|
DrDork posted:Not really, it's unlikely we'll see the types of performance leaps we did with things like C2D -> SandyBridge any time soon; the focus has largely been on efficiency rather than flat out performance. Maybe the next gen or two will see 6C/12T chips finally filtering down into the <$400 price range, but otherwise ~10% per gen improvement seems to be about where Intel wants to keep things.
|
# ? Jul 14, 2017 14:13 |
|
|
# ? May 10, 2024 01:03 |
|
Yeah, that's kinda my point, though: jumping from even a late-stage C2D to SandyBridge was a huge performance gain: Going from a E8500 to a i5-2500, you got about 40-50% faster single-thread performance, and up to double the multi-thread performance (since you went from 2C->4C), and that's before overclocking. Jumping from SandyBridge to Skylake is, comparatively, not nearly as significant a performance jump. This is despite almost twice as much time elapsing between the release of the 2500 and 7500 as between the E8500 and the 2500. The addition of the 2500k still being "more than fast enough" for most users at most things also means that the draw of a only moderately faster chip is pretty muted, compared to back in the early 2000's when a 20-30% gain often translated into clearly discernible benefits to the user experience.
|
# ? Jul 14, 2017 14:45 |
|
Yeah, Intel hasn't been getting 10% better IPC each generation either - my Haswell Refresh at 4.8 is only about 20% faster than my SB@4.6. Although Skylake probably bumped up more than preceding generations because of DDR4, but then Kaby Lake didn't add anything.
|
# ? Jul 14, 2017 14:56 |
|
The single-threaded performance boost is inarguable, but the comparison between E8500 and i5-2500 is a bit disingenuous. The true predecessor to Sandy Bridge i5s would have been one of the Yorkfield Core 2 Quads. In this comparison, the core count does not increase, and when comparing the standard i5-2500, the ability to overclock is lost. A non-K Core i5 2500 is not nearly as competitive with modern processors as an overclocked i5-2500K is. Even then, those 5-10% dinks and dunks have added up over time and the surrounding platforms have improved quite a bit (DDR4, native NVMe support, native USB 3.0/3.1, etc).
|
# ? Jul 14, 2017 15:06 |
|
theperminator posted:I have never seen a home chipset that supports multiple sockets and massive amounts of ram. Skulltrail
|
# ? Jul 14, 2017 15:34 |
|
PBCrunch posted:The single-threaded performance boost is inarguable, but the comparison between E8500 and i5-2500 is a bit disingenuous. The true predecessor to Sandy Bridge i5s would have been one of the Yorkfield Core 2 Quads. It's irrelevant whether the E8500 or a Yorkfield was the direct predecessor to SandyBridge. The point is simply that a decade ago you could expect CPUs to take significant steps between generations, and now you can't. The only "big" improvement in the last several generations was Haswell's noticeably better power use, and that's more for the laptop crowd. The 3 years and small handful of ticks/tocks between the $260 E8500 and the $220 i5-2500k resulted in massive noticeable performance increases--including straight up IPC and single-thread performance. The comparison is perfectly fair in that they were in roughly the same price-point and market segment, just separated by time. The 6 years and correspondingly larger number of ticks/tocks between the i5-2500k and the i5-7500/7600k have not resulted in nearly the same performance boost. The comparison here, too, is fair, since they're again roughly the same price and market segment.
|
# ? Jul 14, 2017 15:46 |
|
PerrineClostermann posted:Skulltrail Yep. Also AMD Quad FX.
|
# ? Jul 14, 2017 15:55 |
|
PBCrunch posted:The single-threaded performance boost is inarguable, but the comparison between E8500 and i5-2500 is a bit disingenuous. The true predecessor to Sandy Bridge i5s would have been one of the Yorkfield Core 2 Quads. In this comparison, the core count does not increase, and when comparing the standard i5-2500, the ability to overclock is lost. i remember gta iv having a palpable perf improvement (going from 'playable but choppy' to 'buttery smooth') from my penryn q9550 to a 3570k. then again that's probably that game's borderline pathological mistreatment of hardware
|
# ? Jul 14, 2017 16:07 |
|
DrDork posted:Not really, it's unlikely we'll see the types of performance leaps we did with things like C2D -> SandyBridge any time soon; the focus has largely been on efficiency rather than flat out performance. Maybe the next gen or two will see 6C/12T chips finally filtering down into the <$400 price range, but otherwise ~10% per gen improvement seems to be about where Intel wants to keep things. When a (now ex-) Intel employee warns that we aren't likely to ever see significant single-threaded performance gains from future x86 processor generations, we should probably listen. His angle suggests that Intel isn't meting out some fixed percentage of a theoretical "architectural IPC gains" pool with each generation; it's more likely pursuing clock speed boosts and process improvements à la Kaby Lake because those are the lowest-hanging fruit it has left. AMD's IPC gains with Ryzen are atypically large and will likely never be repeated, either. If someone is sitting on their Sandy Bridge CPU waiting for another huge 1T performance leap to justify an upgrade, they'll probably be sitting on it until it dies. It's far more likely that many-core, many-thread (8+) chips will become more attainable with time and software developers will simply have to find ways of taking advantage where they can.
