|
Is Prime95 just brutal on Skylakes? It caused like 20C higher temperatures than any other stress test. (I went and bought a 6700k, trying to OC it now)
|
#
?
Jun 7, 2016 20:37
|
|
|
|
| # ? Apr 28, 2024 05:53 |
|
|
Yes
|
|
#
?
Jun 7, 2016 20:45
|
|
|
Smol posted:Is Prime95 just brutal on Skylakes? It caused like 20C higher temperatures than any other stress test. It's the best tool for making sure your overclock is stable and not going to have too high of an average temp under stress. I'd also recommend doing other things while prime95 is running after you think it's stable. While ocing my i7 3770k I ran into a few situations where it could run prime95 for 24 hours+, but if I tried to watch a youtube video while prime95 was going or browse websites prime95 would crash or I'd blue screen. Getting the voltage settings just right was a pain. Khorne fucked around with this message at 20:55 on Jun 7, 2016 |
|
#
?
Jun 7, 2016 20:48
|
|
|
Boiled Water posted:For docks to really take off they really have to be more enticing than building a small gaming box, which currently looks unlikely all things considered. I'm not really sure what you're saying, but I guarantee you that if they stopped overbuilding their EGPU enclosures, they could get prices down and that would be enticing. When I say overbuilding, what I mean is, you don't need 350W set aside for a GPU. Most of the EGPU usage cases aren't going to be top of the line graphics cards, because it doesn't make sense to pair one most likely an Intel 6200U processor. I doubt anyone is going to put a 295X2 into one of these things, so they should build them with the GTX1080 or R480X as the highest card the enclosure is going to accept. That way they can cut down the size, cooling, power supply and price. It is expensive to try accommodate every possible video card, its better for PC manufacturers to try to funnel consumer choices a bit, in order to keep costs down for the majority of the market.
|
|
#
?
Jun 7, 2016 21:15
|
|
|
Twerk from Home posted:I'd agree with Intel. It's not like there's a shortage of quad-core laptops, and I think that most people choosing U CPUs don't really want more CPU performance. With Skylake they introduced relatively cheap mobile i5 quad cores too. Yeah, and the 45W Skylake mobile quads all also have 35W configurable TDP-down. If when designing a system you really need to use less than 35W from the processor, then you should probably stick to the 15W duals because a fast dual with HT is not going to be that far behind a slow 25W quad for most tasks and will cost a lot less due to being smaller. Moreover, many consumer tasks that would really benefit from a quad are games that would also need a better GPU and that will blow out your power targets too. Eletriarnation fucked around with this message at 21:43 on Jun 7, 2016 |
|
#
?
Jun 7, 2016 21:41
|
|
|
Eletriarnation posted:Yeah, and the 45W Skylake mobile quads all also have 35W configurable TDP-down. If when designing a system you really need to use less than 35W from the processor, then you should probably stick to the 15W duals because a fast dual with HT is not going to be that far behind a slow 25W quad for most tasks and will cost a lot less due to using less silicon. Moreover, many consumer tasks that would really benefit from a quad are games that would also need a better GPU and that will blow out your power targets too. The joke is that the duals are more expensive than the quads across the board at the moment 
|
|
#
?
Jun 7, 2016 21:43
|
|
|
Ah, well, today we also get an object lesson that Intel's pricing reflects what people will pay for things more than what they cost to make.
|
|
#
?
Jun 7, 2016 21:45
|
|
|
Eletriarnation posted:Ah, well, today we also get an object lesson that Intel's pricing reflects what people will pay for things more than what they cost to make. CPUs are quite inexpensive in terms of materials and fabrication cost, no? I thought it was on the order of $20 each.
|
|
#
?
Jun 7, 2016 21:46
|
|
|
Yeah, I don't know the cost of materials but I do know that yields are going to generally be worse on larger chips in the same process because you're more likely to encounter a defect as your area goes up. This is a big factor in the early days of a process when defects are more common, but I'm betting that at this point 14nm is pretty well worked out. This is especially true now that they're putting out 22-core Broadwell Xeons, so yields on quad-cores are probably great.
|
|
#
?
