| Forum Index > Tech Support |
When using this resource, please read the opening post. The Tech Support forum regulars have helped create countless of desktop systems without any compensation. The least you can do is provide all of the information required for them to help you properly. | ||
|
Mysticesper
United States1183 Posts
+ Show Spoiler + ![[image loading]](http://i.imgur.com/63pfJ9x.png) This is kind of my progress? I am curious though, is that temperature asymmetry normal, or is the heatsink (NH-D14) fitted better on one half of the chip than the other / thermal paste asymmetry (or something else)? I assume it's fixable (depending), just curious. It would make sense due to the cantilever, at least. It is the first time I managed multiple successful runs of IBT on standard (2x 10, 1x 20) | ||
|
skyR
Canada13817 Posts
| ||
|
Ropid
Germany3557 Posts
Normally, the first and fourth cores are colder as the second and third get cooked by their neighbors. I guess the graphics on the CPU are off for most people. You should compare after you get your graphics card. The GFlops reported by IBT are very low. You didn't yet update your Windows 7? This needs to be done to enable the use of the AVX instruction set of your CPU. After you do that, the temperature of IBT will go through the roof, so be prepared for that. + Show Spoiler [about stress testing and overclocking] + If I were you, I wouldn't touch IBT anymore. What it does is just severely off compared to normal programs. It does nothing useful, just tries to get the highest GFlops score for "linpack". It's an old supercomputer benchmark. The Haswell CPUs squeeze out a good bit more performance out of those AVX instructions compared to past CPUs, but that does not come free. The heat will be massive and you will hit the temperature limits and the CPU will throttle. You should test for heat by encoding videos with x264. That is also strongly optimized, uses AVX and all CPU cores. There's a "x264 benchmark" package floating around with script files to run a test unattended. In the Windows event viewer, there's "WHEA-Logger" warnings that can happen without the PC crashing. You should check for those regularly if you are not yet convinced your PC runs stable. For testing stability, I've seen people say chess engines work very well. Those are optimized by hand, are very good at provoking "WHEA-Logger" events if the PC is not quite stable. They don't use floating point instructions so the CPU won't get super hot. It might be annoying to hunt down the required programs and configure them: "arena chess gui" and engines "stockfish", "rybka", "critter", "houdini" (some old versions of the commercial engines are free). Arena can run automated engine tournaments. I can only confirm those chess engines also work excellent for sniffing out instability on Ivy Bridge. On my CPU, half an hour of chess seems to be a better test than half a day of prime95. | ||
|
Incognoto
France10239 Posts
By the way, just a quick, somewhat random, question that I had laying around in my mind recently. The Noctua NHD14 is a really BIG cooler right? Besides making sure our case is big enough to hold the radiator (in terms of height), will such a big cooler ever collide with RAM? Or will it just collide with RAM with tall heatspreaders? Also, is it normal for me to want to completely take apart my computer just to reassemble it? | ||
|
Mysticesper
United States1183 Posts
I would be afraid of delidding. I am using a radeon 5870 as my dedicated GPU, so the chip shouldnt have to worry about that unless I can turn it off in the bios. I am relelatively new with OCing at a higher level. I've done a few small ones, But ill look into x264 now. Thanks. There is plenty of clearance in the ARC Midi R2 case for the heatsink, and as long as you arent over 44mm (rather standard), it should work out fine. I used G.Skill Snipers. My 4.6 testing resulted in 80-85 temps at somewhere beween 1.28 and 1.3 My 4.7 was pushing beyond 1.3, and it started to peak out at 90. | ||
|
Cyro
United Kingdom20323 Posts
On October 20 2013 16:37 Incognoto wrote: Isn't 90°C a bit high? I don't overclock but if I did I would try to never go over 80°C. 4.7 Ghz isn't too bad a clock though. Did you delid or not? By the way, just a quick, somewhat random, question that I had laying around in my mind recently. The Noctua NHD14 is a really BIG cooler right? Besides making sure our case is big enough to hold the radiator (in terms of height), will such a big cooler ever collide with RAM? Or will it just collide with RAM with tall heatspreaders? Also, is it normal for me to want to completely take apart my computer just to reassemble it? People take them to >90c peaks pretty often, i wouldn't go past maybe low 80's encoding load. Voltage much more important than clock speed for temps, it's kinda a lottery in that you can apply like 1.4v and call it a high overclock, and if that gives you 4.5ghz or 5.0 depends on the chip A lot of RAM with tall heatspreaders does not fit under quite a few air coolers, best to avoid them because of that (the RAM, because tall heatspreaders are mostly there for aesthetics and advertising, not actual function) Maybe  | ||
