low so GFlops are only at 86... I did get one amazing run with 89 GFlops consistently but it crapped out on loop 20 :rofl: (you just have to laugh at the irony).
This time round I'm completely flummoxed. At least the last time the problem presented itself in other software, eg low Everest and Sandra memory bandwidth... but this time everything checks out (including Geekbench, which is what alerted me to the poor memory performance in the first place).
I've left all CPU features enabled, as before, and in addition to that I've enabled C-States for the extra bandwidth, but no change. Will give BIOS 1502 and 0003 a go tomorrow and see how they check out, and then I can play around with PLL and LLC and see what I can do with 4.7GHz HT. :up:
OK I flashed #1 to 1502 and at reboot I got a "bad bios checksum" error... put the CD back in and now it's back to 0401.
Tentatively flashed #2 to 0003 and to my surprise it worked first time and I'm using it now. I'll try reflashing #1 to 1502 again and hope for the best.
EDIT: just tried 1502 a second time and it worked, so now I have two good BIOS versions to play with. Cheers guys. :up:
NP you'll figure it out. QPI LLC you need to enable a jumper on the mobo then you can use it. I just leave it off though.
Ah thanks Cryptik - I'll just leave it off at the moment. Any idea how much QPI/VTT undervolts/overvolts compared to BIOS value, or does it vary between boards? The board is a bit on the big side for my case (damned Asus calling it "ATX"...) so I can't access the voltage measure points but I'm really keen to get DMM voltage readings so I'll try to work something out there. As far as stability overclocking goes, I'll give 0003 a try first. :up:
So I stuck with 1502 (for the moment) and decided to do a bit of relatively tame IMC testing with 3200MHz uncore and 1600MHz 6-7-6-18-60-2T mem and I gotta say it's been a royal pain in the butt trying to get it stable with this board - it's taken me about six hours of repeatedly coming back to a "red" LinX because QPI/VTT wasn't high enough. I'm used to setting 1.275-1.3v on my P6X58D-E, but oh no, this board has different ideas.
I got some really sweet runs (even though they did crap out) - check the consistency of GFlops here:
QPI/VTT: 1.325 (BIOS), 1.28 under load in HW Monitor
(Diagnostic mode, pagefile disabled)
http://img560.imageshack.us/img560/7537/4ghzerr2.png
Part of me thinks that LinX is just tormenting me, because it knows how much I loathe it. :rofl:
And now onto my first successful run with 20 passes. Given the consistency of the above run, I have NO EARTHLY IDEA why GFlops fluctuated so much in the only run that actually passed. The first time I started LinX I got 49 GFlops on the first pass... so I just closed it immediately and fired it up again without rebooting. Like I said, it hates me lol.
QPI/VTT: 1.35 (BIOS), 1.30 under load in HW Monitor
http://img843.imageshack.us/img843/9260/4ghz.png
One thing's for sure: the DMM is coming out tomorrow. Cryptik, any reason why you don't use QPI LLC? There's no way I'm leaving it disabled if droop is as much as 50mV.
I'm also really confused because I was able to bench 4240MHz uncore with 2120MHz C8 memory (at 5GHz no less) with "only" 1.425 VTT in BIOS, which means it was getting about 1.375 under load... yet it wants 1.3v for a measly 3200mhz in LinX?
But I'm definitely not complaining, coming from an i7 920 that needed 1.4 VTT for 3200MHz :down:
Edit: I think 1.3375 (bios) is as low as I'll get VTT without LLC for 3200MHz with tight memory. Not sure if a higher QPI Link freq would need more QPI volts or give better performance, but I left it on "Auto" in BIOS (mostly out of negligence tbh). CPU is at 1.2000 in BIOS and was 1.197-1.204 in CPUZ; Northbridge actively cooled by two silent 120mm fans stacked, which dropped its temp by a good 15C.
Still getting weird GFlop spikes but at least this time it all stayed within 1 GFlop of 77 :up:
CPU: 1.2000
QPI/VTT: 1.3375
http://img545.imageshack.us/img545/9017/4ghzt2.png
Just did some BIOS/software/DMM comparisons and the results are interesting, especially with QPI/VTT.
http://img52.imageshack.us/img52/9178/dmmtesting.png
First and foremost I'm surprised by the innacuracy of the software voltage readout - there was A LOT of vcore fluctuation in CPUZ during LinX testing, from 1.197v spiking as high as 1.224v sporadically. With the DMM the vcore remained solidly on 1.187v throughout, dropping to 1.184v near the end of each loop.
As for QPI/VTT, the actual load value is nearly 60mV lower than the BIOS value, which explains why I could set VTT so low on my previous board. Vdroop aside, vdrop alone is particuarly bad at almost 35mV, compared to only 16mV on the vcore. RAM is a solid 21mV below BIOS value and didn't fluctuate during testing. Impressive. :up:
I'll test QPI/VTT again with LLC enabled and see how I get on. If you guys have a DMM handy and haven't already tried it, I'd suggest giving it a go as the real voltage values can be quite different from what BIOS or software tell you. Cryptik - your 4.5GHz 1.33v submission might be as low as 1.31v under load :shocked: :D
I use qpi llc and 100% calibration on the cpu. I have never seen a situation that it helps to have them off/lower, personally.
As of now I will also be using both. :) I did more DMM testing with QPI LLC enabled and with 1.3000 set in bios, vdrop is only 9mV and vdroop a further 7mV, so 1.284 under load. Brilliant.
And I actually love this board so much. I re-ran my 4.7GHz HT LinX run and it was going good at 90 GFlops until about 20 mins. All I needed to do was lower CPU PLL to 1.7v and it passed with flying colours. Tried enabling C-States for the extra bandwidth but LinX was having none of it, so I just left them off. Measured CPU, QPI and RAM with the multimeter. BIOS values were 1.45, 1.36875, and 1.67 respectively.
4.7GHz HT (142x33), C-states disabled BIOS 1502
CPU: 1.437 real
PLL: 1.7
VTT: 1.352 real
RAM: 1.64 real
All other voltages stock
http://img692.imageshack.us/img692/889/4700crop.png
Looks like I've found a killer daily overclock. :up: Anybody want to try beating 90.6 GFlops with 3700MHz uncore? :D