|
# ? Jul 14, 2017 16:11 |
|
Indeed; physics is a hell of thing to be trying to fight against. I think that "never" would be suspiciously final, but it would probably take a significant shift outside of simple architecture (like switching to a different substrate or other revolutional instead of evolutional physical process changes) to make a real night-and-day increase at this point.
|
# ? Jul 14, 2017 16:23 |
|
DrDork posted:It's irrelevant whether the E8500 or a Yorkfield was the direct predecessor to SandyBridge. The point is simply that a decade ago you could expect CPUs to take significant steps between generations, and now you can't. The only "big" improvement in the last several generations was Haswell's noticeably better power use, and that's more for the laptop crowd. Even Haswell's better power use was just a noticeable inflection point of performance improvement and market uptake for products that were available before. Ivy and Sandy Bridge had 15W -U processors, but most manufacturers still used the -M models for those generations. Looking at Kaby Lake, the 6-9.5W -Y processors are starting to get more respect as being enough for a "standard" laptop and over time we are seeing more systems like the 12" Macbook and Dell/Lenovo's 2-in-1s that use them. We may well see a similar transition from -U to -Y in the traditional laptop form factor over the next few years. This also helps Xeons - the same number of cores in a smaller power envelope also gets you more cores in the same power envelope, which is why we've gone from 4 core Xeons in the Core 2 era to 28 with Skylake-E. It's just not as exciting for desktops because you would like to have proportionally more single-threaded performance instead and physics gets in the way of running processors at 10GHz on air.
|
# ? Jul 14, 2017 16:28 |
|
if IPC really is stuck, id be pretty happy with more power/thermal improvements. im one of those weirdos who gets upset about noise a lot. its pretty cool that you can get high end gaming performance and 8 threads in 65w today. i remember opening the box to my first pentium (a 100 i think?) and scoffing because it needed a FAN. what an unoptimized CAVEMAN product!!!!
|
# ? Jul 14, 2017 17:16 |
|
Cygni posted:if IPC really is stuck, id be pretty happy with more power/thermal improvements. im one of those weirdos who gets upset about noise a lot. its pretty cool that you can get high end gaming performance and 8 threads in 65w today. agreed Paul MaudDib fucked around with this message at 18:02 on Jul 14, 2017 |
# ? Jul 14, 2017 17:22 |
|
Cygni posted:if IPC really is stuck, id be pretty happy with more power/thermal improvements. im one of those weirdos who gets upset about noise a lot. its pretty cool that you can get high end gaming performance and 8 threads in 65w today. Long ago I threw a Corsair H110 on and have never looked back. With decent fans, I can run it silently and still capably cool an overclocked 5820k without issue.
|
# ? Jul 14, 2017 17:27 |
|
only fools use water cooling, intellectuals (me) use the more convenient noctua nh-d15s
|
# ? Jul 14, 2017 17:40 |
|
How much torque can a motherboard typically handle?
|
# ? Jul 14, 2017 17:51 |
|
PerrineClostermann posted:How much torque can a motherboard typically handle? So much torque, the substrate twisted coming off the line
|
# ? Jul 14, 2017 17:56 |
|
Microcenter is starting to clearance 2011-3 hard. Just got an ad showing $300 for a 6850K or two for $500 (my local store is sold out)
|
# ? Jul 14, 2017 18:12 |
|
Really though? This sounds a lot like the netburst Pentium 4 debacle that Intel got stuck with. From what I remember their Israel engineers had a Pentium M architecture which derived more from the P3 era. That architecture turned into the Core series. If I am wrong, feel free to yell at me.