|
Incognoto
France10239 Posts
Secondly, if I were to delid today, I would be putting CLU back on the die, right? So something like this: http://www.overclockers.co.uk/showproduct.php?prodid=TH-013-CL Why can't I put that same stuff between the IHS and the heaksink? Is there a product that could be used for both roles? Not that I'm going to delid, but I like the idea a lot so maybe I will fuck me what would the point of delidding a non-K processor be? Maybe some nice temperature gains though I'm not even overclocking. Though I think my system could be configured to be overall more quiet (not that it's noisy at all). | ||
|
Cyro
United Kingdom20323 Posts
A couple more questions. First of all, I was looking for some RAM for that HTPC and I noticed that it was possible to get a module of 1866 RAM for just a few more € than 1600 RAM (4 Gb). Since my motherboard can support those frequencies, I think I'm going to go for that stick. Good call or not? Higher clockspeeds are done with looser timings usually, or just the same RAM ic's with increased voltage, an 1866mhzc9 RAM that needs 1.65v could be a worse kit than a 1600mhz c9 kit that works on 1.5v Secondly, if I were to delid today, I would be putting CLU back on the die, right? So something like this: http://www.overclockers.co.uk/showproduct.php?prodid=TH-013-CL yea Why can't I put that same stuff between the IHS and the heaksink? You can, but not with an aluminum heatsink, and it's awkward to remove etc, stains IHS (i don't know all details because i never looked into it that much, it's not much better than easier pastes like mx4) ^the temp difference is very small above IHS Is there a product that could be used for both roles? Probably, but you ideally want clu under IHS, and whatever good paste you're comfortable with above IHS. If you wanted every degree, you'd have a lapped IHS with CLU and some time remounting etc what would the point of delidding a non-K processor be? Maybe some nice temperature gains though I'm not even overclocking. Though I think my system could be configured to be overall more quiet (not that it's noisy at all). Macho doesn't have to touch 1000rpm to keep a stock i5 cool, your GPU would be by far the noisiest thing in your system, your CPU fans don't have to spin when cpu is idle (only case fans) not much reason to do it on a stock cpu, unless you are confident. If you can't afford an unlocked cpu and decent z87 board, you probably dont want to be replacing a dead cpu either | ||
|
Mavvie
Canada923 Posts
| ||
|
Cyro
United Kingdom20323 Posts
![[image loading]](http://i.imgur.com/kSyJhBO.png) 80gflops is non-avx level of performance (i think? maybe low clocked ivb with avx1) If you're that low, you have bad performance, maybe running the wrong version of linpack and/or can't run AVX instructions (missing windows 7 service pack 1 which adds OS level support for those cpu instructions) If you're on Haswell, read this: + Show Spoiler + On October 20 2013 16:11 Ropid wrote: The Haswell die looks like this: http://cdn2.wccftech.com/wp-content/uploads/2013/05/Intel-Haswell-Core.jpg Normally, the first and fourth cores are colder as the second and third get cooked by their neighbors. I guess the graphics on the CPU are off for most people. You should compare after you get your graphics card. The GFlops reported by IBT are very low. You didn't yet update your Windows 7? This needs to be done to enable the use of the AVX instruction set of your CPU. After you do that, the temperature of IBT will go through the roof, so be prepared for that. + Show Spoiler [about stress testing and overclocking] + If I were you, I wouldn't touch IBT anymore. What it does is just severely off compared to normal programs. It does nothing useful, just tries to get the highest GFlops score for "linpack". It's an old supercomputer benchmark. The Haswell CPUs squeeze out a good bit more performance out of those AVX instructions compared to past CPUs, but that does not come free. The heat will be massive and you will hit the temperature limits and the CPU will throttle. You should test for heat by encoding videos with x264. That is also strongly optimized, uses AVX and all CPU cores. There's a "x264 benchmark" package floating around with script files to run a test unattended. In the Windows event viewer, there's "WHEA-Logger" warnings that can happen without the PC crashing. You should check for those regularly if you are not yet convinced your PC runs stable. For testing stability, I've seen people say chess engines work very well. Those are optimized by hand, are very good at provoking "WHEA-Logger" events if the PC is not quite stable. They don't use floating point instructions so the CPU won't get super hot. It might be annoying to hunt down the required programs and configure them: "arena chess gui" and engines "stockfish", "rybka", "critter", "houdini" (some old versions of the commercial engines are free). Arena can run automated engine tournaments. I can only confirm those chess engines also work excellent for sniffing out instability on Ivy Bridge. On my CPU, half an hour of chess seems to be a better test than half a day of prime95. If you're looking for a benchmark, use Cinebench r15 | ||
