CoolColJ's test pics thread

Get feedback from others on your works in progress
ieatfish
Posts: 127
Joined: Sun Jan 31, 2010 5:29 pm

Re: CoolColJ's test pics thread

Post by ieatfish » Tue Jun 08, 2010 11:38 am

dag wrote:i7 at 4ghz at 1.18 vcore at 45c full load? Impossible. Try Prime95 with CoreTemp with hyperthreading on. SpeedFan must be wrong.

1.18 volts and 4ghz is perverse. Can't be stable?? My 860 can't run below that on stock.

Edit It took a lot of testing but we finally found the sweet spot. 1.112v on the core and 3.8GHz. I think you'll agree that this is outstanding.
I missed that, yeah, 45C is very low. You should be idling just a bit below that. Try HWMonitor or CoreTemp.
Intel Core-i7 @ 4.0 GHz | GTX 580 | 12 GB RAM

User avatar
dcm
Posts: 663
Joined: Sun Jan 03, 2010 1:55 am

Re: CoolColJ's test pics thread

Post by dcm » Tue Jun 08, 2010 7:25 pm

ieatfish wrote:
dag wrote:i7 at 4ghz at 1.18 vcore at 45c full load? Impossible. Try Prime95 with CoreTemp with hyperthreading on. SpeedFan must be wrong.

1.18 volts and 4ghz is perverse. Can't be stable?? My 860 can't run below that on stock.

Edit It took a lot of testing but we finally found the sweet spot. 1.112v on the core and 3.8GHz. I think you'll agree that this is outstanding.
I missed that, yeah, 45C is very low. You should be idling just a bit below that. Try HWMonitor or CoreTemp.
Or try to touch mobo near cpu :D

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Wed Jun 09, 2010 5:03 pm

Honestly the heatsink is pretty cold, it's not even warm! Likewise with my memory. The Nh-D14 heatsink is massive and maybe a bit of a overkill :lol:
So in that regard I don't really trust in the Coretemp and RealRemp numbers. Speed fan and the Gigabyte software numbers seem more in line with what I can feel. It is a Gigabyte motherboard afterall :)

It's stable on prime95, but it runs the chip hotter than Indigo. Indigo is about 5 degrees cooler when rendering - strange. And playing games is even cooler still!

For some reason I can't run it at 1.18volts anymore, must have been a fluke that day? My settings are far different now, had to bump the vcore to 1.25volts in the bios, but really stable now. It wasn't that stable at 4.2 ghz and it takes ages to find all the combinations to make it so, plus it's far harder to do so with 12 gigs of ram. And when the motherboard hard locks it takes a lot it get it to post again - remove all ram, disconnect power supply etc . Having said that I did get it to run at 4.4 ghz :shock:

Prime 95 with a variety of temp programs at 4.01ghz - no problems passing. And this is with all my fans at the low setting. It idles around 35 degrees today with those other programs, 18 ambient temp, with the heatsink feeling like it's been in the fridge!
CCJ_Prime95.jpg

renders at 4.2ghz 8)
CCJ_Caterpilla_Indigo@4.2ghz.jpg
Attachments
Clay render_4-2ghz.jpg

User avatar
dag
Posts: 189
Joined: Tue Feb 02, 2010 2:28 am

Re: CoolColJ's test pics thread

Post by dag » Thu Jun 10, 2010 4:12 am

This his why SpeedFan is 15 degrees cooler, and why it's wrong.
http://www.tomshardware.com/forum/25818 ... p-speedfan

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Thu Jun 10, 2010 11:37 am

Looks I need to offset speedfan +14 degrees
Ok, but it's funny how cold my motherboard and heatsink feel after a full run of prime95 :shock:
When my p4 hit 60 degrees you could feel it in the heatsink, but on the Noctua in my i7 it feels like it's been in a fridge, at ambient sub 20 temp!

Anyway I figured out why I could run at 1.18volts on the CPU the other day, I only had one set of 6 gig ram in my system. It is definitely easier to run faster clocks, with lower voltage, higher memory speed and more stable overall with one 6 gig kit. When I put in both kits for 12 gigs total, I have to use more voltage, especially on the QPI, reduce the memory multiplier/speed and lossen the timings. So the memory speed and stability is what holds me back.
I wonder if I really need 12 gigs, as even with 6 gigs, 3x supersampling and lots of subdivision, I still have over 3 gigs left when rendering... maybe I should return a 6 gig kit for credit and out it towards a beefy video card...

You can change the CPU multiplier in the i7 930
So what I did was set my CPU multiplier as high as I can (x22) so that my CPU is at 3.8ghz , and turbo kicks in to put it at 4 ghz or more. That way I can keep my base clock low at around 182, which keeps my memory just a touch above stock speed. There isn't much difference between the different memory speeds, only about 1-2% tops. I thought Turbo was only meant to kick in when only 1-2 cores are been loaded, but it's on for sure with all cores loaded.....