|
# ? Jul 14, 2017 18:17 |
|
PerrineClostermann posted:How much torque can a motherboard typically handle? Way way more than the D15 offers. There are solid copper clones of it that people don't have problems with breaking motherboards. The new thinner Kabylake substrate breaks before the motherboard when people have shipped PCs with poor heatsink mounting solutions. (No reports of Noctuas breaking them yet though, they have basically the best mount around) As someone who cares about noise, but doesn't want to compromise on cooling performance I've been using Noctuas for years, I've tried Corsairs and they perform similarly but just take too much airflow to meet what counts as silent for the noise floor in my house. I don't think anyone would seriously argue the H110 is less convenient though. You can fit it into basically any weird case with no difficulty.
|
# ? Jul 14, 2017 18:53 |
|
Paul MaudDib posted:Microcenter is starting to clearance 2011-3 hard. Just got an ad showing $300 for a 6850K or two for $500 (my local store is sold out)
|
# ? Jul 14, 2017 18:58 |
|
DrDork posted:It's irrelevant whether the E8500 or a Yorkfield was the direct predecessor to SandyBridge. The point is simply that a decade ago you could expect CPUs to take significant steps between generations, and now you can't. The only "big" improvement in the last several generations was Haswell's noticeably better power use, and that's more for the laptop crowd. From the perspective of people who would remember 8088 > 80286 > 386 > 486 jumps in PCs, C2Q to Sandy Bridge looks downright laughable in improvements. Especially on workloads that dealt with raw processing power. Generic Monk posted:i remember gta iv having a palpable perf improvement (going from 'playable but choppy' to 'buttery smooth') from my penryn q9550 to a 3570k. then again that's probably that game's borderline pathological mistreatment of hardware I think you mean, the game's reliance on high core counts, which has worked out very well for it in the end.
|
# ? Jul 14, 2017 19:19 |
|
craig588 posted:Way way more than the D15 offers. There are solid copper clones of it that people don't have problems with breaking motherboards. The new thinner Kabylake substrate breaks before the motherboard when people have shipped PCs with poor heatsink mounting solutions. (No reports of Noctuas breaking them yet though, they have basically the best mount around) Where are these solid copper clones of the D15 outside of what Cryorig is trying to do this year with their Cu Series
|
# ? Jul 14, 2017 21:00 |
|
Idly, is it possible to
|
# ? Jul 14, 2017 21:15 |
|
Sidesaddle Cavalry posted:Where are these solid copper clones of the D15 outside of what Cryorig is trying to do this year with their Cu Series I guess they never came out or quickly shut down production. I remember someone sending me a link to a picture of one a couple years ago, but it doesn't seem to exist anymore.
|
# ? Jul 14, 2017 21:33 |
|
Malloc Voidstar posted:only fools use water cooling, intellectuals (me) use the more convenient noctua nh-d15s What do they think that second fan is really going to be doing for them?
|
# ? Jul 14, 2017 21:49 |
|
BangersInMyKnickers posted:What do they think that second fan is really going to be doing for them? either fan alone wouldn't really push the same amount of air through both separated heat sinks, I wouldn't think
|
# ? Jul 14, 2017 21:51 |
|
BangersInMyKnickers posted:What do they think that second fan is really going to be doing for them? It's big enough that you don't need a 3rd fan because it lines up with your existing exhaust fan (You are running a x38 Panaflo, right?). You can run all 3 at 600RPM and still overclock to 4.5GHz or whatever. Taking out a fan means the others have to run faster to meet the same cooling performance. More and bigger fans mean less noise.
|
# ? Jul 14, 2017 21:55 |
|
BangersInMyKnickers posted:What do they think that second fan is really going to be doing for them?
