|
Krailor posted:The actual driver for chip size is pin count; not cores. If you look at a de-lidded chip the die itself is actually a fairly small part of the overall size of the chip. As we get to smaller and smaller processor nodes the overall die size should continue to shrink. Not only is this true, but it actually has negative implications for cooling, since now you're talking more heat concentrated in a smaller area, and past a good TIM/copper cold plate there's already not a ton you can do to promote thermal transfer without getting pretty exotic.
|
#
?
Oct 17, 2016 20:35
|
|
|
|
| # ? Apr 23, 2024 23:26 |
|
|
Which chipset architecture is going to have Intel's support for machine-learning instructions? Is it Kaby Lake or Cannonlake, or?
|
|
#
?
Oct 17, 2016 20:40
|
|
|
ufarn posted:Which chipset architecture is going to have Intel's support for machine-learning instructions? Is it Kaby Lake or Cannonlake, or? Well, we don't have anything confirmed yet but it looked like a further extension of AVX-512. What this means is it might not necessarily make it to desktop Cannonlake ever, and be more of a Knight's Landing / Skylake-EP / future server and HPC chips rather than desktop CPUs.
|
|
#
?
Oct 17, 2016 21:04
|
|
|
DrDork posted:Not only is this true, but it actually has negative implications for cooling, since now you're talking more heat concentrated in a smaller area, and past a good TIM/copper cold plate there's already not a ton you can do to promote thermal transfer without getting pretty exotic. Some of the really cool microfluidics stuff might end up being present in high end head spreaders in a few years, basically a flat plate shaped heat pipe, only about 50x more expensive to fabricate. That or some of the graphene composites they're experimenting on, they can have some completely ridiculous thermal conductivities, shame they're even more expensive. Given how big the die itself is, I wonder if we might see the removal of the heatspreader and direct heatsink contact again on retail parts.
|
|
#
?
Oct 17, 2016 21:46
|
|
|
Methylethylaldehyde posted:Given how big the die itself is, I wonder if we might see the removal of the heatspreader and direct heatsink contact again on retail parts. Is the actual die big, though? I thought the consensus is that the package is huge due to the pin-out, but that the die itself was probably still quite compact. I kinda doubt we'll see naked dies in any event, though; these are gonna be pricey as gently caress, and largely aimed at enterprise type stuff where moderately higher temperatures can be combated by moderately faster fans for more noise. I mean, I don't see too many $2000+ chips ending up in home setups where people actually care about an extra few dB. And in exchange Intel gets fewer complaints about people breaking the fuckers one way or another. Graphene would be cool, but we've been playing with that poo poo for a decade and still no one can manufacture it on any sort of scale for prices that aren't eye-watering, so I'm not holding my breath on that one.
|
|
#
?
Oct 17, 2016 22:45
|
|
|
DrDork posted:Is the actual die big, though? I thought the consensus is that the package is huge due to the pin-out, but that the die itself was probably still quite compact. I suspect one reason the socket is so physically large due to Knight's Landing support, where they can make the chips as large as their large-die yields allows. Hopefully they prove me wrong and use the extra space in Xeons for some ridiculously large caches.
|
|
#
?
Oct 18, 2016 06:20
|
|
|
Chuu posted:I suspect one reason the socket is so physically large due to Knight's Landing support, where they can make the chips as large as their large-die yields allows. Hopefully they prove me wrong and use the extra space in Xeons for some ridiculously large caches. Behold the 40 core Xeon E7240 Codenamed 'gently caress you, pay me', with an eye watering 512MB of shared L4 cache. We'll probably see some interesting real world improvements in things when you can buffer huge chunks of data to cache before chewing on it. Just think of how fast multithreaded bubblesort would be with the entire dataset in L4 cache.
|
|
#
?
