|
Josh Lyman posted:My Antec Trupower Classic has 2x +12V rails That’s still 360 W on either rail, though, which ought to be enough? 
|
#
?
Mar 4, 2017 21:13
|
|
|
|
| # ? May 30, 2024 15:46 |
|
|
fat bossy gerbil posted:Intel processors with a K in the number (7600k etc.) can overclock, chips without the k don't. If you want to overclock get a Z series chipset, if you know you wont overclock and need an affordable board get an H series. Don't buy the B/H-series boards ever. The $10 you save now isn't worth permanently locking yourself out of the capability to switch to an unlocked processor. track day bro! posted:Stupid question, I should be able to run a 5820k and a 1070 on a good 550w seasonic psu at std clocks right? Yes, you'll be fine.
|
|
#
?
Mar 4, 2017 21:36
|
|
|
redeyes posted:Make sure and get a 1x 12v rail model. You do not want the older multiple 12v rail ones with modern video cards. Why's that? Also is it going to make a difference with a 150W card like the 1070?
|
|
#
?
Mar 4, 2017 21:44
|
|
|
I kinda hope intel responds to the hopeful pressure from amd by sticking fpgas in the chip (whether as an increase in relative die size or replacing a core). While I'm almost certain the 'programs' for them would be an added cost ('App Store'), it'd be nice to be able to get hardware decode for like vp11 or h266 or (less nicely) whatever horrible drm is implemented in the future.
|
|
|
#
?
Mar 4, 2017 21:50
|
|
|
Talking about power supplies, one thing that was recently discovered at my work is that the 6 pin connectors for the molex 4 pin/sata power are NOT the same between manufacturers. We are moving from using corsair bx 750s to evga 750g3s at work and a new guy plugged a cable that uses a molex power connector in using the corsair ribbon cables and kerpow. Turns out the +5v and +12v are swapped, but the keying works fine. Nothing expensive was blown up but was still a  moment, that they'd be keyed the same. Even still it should be obvious to only use the cables that come with the power supply but I would have expected more idiot proofing. Anyway just thought that was interesting.
|
|
#
?
Mar 4, 2017 22:05
|
|
|
^ Yeah, modular power connectors are not standardized in any way so you need to make sure you keep original cables for each unit or use known-compatible replacements.ConanTheLibrarian posted:Why's that? Also is it going to make a difference with a 150W card like the 1070? Multi-rail supplies are more complex internally and have more limitations for no real gain. It shouldn't make a substantial difference though with something like a 1070 as long as you don't load everything on one rail, and usually EPS12V (for the CPU) and PCIe connectors won't be on the same rail if there's more than one.
|
|
#
?
Mar 4, 2017 22:30
|
|
|
The whole multi rail being bad throws me every time. Years ago everyone was paying extra for dual rail and was heralded as the best thing. Then years pass and I find out its a bad thing.
|
|
#
?
Mar 4, 2017 23:19
|
|
|
Captain Hair posted:The whole multi rail being bad throws me every time. Years ago everyone was paying extra for dual rail and was heralded as the best thing.
|
|
#
?
Mar 4, 2017 23:35
|
|
|
Allow me to introduce you to the Dell 5 rail monstrosity  I pulled a bunch of them out of decommissioned workstations that were getting thrown out and I still have some of them in use today. I'd never ever recommend them to someone, but for free they're alright and they're my favorite reminder of the bad old days of split rails.
|
|
#
?
Mar 4, 2017 23:49
|
|
|
Multi-rail designs are perfectly fine as long as the rails are big enough for your usage. Single-rail PSUs are simpler and easier, but there is nothing inherently wrong with a multi-rail PSU and you shouldn't throw yours out just for that. OklahomaWolf of JonnyGuru fame famously prefers multi-rail designs, even if they have all but disappeared from the market.
|
|
#
?