|
Mavvie
Canada923 Posts
I'm on win8. Obviously my processor should be slower than an i7, but I wouldn't have expected by a factor of like 4. Okay, I downloaded linpack 11.1 and ran the benchmarks/linpack_xeon64.exe, and in the output it's reporting like 40 GFLOP/s. + Show Spoiler [Screenshot] + ![[image loading]](http://i.imgur.com/kaYKKnH.png) | ||
|
riotjune
United States3394 Posts
![[image loading]](http://i.imgur.com/IrAUc0E.jpg) The Fractal case manual didn't come with pictures, but since it seems to be a popular case there were plenty of youtube videos on it. Though it's a budget build, it's still a huge step up from my previous laptop. If I knew booting from an SSD was so fast I would've done it sooner and saved hours of my life. Thanks especially to skyR for his really helpful tutorial (and you were right, this was a learning experience, and it was fun), and to itzsnypah for picking out the components. I'd buy you guys a drink or something (or here, have TL+ instead )I probably won't buy a prebuilt ever again. I recommend anyone who has no idea how to build a computer (like me) to start here, there's a lot of people willing to help. | ||
|
Cyro
United Kingdom20323 Posts
I guess 80gflops is probably a normal avx1 result for stock ivy bridge. It might go up a little on ivy bridge if you disable hyperthreading, but you don't want to do that for the vast vast majority of applications, it's just a niche case where maybe higher performance with ht forced off Again though, Linpack is not a good benchmark - use Cinebench r15 and other programs for that Gz riotjune! | ||
|
Mavvie
Canada923 Posts
I didn't know that Haswell has avx2, but I doubt that most programs are built with those optimizations/instructions anyway? I'm going to disable hyperthreading again actually, almost everything I do runs slower with it on (web browser stuff, sc2, non-concurrent web-dev stuff). Yeah, I was only curious about FLOPS because I read a bit about Folding@Home and was curious as to how much a peta-FLOP truly is in terms of modern-day processors. Turns out, it's a lot. | ||
|
Cyro
United Kingdom20323 Posts
I didn't know that Haswell has avx2, but I doubt that most programs are built with those optimizations/instructions anyway? A few things do, but not much stuff can be sped up by avx/avx2. In terms of video encoding, avx2 support over avx gives haswell a ~5% boost in performance, before taking into account other architectural improvements etc, so the difference between haswell and ivy bridge etc is bigger there than in most other cases If you have to do some work composed of task A and B - task A takes 90 seconds task B takes 10 seconds You need to do both of them, therefore if you can do task B twice as quickly, you're only doing the task as a whole in 95 seconds instead of 100 seconds. That's the kind of situation with avx2 being ridiculously fast Linpack is just writing the entire workload as task B and saying hey, look at this | ||
|
Incognoto
France10239 Posts
On October 20 2013 23:15 Cyro wrote:+ Show Spoiler + A couple more questions. First of all, I was looking for some RAM for that HTPC and I noticed that it was possible to get a module of 1866 RAM for just a few more € than 1600 RAM (4 Gb). Since my motherboard can support those frequencies, I think I'm going to go for that stick. Good call or not? Higher clockspeeds are done with looser timings usually, or just the same RAM ic's with increased voltage, an 1866mhzc9 RAM that needs 1.65v could be a worse kit than a 1600mhz c9 kit that works on 1.5v Secondly, if I were to delid today, I would be putting CLU back on the die, right? So something like this: http://www.overclockers.co.uk/showproduct.php?prodid=TH-013-CL yea Why can't I put that same stuff between the IHS and the heaksink? You can, but not with an aluminum heatsink, and it's awkward to remove etc, stains IHS (i don't know all details because i never looked into it that much, it's not much better than easier pastes like mx4) ^the temp difference is very small above IHS Is there a product that could be used for both roles? Probably, but you ideally want clu under IHS, and whatever good paste you're comfortable with above IHS. If you wanted every degree, you'd have a lapped IHS with CLU and some time remounting etc what would the point of