Oct 18, 2016 09:07
|
|
|
Twerk from Home posted:Is there any reason not to expect the pattern to continue that the distance between max turbo clock on a stock chip vs. reasonable overclocks achievable will shrink? If the i7-7700K is shipping with a 4.5GHz turbo, then 7600K overclocks should be in the 4.6-4.7 GHz range. That's still better than most 6600Ks get, as far as I've heard! Plenty of posters here have gotten 6600Ks up to 4.6, including me. I wouldn't be surprised if i could push it up to 4.7-4.8, but I'm sitting at 1.37 vcore and don't really want to go into the 1.4 range.
|
|
#
?
Oct 18, 2016 09:11
|
|
|
The codename for that 40-core chip would be "Classified" since the NSA and DoE would claim them all.
BIG HEADLINE fucked around with this message at 09:15 on Oct 18, 2016 |
|
#
?
Oct 18, 2016 09:12
|
|
|
BIG HEADLINE posted:The codename for that 40-core chip would be "Classified" since the NSA and DoE would claim them all. No, "Classified" would be the specially spec'd version that costs 3x as much for no meaningful gains because defense budgets.
|
|
#
?
Oct 18, 2016 11:47
|
|
|
let i hug posted:No, "Classified" would be the specially spec'd version that costs 3x as much for no meaningful gains because defense budgets. Uh, excuse me, but those chips would absolutely be different. You see, the government-use chips are painted black because reasons, and that makes them worth the extra price obviously.
|
|
#
?
Oct 18, 2016 12:53
|
|
|
When someone says they want a low-profile HSF, you just paint it black and hand it back to them.
|
|
#
?
Oct 18, 2016 13:05
|
|
|
For an extra 25% I'll anodize it vice spray-paint.
|
|
#
?
Oct 18, 2016 16:04
|
|
|
I don't know if y'all actually work with the Government on technology projects but blame the contractors for those markups, not the government. When obsolescence is a problem, things get expensive fast. That markup is the assumed risk on the part of the contractor.
|
|
#
?
Oct 18, 2016 21:35
|
|
|
And in a way isn't Tech stuff, enterprise or otherwise, usually bought and then used until it practically falls apart in a lot of Government? Cost a lot at the start, and even more in the long run to support a system 20+ years past its EoL date?
|
|
#
?
Oct 18, 2016 21:59
|
|
|
EdEddnEddy posted:And in a way isn't Tech stuff, enterprise or otherwise, usually bought and then used until it practically falls apart in a lot of Government? Cost a lot at the start, and even more in the long run to support a system 20+ years past its EoL date? You might be conflating user workstations and configuration-managed systems like ewar suites.
|
|
#
?
Oct 18, 2016 22:08
|
|
|
SuperDucky posted:I don't know if y'all actually work with the Government on technology projects but blame the contractors for those markups, not the government. Well, that and generally really lovely contracting agents who have minimal reason to actually shop around for the best prices, or are rail-roaded into one company/product type by management requirements that are so specific that mysteriously only one company qualifies. EdEddnEddy posted:And in a way isn't Tech stuff, enterprise or otherwise, usually bought and then used until it practically falls apart in a lot of Government? Cost a lot at the start, and even more in the long run to support a system 20+ years past its EoL date? Most small tech stuff in government/big business gets replaced on a fairly regular cycle (desktops/laptops often are swapped out every 2-3 years). The big back-end systems, on the other hand, are generally so mission-critical that no one wants to be The Guy who thought it'd be a good idea to futz with a system that's working perfectly well and have it break. So they don't get updated until they absolutely are forced to. Which, much like the difference between going to the doctor's for preventative health vice only going in when you think you might be dying, is a really great way to end up spending a poo poo ton of money on something that could have been addressed earlier for far less, and now has a much worse chance of actually turning out well. But hey, when you're just chasing next quarter's numbers...
|
|
#
?