Mar 5, 2017 02:52
|
|
|
In theory I guess a really heavy load on one rail might cause voltage droop, but with a multi-rail design that would be confined to just that one rail. Single-rail would droop less in the first place though. The bigger problem with multi-rail is when you get into stuff like the 295x2 where it needs 28 amps on each power connector. Some designs have unbalanced rails, like 40 amps on one rail and 20 on the other, or 30-20-20. Enough power in theory but it's split across a pair of rails. edit: or yeah like that 5-rail design, what the gently caress
|
|
#
?
Mar 5, 2017 03:03
|
|
|
Paul MaudDib posted:Don't buy the B/H-series boards ever. The $10 you save now isn't worth permanently locking yourself out of the capability to switch to an unlocked processor. The DDR4 support above 2133/2400 on the Z-boards is arguably just as critical as CPU OCing.
|
|
#
?
Mar 5, 2017 14:17
|
|
|
redeyes posted:Make sure and get a 1x 12v rail model. You do not want the older multiple 12v rail ones with modern video cards. The reviewer over at johnnyguru constantly says that the idea that multi rail PSUs are worse than single is a complete myth, and that multi rail is often better at higher wattages.
|
|
#
?
Mar 5, 2017 14:27
|
|
|
The Lord Bude posted:The reviewer over at johnnyguru constantly says that the idea that multi rail PSUs are worse than single is a complete myth, and that multi rail is often better at higher wattages. Well, my most recent AMD RX 480 would NOT run on anything less than about 40a on the single 12v line. Could be something else but after reading around a bit, all new designs are single rail and I suspect for a reason.
|
|
#
?
Mar 5, 2017 14:54
|
|
|
priznat posted:
I'm more interested in what you guys are building at work. We used to custom build workstations in the early 2000s because because parts weren't great or even consistent in prebuilts but these days I assumed everyone ordered a dell precision or whatever and called it a day. Do you work in IT or at a boutique system builder?
|
|
#
?
Mar 5, 2017 16:36
|
|
|
buffbus posted:I'm more interested in what you guys are building at work. We used to custom build workstations in the early 2000s because because parts weren't great or even consistent in prebuilts but these days I assumed everyone ordered a dell precision or whatever and called it a day. Do you work in IT or at a boutique system builder? We're doing hardware development and use off the shelf PSUs to power the dev boards and infrastructure. We had some issues with the corsairs so are moving away from them. We also order a variety of servers to interop with from the major manufacturers and the occasional custom build pc just to get a wide variety of chipsets and cpus to test against.
|
|
#
?
Mar 5, 2017 17:46
|
|
|
.
sincx fucked around with this message at 05:55 on Mar 23, 2021 |
|
#
?
Mar 6, 2017 10:44
|
|
|
What's your PC usage look like? Existing kaby lake plus fast ddr4 most certainly is a large upgrade from a typical 2600k system in gaming terms, especially in the min fps area. (And a typical 7700 gets to 5ghz without much effort). If it's not games, then any of the 6+ core CPUs are upgrades, even if I wouldn't buy in to X99 at the moment.
|
|
#
?
Mar 6, 2017 14:22
|
|
|
I have a weird CPU question. As you go smaller and smaller nanomater fabrications, you can either put more stuff on the same size die, or use the same amount of stuff on a smaller die, yes? But isn't there some surface area rule that the smaller you go the harder it is to transfer heat? Ie, a 100w die with a surface area of 100mm2 would not be able to transfer as much heat as a 100w die with a surface area of 150mm2? In such a situation would it be possible to take the blueprint of a 10nm die and sort of spread it out a bit? Imagine zooming in but keeping each line drawn the same thinness of 10nm? I can't words.
|
|
#
?
Mar 7, 2017 12:35
|
|
|
The problem with spreading it out means you get limited to lower clock speeds because of the extra distance and at that point you've lost most of the benefit of node shrinking, more spread out also means less room for extra transistors for more optimized execution units and keeping the die big means the dies per wafer count goes down as well. Maybe if you have the budget of the DOD or something to spend on a few hundred processors you could commission something like that, but a design like that doesn't line up with consumer demands.
|
|
#
?