|
# ? Jul 14, 2017 21:56 |
|
Sidesaddle Cavalry posted:
Direct answer: Probably. There's 2 M.2 ports (which can each be broken out to a PCIe 3.0x4) and a Thunderbolt port so in theory you have plenty of expansion capability, you could just plug in whatever you wanted in a PCIe enclosure. A USB adapter would be less preferable since it involves more CPU overhead. But, the definitive feature of Skull Canyon is its graphics processor (Crystalwell Iris Pro 580) and you are shelling out a substantial premium for it. Why would you want to use a graphics-oriented platform for a pfsense box? There are dedicated mini-ITX and thin-ITX boards for LGA1150 and LGA1151 that provide more NICs and other features that are more relevant to a pfsense box. I actually just posted in the Home Networking thread about one I ran into while window-shopping on Newegg, check this puppy out (and it can be expanded with daughterboards for even more NICs if you want!) You can get full-passive thin-ITX or mini-ITX cases as well, so I think you'd want to justify why you would want to go with Skull Canyon over the alternatives. There's also the Netgate SG-1000 or PC Engines APU2, both of which are custom-designed for these tasks (also wifi AP or cellular bridge in the case of the PC Engines boards). Paul MaudDib fucked around with this message at 06:17 on Jul 15, 2017 |
# ? Jul 14, 2017 22:13 |
|
Aha. The passive thin-ITX cases look a lot nicer, too. Would be cool to see one of those boards with a nice switch for some of those RJ-45s instead of individual NICs for every port but that's not really the kind of load one would use pfSense for
|
# ? Jul 14, 2017 22:24 |
|
craig588 posted:It's big enough that you don't need a 3rd fan because it lines up with your existing exhaust fan (You are running a x38 Panaflo, right?). You can run all 3 at 600RPM and still overclock to 4.5GHz or whatever. Taking out a fan means the others have to run faster to meet the same cooling performance. More and bigger fans mean less noise. Putting fans in front of them isn't additive to their airflow, you're creating back pressure by shoving the air in to the blades of the next fan spinning at the same rate. There's not nearly enough material obstruction or distance to justify any of that and its idiotic.
|
# ? Jul 14, 2017 22:29 |
|
Sidesaddle Cavalry posted:Aha. The passive thin-ITX cases look a lot nicer, too. Would be cool to see one of those boards with a nice switch for some of those RJ-45s instead of individual NICs for every port but that's not really the kind of load one would use pfSense for If you wanna chat about this further, we should probably take it to the SFF/mITX or Home Network threads as applicable. But I am actually thinking fairly seriously about getting one of those Akasa cases even if it's wider than I'd prefer. My fiance and I both really like that spaceship heatsink design, and it has Thunderbolt 3 as an expansion option for an external GPU. There's only one other mITX/thin ITX board with Thunderbolt, and that one is LGA1151 not 1150, so I couldn't use a Crystalwell 5775C for the Iris Pro 580. So any other similarly-tiny option drops the graphics quality. There is a Skylake with the Iris Pro 580 (6785R), but it's a BGA chip, not socketed. I wonder if Intel is ever going to do a Skull Canyon followup now that Kaby Lake has the updated iGPU that supports Netflix's new DRM - that's my only hangup on Skull Canyon. Paul MaudDib fucked around with this message at 22:38 on Jul 14, 2017 |
# ? Jul 14, 2017 22:31 |
|
BangersInMyKnickers posted:Putting fans in front of them isn't additive to their airflow, you're creating back pressure by shoving the air in to the blades of the next fan spinning at the same rate. There's not nearly enough material obstruction or distance to justify any of that and its idiotic. You're certainly not going to get a huge performance bump, but since it's not a sealed system, most of that back-pressure just leaks out the sides, so you will get an increase in airflow/cooling on at least the front radiator surface. Whether the difference is worth the $20 for another fan is a different issue (probably not). I still love my H110, though. It's silent, keeps the interior of my case nice and tidy, and it allows me to vent the heat straight up out of the top of my case. This last part is super important during the winter because my cat loves lounging on top of the case, and really enjoys the built in heating pad for her.
|
# ? Jul 14, 2017 22:55 |
|
Paul MaudDib posted:I wonder if Intel is ever going to do a Skull Canyon followup now that Kaby Lake has the updated iGPU that supports Netflix's new DRM - that's my only hangup on Skull Canyon. Get an NVIDIA Shield TV for your 4k Netflix and enjoy that sick passive Skull Canyon setup for everything else.