Oct 18, 2016 22:21
|
|
|
DrDork posted:Well, that and generally really lovely contracting agents who have minimal reason to actually shop around for the best prices, or are rail-roaded into one company/product type by management requirements that are so specific that mysteriously only one company qualifies. College/uni that I used to go to was "upgraded" throughout entirely by fujitsu. What's that a printer needs a new cartridge? Whelp 30 days minimum until their qualified engineer comes to change it. No, you're not allowed to unplug it and move it out the way till it's fixed even if it is just a usb cable. So about a year down the line several computers, printers etc would get sticky notes of shame for several months until the magic repairman could come to change a cartridge or update software. Classes had to be split across multiple rooms so everyone had a working computer. This was also an IT establishment, so the tutors were super pissed when they were no longer allowed to do any maintenance whatsoever. They were also required (but often didn't bother) to get prior agreement to any 3rd party devices they would be using, such as wireless mouse pointer wand things or speakers. Went there for my networking degree which was taught to me literally using stick men. I got in trouble twice for drawing smiley faces on them 
|
|
#
?
Oct 18, 2016 22:53
|
|
|
Captain Hair posted:They were also required (but often didn't bother) to get prior agreement to any 3rd party devices they would be using, such as wireless mouse pointer wand things or speakers. The name for this class of device is "presenter". "I lost my wireless presenter and had to get another one."
|
|
#
?
Oct 18, 2016 23:45
|
|
|
Y'all are talking about what we call "prime contractors" in the biz. EdEddnEddy posted:And in a way isn't Tech stuff, enterprise or otherwise, usually bought and then used until it practically falls apart in a lot of Government? Cost a lot at the start, and even more in the long run to support a system 20+ years past its EoL date? Potato Salad posted:You might be conflating user workstations and configuration-managed systems like ewar suites. Yes, in both cases. Ironically, you're more likely to see upgrades to the more integrated things shipborne, airborne, etc over its lifetime than the end-user, non-actual-deployment-risk stuff over a system's projected lifetime. The government penny-pinchers are weird that way. You'll have 20 year old desktop-class equipment that does maintenance/health checks on state of the art, classified fire control stuff over serial or some bullshit like that. DrDork posted:Well, that and generally really lovely contracting agents who have minimal reason to actually shop around for the best prices, or are rail-roaded into one company/product type by management requirements that are so specific that mysteriously only one company qualifies. Captain Hair posted:literally third party, first line integrators.txt
|
|
#
?
Oct 19, 2016 00:58
|
|
|
Chuu posted:I suspect one reason the socket is so physically large due to Knight's Landing support, where they can make the chips as large as their large-die yields allows. Hopefully they prove me wrong and use the extra space in Xeons for some ridiculously large caches.
|
|
#
?
Oct 19, 2016 02:41
|
|
|
Captain Hair posted:College/uni that I used to go to was "upgraded" throughout entirely by fujitsu. This sounds absolutely horrible, as well as a great example of what happens when someone (or a committee) with minimal experience in writing contracts attempts to negotiate a contract with a company with lots of experience writing contracts. It's so easy to miss little things that turn out to be incredibly loving annoying later on (like for your uni, apparently no clause about maximum ticket-to-resolution times). Which is a big part of the issue: a lot of tech contracts are written or approved by people/committees with minimal experience reviewing contracts and often minimal understanding of the underlying issues the contract needs to address; i.e., they know they need printers, but often don't understand the need for, say, the ability to keep toner/fuzers on-hand for quick replacement.
|
|
#
?
Oct 19, 2016 02:55
|
|
|
japtor posted:There's also that thing about "optional accelerators" in the form of integrated FPGAs and IGPs and god knows what else. The Altera purchase raises a lot of questions. What was that quote about Saudi Princes upthread, though?
|
|
#
?
Oct 19, 2016 16:07
|
|
|
SuperDucky posted:The Altera purchase raises a lot of questions. What was that quote about Saudi Princes upthread, though? Saudi princes have assloads of cash and a habit of buying the most expensive thing regardless of whether its worth it. So the joke is that you probably wouldn't benefit enough to justify the cost.
|
|
#
?
Oct 20, 2016 03:51
|
|
|
fishmech posted:Saudi princes have assloads of cash and a habit of buying the most expensive thing regardless of whether its worth it. It still boggles my mind what they've done to GS:GO pretty much single-handedly. I wonder what other much less visible markets they've had the same effect in.
|
|
#
?