I definitely have one of the better binned chips here.
When the prices for 2nd hand i7 980x drops, then I'll have an Extreme Edition 6 core chip to play with that will allow me to use a higher mulitpliers to negate the ram issue, and an extra 50% of CPU grunt :)
Last edited by CoolColJ on Thu Jun 10, 2010 2:19 pm, edited 1 time in total.

StompinTom
Indigo 100
Posts: 1821
Joined: Mon Sep 04, 2006 3:33 pm

Re: CoolColJ's test pics thread

Post by StompinTom » Thu Jun 10, 2010 1:14 pm

CoolColJ, what kind of RAM do you have? I'm looking to grab a OCZ DDR3 PC3-10666 Gold 12GB Triple Channel Kit so that I can eventually grab another one to end up with 24 Gb/RAM. I'll be matching that with an i7 930 on an Asus board. Do you know if RAM types or even size per stick (i.e. 6 sticks of 2Gb vs. 3 sticks of 4Gb) have an effect on overall performance?

Can't wait to get mine set up and overclocked :)

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Thu Jun 10, 2010 2:31 pm

StompinTom wrote:CoolColJ, what kind of RAM do you have? I'm looking to grab a OCZ DDR3 PC3-10666 Gold 12GB Triple Channel Kit so that I can eventually grab another one to end up with 24 Gb/RAM. I'll be matching that with an i7 930 on an Asus board. Do you know if RAM types or even size per stick (i.e. 6 sticks of 2Gb vs. 3 sticks of 4Gb) have an effect on overall performance?

Can't wait to get mine set up and overclocked :)
Corsair 3x2gb TR3X6G1333C9 6GB KIT DDR3 - I have 2 of these

3 sticks of 4 will definitely be easier to overclock and get more stable vs 6 sticks. The memory speed will alway run at the slowest of all the sticks, so the more you have the more complicated things become.
Especially since yours will be a set kit of 12 gigs which will be tested to run together, instead of my 2 separate kits. Do you really need 24 gigs?! Because I doubt you will have much luck getting a good overclock with that much ram :)

I think total amount of memory as well. I had to loosen my timings when using 12gb, up the QPI and CPU voltage to get rid of the BSODs. Even then I was still getting em when idling. But loosening the memory timing from 9-9-9-24 to 10-10-10-28 seems to have fixed that, fingers crossed!

Do a system drive image before you start overclocking. Overclocking on an i7 is quite a bit more complex than in the past. Memory settings, timings and voltage willl cause all sorts of issues, and the auto settings on the MB tend to use values that will cause issues. So best to manually input the proper settings for your sticks, especially voltage. MB tend to default to 1.5 volts, whereas most high end ram use 1.65
And do the research :)

http://www.maximumpc.com/article/featur ... its_limits
http://www.xbitlabs.com/articles/cpu/di ... cking.html

User avatar
lycium
Developer
Posts: 1197
Joined: Wed Sep 12, 2007 7:46 am
Location: Leipzig, Germany
Contact:

Re: CoolColJ's test pics thread

Post by lycium » Thu Jun 10, 2010 6:42 pm

I must strongly recommend against those Corsair 1.333GHz C9 sticks, I bought 6GB of them when I first got my i7, and they *ahem* totally suck. 1 of the 3 sticks flat out doesn't work @ 1.3GHz on my Asus P9T motherboard, and the others are a bit flaky in their new home, which has been seeing some random bluescreen reboots lately. Couldn't get a warranty on them as they memtest fine :(

I bought that memory on November 10th 2008 (before i7s were formally released, to give you an idea of how old these chips are) and dealt with very poor performance and 4GB for a while, until I got 12GB of "Team Xtreme" (yeah I know) 1.6GHz C8 memory, which I have found to be excellent in terms of price, performance, stability and they use only 1.6v.

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Thu Jun 10, 2010 7:11 pm

Ok that's why they were going cheap and end of line :)

Well it's working fine now, since I set my QPI to Slow mode, something under 18x (or 36x on Gigabyte Motherboards). The slow mode didn't effect my games like I read on the net, still got 180fps on CS Source stress test like before.

6 sticks does stress the system more when overclocking though, this I know of. But I haven't had any BSODs since I increased my CPU voltage to 1.3 volts in the bios, QPI/VTT to 1.36, dropped uncore a bit lower and set QPI to Slow mode. It's rock solid now. Before I did this, I was seeing hard faults in the resource monitor for memory, you can bring up in the Windows 7 task manager, right now there is zero.

4Ghz with Turbo off is solid with base clock of 191. Might try to see if 4.2 ghz can be stable now :)
I'll try it with the motherboard software in windows


edit - hard faults are more to do with swap files than ram
And I have 6 gigs left after Windows7 uses some to cache. So maybe 12 gigs is not so bad afterall, especially when alt-tabbing out of games and switching programs - lightning fast!