Jun 7, 2016 21:53
|
|
|
6700k @ 4.6 temps hovering around 65-75 while stress testing sound about right? No idea what this platform usually does.
|
|
#
?
Jun 7, 2016 22:05
|
|
|
Smol posted:6700k @ 4.6 temps hovering around 65-75 while stress testing sound about right? No idea what this platform usually does. You are safe till 105c
|
|
#
?
Jun 7, 2016 22:55
|
|
|
VulgarandStupid posted:I'm not really sure what you're saying, Essentially: If an enclosure costs $600 sans graphics card and building a whole computer sans graphics card costs me less, then why should I ever bother looking at enclosures?
|
|
#
?
Jun 7, 2016 23:16
|
|
|
Boiled Water posted:Essentially: If an enclosure costs $600 sans graphics card and building a whole computer sans graphics card costs me less, then why should I ever bother looking at enclosures? Exactly. And considering all whats inside even the most fancy shmancy enclosure so far, there is no reason for it to cost even half that.
|
|
#
?
Jun 7, 2016 23:19
|
|
|
Boiled Water posted:Essentially: If an enclosure costs $600 sans graphics card and building a whole computer sans graphics card costs me less, then why should I ever bother looking at enclosures? I've been saying the same thing, too. But I'm not primarily a laptop user. They need to come down in price, then they will be an attractive option for laptop users.
|
|
#
?
Jun 7, 2016 23:19
|
|
|
VulgarandStupid posted:I've been saying the same thing, too. But I'm not primarily a laptop user. They need to come down in price, then they will be an attractive option for laptop users. As primarily a laptop user I don't see the appeal. It's the same thing that should be nice about gaming laptops. However, as stated in the SH/SC thread on laptops: Gaming on the go is a unicorn. I mean sure some people just wanna chase that dragon but for most people reality sets in before buying it.
|
|
#
?
Jun 7, 2016 23:24
|
|
|
Also with new laptops like the Asus Strix and Gigabyte Aero coming into play in the 14-15" range that are near ultrabook size while having the power of those 17.3" beast of the past, especially when the Nvidia mobile (or desktop even) 1070/1080 build start dropping, the Ultrabook + Dock setup looses its appeal.
|
|
#
?
Jun 7, 2016 23:34
|
|
|
Boiled Water posted:As primarily a laptop user I don't see the appeal. It's the same thing that should be nice about gaming laptops. However, as stated in the SH/SC thread on laptops: Gaming on the go is a unicorn. I mean sure some people just wanna chase that dragon but for most people reality sets in before buying it. If gaming laptops weren't huge, gaudy monstrosities, there would be plenty more of them in consumer hands. But the appeal of an ultrabook and eGPU combo is that you're productive on the go, and can game at home. It won't be quite the top level gaming experience, but better for many consumers than owning two machines. I don't think there will be a high percentage of users taking their eGPU with them when they go somewhere but, there are enough docking stations out there to tell you that this is some what of a big deal.
|
|
#
?
Jun 7, 2016 23:42
|
|
|
Subjunctive posted:CPUs are quite inexpensive in terms of materials and fabrication cost, no? I thought it was on the order of $20 each. I don't think so. Intel's gross margin is in the low 60% range, meaning their cost of goods sold is in the high 30% range of revenue. On average, a part selling for $200 (wholesale) would cost $60ish or so in materials/fabrication cost. Of course, they've got different profit margins across different product lines, probably some with margins above 60% and probably some below 60%. That's not even considering what it costs to invent, develop, market, sell, and support the product, which is about another half of the gross profit. http://www.intc.com/results.cfm
|
|
#
?
Jun 8, 2016 00:55
|
|
|
Subjunctive posted:CPUs are quite inexpensive in terms of materials and fabrication cost, no? I thought it was on the order of $20 each. Sand is cheap, fabs are not.
|
|
#
?
Jun 8, 2016 06:10
|
|
|
Rare metals often sourced from locations with endless war, monstrously toxic chemicals with expensive storage and disposal, delicate multi-million-dollar fabrication gas chambers or etchers or whatever else in rows upon rows upon rows... If you ever get a chance to tour a fabrication facility's floor, take it.
|
|
#
?