Oct 20, 2016 05:08
|
|
|
Yeah, I didn't mean anything by the 'Saudi Prince' comment other than I really kind of suspect the barrier to entry for a Sky-E system will be somewhere around $2.5k minimum for something that'll be worthwhile. Hell, I'm just guessing ballpark that the mid-range chip (with extra PCIe lanes) will be ~$799, before the motherboard, which will undoubtedly be probably $299 for the 'bare bones' version, then 6x8GB DDR4, which will be the *cheapest* element in the system. 
BIG HEADLINE fucked around with this message at 07:40 on Oct 20, 2016 |
|
#
?
Oct 20, 2016 07:36
|
|
|
The crazy Skylake EP socket made me wonder if it'd be too big and/or hot for the Mac Pro (...if it ever gets updated), then I remembered Skylake X (i.e. Skylake E). Looking it up again brought up Skylake W, which seems like more of a direct successor to the current 1-2 socket EPs (well, minus dual socket support). I'm guessing EP will be limited to servers and that ilk with W being for desktop workstations. That doesn't rule out some crazy rear end LGA 3647 desktop board to take advantage of super rich users, but I imagine Skylake X will be the basis of whatever "normal" super high end system. I'm sure prices will still be ridiculous either way though.
|
|
#
?
Oct 20, 2016 08:59
|
|
|
BIG HEADLINE posted:Yeah, I didn't mean anything by the 'Saudi Prince' comment other than I really kind of suspect the barrier to entry for a Sky-E system will be somewhere around $2.5k minimum for something that'll be worthwhile.
|
|
#
?
Oct 20, 2016 12:34
|
|
|
fishmech posted:Saudi princes have assloads of cash and a habit of buying the most expensive thing regardless of whether its worth it. So the joke is that you probably wouldn't benefit enough to justify the cost.  japtor posted:The crazy Skylake EP socket made me wonder if it'd be too big and/or hot for the Mac Pro (...if it ever gets updated), then I remembered Skylake X (i.e. Skylake E). Looking it up again brought up Skylake W, which seems like more of a direct successor to the current 1-2 socket EPs (well, minus dual socket support). I'm guessing EP will be limited to servers and that ilk with W being for desktop workstations. See also BIG HEADLINE posted:Yeah, I didn't mean anything by the 'Saudi Prince' comment other than I really kind of suspect the barrier to entry for a Sky-E system will be somewhere around $2.5k minimum for something that'll be worthwhile. Maybe. For a low-end single-socket build. Maybe. SuperDucky fucked around with this message at 23:30 on Oct 20, 2016 |
|
#
?
Oct 20, 2016 23:24
|
|
|
You know, I check this deal website daily...HP is the only vendor sleazy enough to be marketing the *dual core* mobile Kaby Lakes in volume and selling them for more than even some mobile i7 Skylake quads are selling for because *seven is more than six*!
|
|
#
?
Oct 21, 2016 02:23
|
|
|
BIG HEADLINE posted:You know, I check this deal website daily...HP is the only vendor sleazy enough to be marketing the *dual core* mobile Kaby Lakes in volume and selling them for more than even some mobile i7 Skylake quads are selling for because *seven is more than six*! It's not about this at all, though. If you look at the list price of CPUs from Intel, the quad cores and dual cores are the same price. The i7-7Y75 is a 4.5W dual core core m part with a $393 list price. The difference between dual and quad core is almost entirely a question of thermals. List price on a quad core i5-6300hq is $250, while an i5-7200U which is a dual core costs $281. The laptop OEMs aren't out to screw you, it's just that quad vs dual core isn't something that makes chips more expensive. Intel also increased list prices with Kaby Lake, so the dual core Kaby Lake CPUs are outright more expensive than even i7 quad core Skylakes. Twerk from Home fucked around with this message at 02:41 on Oct 21, 2016 |
|
#
?
Oct 21, 2016 02:37
|
|
|
Ah. So one could say that Intel's finally decided to test the waters in regards to charging more, since AMD's not making any substantive noise about how incredible Zen is going to be?
|
|
#
?