delidding a non-K processor be? Maybe some nice temperature gains though I'm not even overclocking. Though I think my system could be configured to be overall more quiet (not that it's noisy at all). Macho doesn't have to touch 1000rpm to keep a stock i5 cool, your GPU would be by far the noisiest thing in your system, your CPU fans don't have to spin when cpu is idle (only case fans) not much reason to do it on a stock cpu, unless you are confident. If you can't afford an unlocked cpu and decent z87 board, you probably dont want to be replacing a dead cpu either No the voltage is the same, 1.5 for both sticks. Difference is a couple €. http://www.amazon.fr/Crucial-BLT4G3D1608DT1TX0CEU-PC3-12800-Ballistix-Tactical/dp/B006YG96G8/ref=sr_1_3?ie=UTF8&qid=1382284781&sr=8-3&keywords=Crucial 4 Go ddr3 1600 http://www.amazon.fr/gp/product/B006YG9BPY/ref=ox_sc_act_title_1?ie=UTF8&psc=1&smid=A1X6FK5RDHNB96 I'm guessing I'll go with the 1866 and call it a day, seriously what is €1? It's a croissant, that's what. I can afford plently of croissants so that solves that. I guess I won't delid my processor. The computer seriously isn't noisy in the slightest, even running FC3 at ultra settings. I think that what I'll do at most is buy a tube of thermal paste and replace the thermal strips on the stock cooler (it's called macho?), just for kicks. I'll overclock skylake when it comes out! Edit: @Riotjune, looks sexy, it reminds me a lot of my own set up. Card looks like a Sapphire something? I immediately recognized the Core 1000's interior. | ||
|
Cyro
United Kingdom20323 Posts
That RAM is single 1x4gb sticks i believe, cant read french. You'd want 2x2 or 2x4gb | ||
|
Mavvie
Canada923 Posts
On October 21 2013 01:06 Cyro wrote: Most stuff won't actually run slower just because HT is on. They might not utilize it, but it's mostly exceptions that actually lose speed because of it A few things do, but not much stuff can be sped up by avx/avx2. In terms of video encoding, avx2 support over avx gives haswell a ~5% boost in performance, before taking into account other architectural improvements etc, so the difference between haswell and ivy bridge etc is bigger there than in most other cases If you have to do some work composed of task A and B - task A takes 90 seconds task B takes 10 seconds You need to do both of them, therefore if you can do task B twice as quickly, you're only doing the task as a whole in 95 seconds instead of 100 seconds. That's the kind of situation with avx2 being ridiculously fast Linpack is just writing the entire workload as task B and saying hey, look at this About HT...I'm talking about single-threaded applications. I was under the impression that they ran at slightly above 1/2 the speed, you could just run twice as many of them. Perhaps it's confirmation bias, but I thought I've noticed HUGE performance gains in single-threaded programs by disabling it. Maybe I'm totally wrong though haha!  Okay, the avx2 stuff makes sense. I don't do video encoding anwyays, so I won't notice a difference. Edit: Funny typo haha, this would be some slow RAM! ![[image loading]](http://i.imgur.com/DSDAmDL.png) | ||
|
Cyro
United Kingdom20323 Posts
About HT...I'm talking about single-threaded applications. I was under the impression that they ran at slightly above 1/2 the speed, you could just run twice as many of them That's not how hyperthreading works. I've benchmarked it on/off to be within a tenth of a percent in sc2, though i have it off for playing because it makes the stutters in the sc2 engine far worse for me, i dunno if other people have those issues (would have to look at their monitor) I don't do video encoding anwyays, so I won't notice a difference. CPU load from streaming = almost entirely x264 (awesome video encoder)I'm gonna say cinebench r15 a third time though, totally kickass program | ||
|
Mavvie
Canada923 Posts
In task manager, HT on shows my program uses only 14% of the CPU, whereas with HT off it shows 28%, but they actually get processed/run at the same speed! Today I learned...This actually scales really well, doesn't it? If you're running a program that only supports 1-4 cores, you get 100% speed, while if you run a program that supports all 8 you can (sometimes?) get up to 130%. Pretty cool solution, I'm actually really impressed right now haha. x264 isn't its own instruction set though, it's just an encoder right? Which instruction set does it use? I've used Cinebench before! It's pretty cool, but my results were kinda sketchy. My OpenGL test results were the same if I had my GPU core clock at 400MHz or 1200MHz, so I think that I was doing it wrong.. | ||
| ||
StarCraft 2 Dota 2 Counter-Strike Super Smash Bros Heroes of the Storm Other Games |
|
Replay Cast
Wardi Open
Monday Night Weeklies
OSC
WardiTV Winter Champion…
PiGosaur Cup
Replay Cast
WardiTV Winter Champion…
Replay Cast
PiG Sty Festival
[ Show More ] |
|
|
EPT