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Thu Jun 10, 2010 10:12 pm

Have my PC running at 4.2ghz rock, solid stable! Now it will crash at 4.3 ghz without changing anything and just increase the base clock, but I'm not willing to go any higher with my CPU voltage and so on.

bidirectional Path tracing with v2.4.3 for 12 mins
CCJ_InteriorTest_BPT_12mins.jpg
Last edited by CoolColJ on Thu Jun 10, 2010 10:38 pm, edited 1 time in total.

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Thu Jun 10, 2010 10:29 pm

I did these tests in v2.4.3 .
I'll try v2.4.4 next to see if there are any major changes....

Around 10-12 mins each
And even though Path tracing gets over 1.2 million samples/sec, it fails at interior lit scenes :o
im1276156653.jpg
Path tracing
Bidirectional Path tracing - cleanest image of the lot, and less than half the sample rate of PT! Has problems lighting up the glass door cupboards at the back though....
im1276157328.jpg
BPT
Plain MLT - not bad, a bit of artifacts, but v2.4.4 might fix that. Sample rate was pretty high, and it renders way better than Path tracing, and lights up the rear glass dorred cupboards with ease.
Quite a lot brighter hough, by about 1.5-2 EV, I had already lowered EV by 1 and it's still bright than the other images.
im1276158136.jpg
MLT
Bidirectional MLT, the slowest of the lot in terms of samples/sec, but cleaner than MLT, and should light up the glass doored cupboards better than BPT
Attachments
im1276159067.jpg
BMLT

StompinTom
Indigo 100
Posts: 1821
Joined: Mon Sep 04, 2006 3:33 pm

Re: CoolColJ's test pics thread

Post by StompinTom » Fri Jun 11, 2010 12:36 am

Damn, 12 minutes! Did you change the resolution at all? That's pretty impressive, considering that it's a "real" scene versus just some simple objects thrown together for a test.

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Fri Jun 11, 2010 10:21 am

StompinTom wrote:Damn, 12 minutes! Did you change the resolution at all? That's pretty impressive, considering that it's a "real" scene versus just some simple objects thrown together for a test.
No that's the resolution I rendered them in.

While my PC was stable at 4.2ghz last night, after a cold 5 degree morning boot up it isn't so, I guess it needs to be warmed up first :)
All the BSODs state memory management issue. Back down to 4.09ghz, and relaxed memory timings. Although relaxing the memory from 9-9-9-24 to 10-10-10-28 may allow me back up to 4.2ghz, I'll see.

correction, my memory is Corsair CMX6GX3M3A1600C9, so it's not that old.I have two 3x2GB kit.
I should have went for some 3x4GB kit instead, but the shop didn't have anything like that.

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Fri Jun 11, 2010 2:00 pm

some renders with v2.4.4

I certainly notice the difference in sampling now, and caustics seem to resolve better in the water scenes I tried out. MLT and Path tracining no longer give me really high samples rates though! Basic MLT and Bidirectional MLT is almost about the same samples/sec wise, and Bidirectional PT is higher than Path tracing sometimes!
Something fishy about that :?

All rendered for 15 mins each n an i7 930 @ 4.1-4.2ghz

Interior scene - one 65k blackbody light
top row - PT, BPT. Bottom row - MLT, BMLT
Well the overall best is probably BMLT, it can light up the objects in the glass door cupboards, although BPT is cleaner. PT fails, and MLT while not bad is a bit blotchy, but less so than in v2.4.3.
CCJ_IndigoV2.4.4_comparo2_PT-BPT-MLT-BMLT.jpg
lit by Sunlight, no portals
top row - PT, BPT. Bottom row - MLT, BMLT
The MLT ones win for sure. Much cleaner and better reflections and caustics. Now PT is much faster than before, but can't do the caustics and reflections as well, creates fireflies instead. Much cleaner than BPT, which surprised me!
Look at the sample rates though, weird! BPT had higher sample rates than PT, that was never the case before. And MLT is closer to BMLT Sample rate wise. It used to be way higher
CCJ_IndigoV2.4.4_comparo_PT-BPT-MLT-BMLT.jpg
Combine the new PT code with bidirection, and add MLT for only the parts that need it and you have a winner :)

Note how different all the sample rates and samples per pixel, yet the higher ones don't always end up being the cleanest!

User avatar
CoolColJ
Posts: 1734
Joined: Mon Jun 25, 2007 1:47 pm

Re: CoolColJ's test pics thread

Post by CoolColJ » Fri Jun 11, 2010 5:35 pm

Path tracing in v2.4.4 is pretty fast outdoors. These renders cleaned up so fast, that they almost look biased :P

I also regained my 1+ million samples /sec by going outdoors. Not even 1000 samples and clean already
around 5 mins
quickrender.jpg
quickrender2.jpg

Post Reply
685 posts

Who is online

Users browsing this forum: No registered users and 2 guests