Jun 8, 2016 12:00
|
|
|
PCjr sidecar posted:Sand is cheap, fabs are not. The actual cost of fabrication is cheap. The cost of developing a new node and building a fab for it is monstrously expensive, just like it's monstrously expensive to design a CPU and get litho masks made for it. The first hamburger costs a million dollars, after that they're fifteen cents.
|
|
#
?
Jun 8, 2016 18:49
|
|
|
Smol posted:Is Prime95 just brutal on Skylakes? It caused like 20C higher temperatures than any other stress test. The newest versions of P95 use the new avx instructions which apparently transform skylakes into space heaters. I have read that it is better to use v26.6 for more reasonable temps since few real world situations utilize avx 2.0 or whatever it is that drives skylake to run crazy hot. This resulted in a ~20 degree drop for me at 4.7GHz and 1.32V on my 6700k.
|
|
#
?
Jun 9, 2016 00:42
|
|
|
AVX instructions cranking the heat up to 11 isn't just a Skylake thing, either, all the chips that support AVX throttle down more when using them.
|
|
#
?
Jun 9, 2016 15:10
|
|
|
What uses AVX instructions in real-world workloads?
|
|
#
?
Jun 9, 2016 15:32
|
|
|
Finding prime numbers of 2n-1 form?  Seriously though, is prime95 a good benchmark for any kind of performance other than thermal load? Potato Salad fucked around with this message at 15:52 on Jun 9, 2016 |
|
#
?
Jun 9, 2016 15:48
|
|
|
Now considering AVX is the new Intel Instructions for "stuff" wouldn't it being used at that level be the chip getting hot because it is actually working hard (If say it was ever used for something like encoding)? How efficiently it is working is another test I guess. Also on my cooler note, turns out I was just running the H100 on level 2 instead of 3 so once I could reach in and hit the button on the pump, level 3 is a good bit louder, but dropped my max temps from 80Cish to 68ish so far. Now nothing in my case should go above 70C at load which makes me happy even though I'd like something quieter.
|
|
#
?
Jun 9, 2016 15:51
|
|
|
Twerk from Home posted:What uses AVX instructions in real-world workloads? Libyuv, used by hangouts and chrome for software encode/decode will use it if it's available, and fall back to ssse3 or sse2 if necessary.
|
|
#
?
Jun 9, 2016 15:54
|
|
|
Fuzzy Mammal posted:Libyuv, used by hangouts and chrome for software encode/decode will use it if it's available, and fall back to ssse3 or sse2 if necessary. Interesting. I thought that recently there had been a huge push to use dedicated h264 / h265 / vp9 hardware encode / decode. I guess that Intel stuff doesn't have hardware VP9 encode and google loves vp9, so software encoding still matters.
|
|
#
?
Jun 9, 2016 16:46
|
|
|
Twerk from Home posted:What uses AVX instructions in real-world workloads? I think PSCX2 will use AVX if you tell it to.
|
|
#
?
Jun 9, 2016 17:38
|
|
|
Potato Salad posted:Rare metals often sourced from locations with endless war, monstrously toxic chemicals with expensive storage and disposal, delicate multi-million-dollar fabrication gas chambers or etchers or whatever else in rows upon rows upon rows... I interned at a plant that built steering racks in highschool. The hermetically sealed cell that built stuff for the (then-super secret, hadn't-been-revealed-yet) Camaro was awesome. I can only imagine the PPE requirements to work in a fab.
|
|
#
?
Jun 9, 2016 18:21
|
|
|
havenwaters posted:I think PSCX2 will use AVX if you tell it to. I looked up someone's tests and you can use it but there's not much benefit over SSE4.1. HMS Boromir fucked around with this message at 18:27 on Jun 9, 2016 |
|
#
?
Jun 9, 2016 18:25
|
|
|
SuperDucky posted:I interned at a plant that built steering racks in highschool. The hermetically sealed cell that built stuff for the (then-super secret, hadn't-been-revealed-yet) Camaro was awesome. I can only imagine the PPE requirements to work in a fab. Well there is also the Intel ARK that shows that the materials used to make said products come from "Conflict Free" locations, so there is that too now.
|
|
#
?