Mar 7, 2017 12:49
|
|
|
Hmm, DoD milspec processors running at high power levels you say? https://arstechnica.com/information-technology/2016/06/cheaper-better-faster-stronger-ars-meets-the-latest-military-bred-chip/
|
|
#
?
Mar 7, 2017 13:35
|
|
|
Ak Gara posted:I have a weird CPU question. You could, there just wouldn't be a point. As craig588 notes, it would limit it's performance and cost more than a completely shrunk die. The only benefit would be potentially better thermal dissipation, and frankly that's not really a problem in most cases these days. Maybe if there were some very particular special application that called for it, but otherwise it's a solution in search of a problem.
|
|
#
?
Mar 7, 2017 14:47
|
|
|
Ak Gara posted:I have a weird CPU question. When you put the same amount of stuff on the smaller die, you use less electricity. The smaller transistors and wires use less electricity than the larger ones.
|
|
#
?
Mar 7, 2017 15:11
|
|
|
I have an interesting problem: My supervisor has a factor analysis problem which can chew through as many cores as can be thrown at it. Question is, how many cores is that in workstation form, and where would I order such a thing? AWS and any cloud services are out of the question, so is building own + the troubleshooting that goes with it. Also has to be available to order in Denmark.
|
|
#
?
Mar 8, 2017 21:33
|
|
|
Boiled Water posted:I have an interesting problem: My supervisor has a factor analysis problem which can chew through as many cores as can be thrown at it. Question is, how many cores is that in workstation form, and where would I order such a thing? If the only goal is to maximize number of cores in a workstation, you can get something like an HP Z840 Workstation or Dell Precision Tower 7000 Series (7910) configured with dual Xeon E5-2699 v4 processors (22 cores each, total 44 cores). It's possible that just maximizing number of cores may not be the best performance configuration for that application, however.
|
|
#
?
Mar 8, 2017 22:02
|
|
|
I would benchmark a similar problem to determine your needs. Find a smaller, similarly structured problem and determine computation time against number of cores, single core speed, total memory, memory bandwidth, in order to find your bottleneck. For instance, I do hobby CFD, and research has shown you can throw cores at it for marginal improvements, but the greatest gains come from higher memory bandwidth.
|
|
#
?
Mar 8, 2017 22:25
|
|
|
Thanks duders. The problem is Parallel Factor Analysis in parallel, ie. a million models at the same time through some matlab script. My only other question beyond "how do i maximize core" would be why isn't this written in a CUDA-like fashion? Only the mind that boggles knows, and it's not telling.
|
|
#
?
Mar 8, 2017 23:07
|
|
|
Boiled Water posted:Thanks duders. The problem is Parallel Factor Analysis in parallel, ie. a million models at the same time through some matlab script. My only other question beyond "how do i maximize core" would be why isn't this written in a CUDA-like fashion? Only the mind that boggles knows, and it's not telling. https://www.mathworks.com/discovery/matlab-gpu.html
|
|
#
?
Mar 8, 2017 23:13
|
|
|
Intel has revealed the pricing for their Retail Edge Spring Deal. 7700k for $280, $230, or $200 depending on your retail edge level. These were last holiday deals. Deal 1 - Intel i7-6850K Legends: $169.00 USD Masters: $179.00 USD Apprentices: $199.00 USD Deal 2 - Intel i7-6700K Legends: $109.00 USD Masters: $119.00 USD Apprentices: $134.00 USD
|
|
#
?
Mar 9, 2017 00:09
|
|
|
Is it silly to anyone else that it is a better deal for Retail Edge people then Intel employees?
|
|
#
?
Mar 9, 2017 00:19
|
|
|
EdEddnEddy posted:Is it silly to anyone else that it is a better deal for Retail Edge people then Intel employees? The thing is if you're not a Master level, the price isn't work it. You spend almost $20 on shipping and then you have sales tax. You could get a 7700k after taxes for less than the apprentice level pricing.
|
|
#
?