Oct 21, 2016 03:00
|
|
|
they've always been doing this since ivy bridge 4.5-28w dual core "big arch" SKUs are always way more expensive than 45w quad core SKUs, which themselves are way more expensive than desktop SKUs there's a reason why the desktop chips are cheaper, namely being leaky poo poo
|
|
#
?
Oct 21, 2016 03:03
|
|
|
BIG HEADLINE posted:Ah. So one could say that Intel's finally decided to test the waters in regards to charging more, since AMD's not making any substantive noise about how incredible Zen is going to be? What you're really seeing is confidence from the OEMs that most users don't need 4 cores. Dual core laptops are thinner, cooler, quieter, and have better battery life.
|
|
#
?
Oct 21, 2016 03:13
|
|
|
When are Apollo Lake boards (eg Asrock J4205) actually going to be in stores for purchase and will they compete with the Athlon 5350 performance wise for lightweight Linux workstation usage (LUbuntu with IntelliJ)?
|
|
#
?
Oct 28, 2016 02:29
|
|
|
Paul MaudDib posted:When are Apollo Lake boards (eg Asrock J4205) actually going to be in stores for purchase and will they compete with the Athlon 5350 performance wise for lightweight Linux workstation usage (LUbuntu with IntelliJ)? I've seen these kinds of things used as low power NASes or media centers, I'd think that with any type of compiling code and especially with heavy IDEs like IntelliJ you'd want something with a good bit faster single threaded. I'm also interested in the segment, though and excited to see Apollo Lake!
|
|
#
?
Oct 28, 2016 02:42
|
|
|
Twerk from Home posted:I've seen these kinds of things used as low power NASes or media centers, I'd think that with any type of compiling code and especially with heavy IDEs like IntelliJ you'd want something with a good bit faster single threaded. I'm also interested in the segment, though and excited to see Apollo Lake! Ah, well. I've used Lubuntu before and it really helped on trashtastic systems (my Compaq CQ56 with an AMD V140 processor, which amounts to the last dying gasp of the single-threaded Athlon 64 at 2.3 GHz). I'd hoped that IntelliJ might be OK on a 5350. So you think I'm more in the market for something like a i3-6100 then? MicroCenter has those for $110, plus the cheapest H110 motherboards are $70, minus the $30 discount. So about $150 before tax, vs $30 for the 5350. I wish there wasn't such a big jump to the i5s. All they have is the i5-6500 starting at $180 and I'd probably just say gently caress it and jump up to the 6600K for $200 at that point. Not gonna do AMD's desktop chips though. What is the expected street date of Kaby Lake? Am I making a mistake by doing this now? As previously mentioned I do have a use for HEVC/H265 Main10 decode, and I'd rather have a processor with a clear path I could retire it into.
|
|
#
?
Oct 28, 2016 02:53
|
|
|
Paul MaudDib posted:Ah, well. I've used Lubuntu before and it really helped on trashtastic systems (my Compaq CQ56 with an AMD V140 processor, which amounts to the last dying gasp of the single-threaded Athlon 64 at 2.3 GHz). I'd hoped that IntelliJ might be OK on a 5350. People always overlook the big core Pentiums / Celerons for some reason. If you're after a good experience while pinching pennies, look at the Pentium G4400 and its ilk. The i3-6100 is a fine CPU and good value too though.
|
|
#
?
Oct 28, 2016 03:07
|
|
|
Kaby Lake isn't due until late December/early January, last I heard. The Z270 boards are ready and have already been showcased, but my guess is Intel doesn't want to put the new chip out when stock of Skylake-E still need clearing. Early reports show Kaby Lake is ~10-15% faster than Skylake in *some* instances, but it's an optimization step, not a wildly different chip, hence the fact that current-gen boards are drop-in ready.
|
|
#
?
Oct 28, 2016 03:39
|
|
|
|
| # ? Apr 23, 2024 23:26 |
|
|
I was trying to work out why my computer was crashing on even small overclocks. I also think I found out why my system was running so poo poo before, the fan on my big cooler has seized. I'm going to try getting it fixed, but just in case how much of an overclock can I expect on the fan that I get with the i5 4690k CPU?
|
|
#
?
Oct 28, 2016 07:11
|
|