Jun 9, 2016 18:59
|
|
|
Potato Salad posted:Finding prime numbers of 2n-1 form? Prime95 uses FFTs to implement large integer multiplication and FFTs are used in a wide variety of scientific and engineering applications.
|
|
#
?
Jun 9, 2016 19:05
|
|
|
MaxxBot posted:Prime95 uses FFTs to implement large integer multiplication and FFTs are used in a wide variety of scientific and engineering applications. So basically you can use Prime95 to test your overclocking in a situation where you would not be overclocking. A while ago I read a comment that you shouldn't use P95 to test your OC and I understood the argument in the way, that if you limit your OC to where Prime95 is stable you will not get full power out of your CPU. You can OC to a degree where P95 is crashing or causing the CPU to thermal throttle, but games or anything else you do would still work perfectly at that stage.
|
|
#
?
Jun 9, 2016 21:03
|
|
|
How does Prime95 compare to using something like IntelBurnTest to set an OC?
|
|
#
?
Jun 9, 2016 21:08
|
|
|
Instant Sunrise posted:How does Prime95 compare to using something like IntelBurnTest to set an OC? Intel Burn Test itself isn't completely reliable- The 3770k I had was fine running IBT, but would BSOD when encoding a video in Premier. The reality is that most people's OCs probably have hidden landmines of some form of instability waiting there, because none of us can or care to obviously run the type of tests that are able to conclusively determine full OCed functionality. But if it works most of the time or for any of the programs that we do use, good enough! Re: AVX, I thought the reason that use of those instructions caused so much extra heat was that it was just a matter of the die space devoted to implementing them- More area with current running through it, more heat cranked out.
|
|
#
?
Jun 9, 2016 21:23
|
|
|
For OC Testing I just defaulted to OCCT which seems to have the most reliable results as long as you move the Stop Temps up a bit. Good way to find bad ram too apparently.
|
|
#
?
Jun 9, 2016 21:32
|
|
|
Gwaihir posted:Re: AVX, I thought the reason that use of those instructions caused so much extra heat was that it was just a matter of the die space devoted to implementing them- More area with current running through it, more heat cranked out. Pedantic response, but generally instructions don't activate their own die space, but get decoded down into execution units. I doubt they have significant additional execution units devoted to only these instructions, but it is quite possible that these instructions allow many more execution units to fire on a set of data at the same time, which would drive power consumption and heat. *Checks Wikipedia*, yeah, wider bus plus additional operand commands for execution. Yeah, that'd light up more of the execution units.
|
|
#
?
Jun 9, 2016 21:35
|
|
|
Potato Salad posted:Finding prime numbers of 2n-1 form? Absolutely not, Prime95 SmallFFT is basically a power virus - it runs drastically hotter than any real-world workload, and in extreme cases it may actually damage your processor. Not that I would worry about a 10 minute test run but I wouldn't leave a machine doing it nonstop for hours as a stability test or anything. Thermal throttling will probably catch any problems before they get too bad, but SmallFFT generates a fuckload of heat in some specific units and who knows if the thermal measurement points are close enough to catch a localized hotspot in time? It's just too much of a risky test unless you have a good reason, it's not like there aren't other burn-in programs or benchmarks or whatever. Paul MaudDib fucked around with this message at 22:57 on Jun 9, 2016 |
|
#
?
Jun 9, 2016 22:49
|
|
|
|
| # ? Apr 28, 2024 05:53 |
|
|
Gwaihir posted:Intel Burn Test itself isn't completely reliable- The 3770k I had was fine running IBT, but would BSOD when encoding a video in Premier. This is totally correct and it really does take some time to be confident that an OC is really totally stable. Not every workload stresses every functional unit to the max (that's probably impossible, actually) and you never know when the combination of unit X and total load Y is going to cause enough ripple or droop or something to gently caress up unit Z. Or, such is my understanding. I actually don't like generating data for archival purposes (things like video encoding or audio recompression) while overclocked. I've run into it on my 780 Ti, some workloads would barf at 1 GHz while others would happily crank along 200mhz faster.
|
|
#
?
Jun 9, 2016 22:53
|
|