|
# ? Jul 14, 2017 23:03 |
|
BangersInMyKnickers posted:Putting fans in front of them isn't additive to their airflow, you're creating back pressure by shoving the air in to the blades of the next fan spinning at the same rate. There's not nearly enough material obstruction or distance to justify any of that and its idiotic. It's quieter, that's good enough for me. Edit: I was checking with people to make sure I hadn't missed something and there was a new generation of quieter single fan options and got reminded of this https://www.kickstarter.com/projects/1489140137/nsg-s0-worlds-first-fanless-chassis-for-high-perfo I don't do kickstarter preorders but I'll probably be early to get one of those if the reviews come in and prove to be as good as the previews and they show up for sale at Newegg or Amazon. Around the same cost as a good CPU+GPU water loop but without the maintenance. craig588 fucked around with this message at 01:28 on Jul 15, 2017 |
# ? Jul 15, 2017 00:40 |
|
SamDabbers posted:Get an NVIDIA Shield TV for your 4k Netflix and enjoy that sick passive Skull Canyon setup for everything else. Netflix DRM is one thing but Kaby's a little more efficient than Skylake especially in mobile SKUs. I just would not be that surprised to see a drop-in replacement with a refreshed chip soon, and I'm nervous about ending up with "the lovely one". Especially if Intel goes up against some decent APUs (although loving lol about the chances of that given the apparent performance of the current Vega part - hopefully AMD has fixed them in their Vega 11 die). I mean, there's other options for a super small TV PC. I could get something for the back with a mITX or thin-ITX, and the 5775C still exists (kinda). Just not with Thunderbolt, and often kinda loud. Or I can go to a bigger case with a dGPU, which is also a step upwards in performance, but high-power tiny builds suck rear end to do (did a 4690K with a H75 AIO and a 780 Ti EVGA SC FE-style blower card in a RVZ01, worked fine but I removed+sold the 780 Ti a while back). They suck even worse with non-1151 or 1150 parts, 2011-3 or AM4 would be hell to build and even worse to debug if you had issues with PCIe so your display was down. That led me to looking at passive builds since the design and layout are much more pre-ordained than a from-parts mITX build. You can fit mITX size with a thermal envelope of X watts, you can fit a GPU with a thermal envelope of Y watts, you design around it, connect this heatpipe system and go. There's a mATX chassis that can do 90W+90W with a mITX mobo, that's enough for a slightly underclocked Haswell-E/BW-E or Ryzen 7, and you can underclock GPUs as far as you want. My 1080 hits close to 1070 performance at those thermal envelopes, you could also underclock a 1080 Ti and hit probably 1080 performance. It's a conceptually cool workstation build but we don't have space for it, a 200W passive PC is pretty big (17" wide - almost rackmount sized). That Akasa Galactico case gets me a nice thin mini-PC with a decent iGPU, but without the loud fans of the original Skull Canyon design, and Thunderbolt for future expansion with an eGPU if we want. And that case is pretty loving rad, me and the boss really like the angular heatsink design, please excuse our terrible taste (downside: I don't think you could stand it up on end, you would have to wall-mount it if you wanted to get it out of the way) If the Skull Canyon replaces our living room PC (the RVZ01) then she could have that for her photo editing PC, and I'll toss a 1060 or 480 in the RVZ01 for her display and gaming. On paper I could rebuild the RVZ01 as a Ryzen 7 case for photo editing but there's just no way I'm taking that thing apart and upgrading the motherboard ever, it's just too loving fiddly. I'll sell the thing and buy another one before I take it apart. Tiny high-power builds can work but they suck rear end to plan and do - which is one of the things that makes the Skull Canyon case appealing, it's a kit made for a specific model. Paul MaudDib fucked around with this message at 06:20 on Jul 15, 2017 |
# ? Jul 15, 2017 04:02 |
|
Paul MaudDib posted:That led me to looking at passive builds since the design and layout are much more pre-ordained than a from-parts mITX build. You can fit mITX size with a thermal envelope of X watts, you can fit a GPU with a thermal envelope of Y watts, you design around it, connect this heatpipe system and go. There's a mATX chassis that can do 90W+90W with a mITX mobo, that's enough for a slightly underclocked Haswell-E/BW-E or Ryzen 7, and you can underclock GPUs as far as you want. My 1080 hits close to 1070 performance at those thermal envelopes, you could also underclock a 1080 Ti and hit probably 1080 performance. That owns really hard. I wonder if there is any way to connect the heatsink bare - die to skylake or kaby.
|
# ? Jul 15, 2017 04:27 |
|
Your avatar reminds me of the other selling point: your cat is gonna loving love this thing This also sounds like a good justification to get yourself a better monitor stand/arm so you don't have to stand the monitor on the PC and subsequently have it destroyed in a furtive attempt to get to the warm.
|
# ? Jul 15, 2017 04:34 |
|
|
# ? May 10, 2024 01:03 |
|
Paul MaudDib posted:Your avatar reminds me of the other selling point: your cat is gonna loving love this thing She totally would - it's an ideal cat warmer That case reminds me of the old Zalman TNN500 - I wanted one of those a lot, and missed it. I'll get one of these for my next build. Thanks for the heads up, I didn't think manufacturers would make another passive case for desktop hardware, and this owns.
|
# ? Jul 15, 2017 05:18 |