Mar 9, 2017 00:35
|
|
|
SlayVus posted:The thing is if you're not a Master level, the price isn't work it. You spend almost $20 on shipping and then you have sales tax. You could get a 7700k after taxes for less than the apprentice level pricing. Maybe you were looking at the 6850k pricing by mistake? edit: oh, I'm confused. I misread and thought the deal from last holiday was the current deal. I wonder why the current deal is so much worse than historical deals. Is the current deal only on 7700k? EdEddnEddy posted:Is it silly to anyone else that it is a better deal for Retail Edge people then Intel employees? Col.Kiwi fucked around with this message at 01:08 on Mar 9, 2017 |
|
#
?
Mar 9, 2017 01:03
|
|
|
134 is for the 6700k. 7700k is $280 apprentice.
|
|
|
#
?
Mar 9, 2017 01:06
|
|
|
Watermelon Daiquiri posted:134 is for the 6700k. 7700k is $280 apprentice. Which is only a $19 savings over Micro Center's MSRP, and Intel's probably going to charge sales tax as well since they've got property in a fair amount of states.
|
|
#
?
Mar 9, 2017 01:13
|
|
|
BIG HEADLINE posted:Which is only a $19 savings over Micro Center's MSRP, and Intel's probably going to charge sales tax as well since they've got property in a fair amount of states.
|
|
#
?
Mar 9, 2017 01:14
|
|
|
I thought of an interesting question: Why can't CPU's run at even lower frequencies to save more power? Mine runs at 800 MHz at lowest.
|
|
#
?
Mar 9, 2017 02:17
|
|
|
SinineSiil posted:I thought of an interesting question: Why can't CPU's run at even lower frequencies to save more power? Mine runs at 800 MHz at lowest. Past a certain point, reducing the operating frequency won't save enough power to be measurable, since the supporting electronics still need a certain amount of power to operate anyway. Modern CPUs will temporarily shut down whole cores in situations where the clock has already hit the minimum, because that way will save power in a more effective way than say, trying to reduce the clock speed to 200 MHz. Also, depending on the hardware it's connected to, reducing all the CPU cores to too slow a speed might cause problems for communicating with other system components that don't downclock as far, like RAM or the expansion bus cards. This is less of a problem nowadays than it used to be.
|
|
#
?
Mar 9, 2017 02:25
|
|
|
fishmech posted:Past a certain point, reducing the operating frequency won't save enough power to be measurable, since the supporting electronics still need a certain amount of power to operate anyway. Modern CPUs will temporarily shut down whole cores in situations where the clock has already hit the minimum, because that way will save power in a more effective way than say, trying to reduce the clock speed to 200 MHz. Thanks for such good explanation!
|
|
#
?
Mar 9, 2017 02:30
|
|
|
EdEddnEddy posted:Is it silly to anyone else that it is a better deal for Retail Edge people then Intel employees? What's sillier is that once you apply the bundle discount with a motherboard, Microcenter is the better deal. That 7700k i just not a great deal unless you're a Master or whatever. I haven't been in retail for a while though, so I don't really care.
|
|
#
?
Mar 9, 2017 02:35
|
|
|
|
| # ? May 30, 2024 15:46 |
|
|
EdEddnEddy posted:Is it silly to anyone else that it is a better deal for Retail Edge people then Intel employees? SinineSiil posted:I thought of an interesting question: Why can't CPU's run at even lower frequencies to save more power? Mine runs at 800 MHz at lowest. The important concept here is the "race to idle," any "active" state is going to be so much more power hungry than parking everything at C1E or lower that you want the core to get whatever work it has done so it can go to sleep. If the only thing the system is doing is updating a flashing cursor, it can be sleeping for huge fractions of a second between each little update. I think you can check for core and platform C/S states with perfmon, but it can get a little tricky because "plot these metrics live" can be more work than you were otherwise doing. There would be more complications to making a super-slow core. The uncore is going to be chugging along full speed, and validating cases where you have a x16 PCIe card shoveling data into a 100MHz core will cause backpressure issues that wouldn't exist at 800MHz.
|
|
#
?
Mar 9, 2017 02:37
|
|
