|
I was assuming 1700s did 3.9 and 1800xes did 4GHz on average, but it's still not much of a difference. 1700s doing 3.8GHz and 1800xes doing 3.9GHz is still less than a 5% improvement for a massive jump in cost. Even at only 3.8GHz I'd still recommend a 1700 if someone actually could use 8 cores. With the embargo lifting soon on Coffeelake that might be about to change, but for now it's a good cheap option for 8 real cores even if each core is slower.
|
# ? Sep 29, 2017 20:32 |
|
|
# ? May 28, 2024 05:42 |
|
Nvidia's bro CEO cracks me up every time I see him strutting around in skinny jeans or leather jackets
|
# ? Sep 29, 2017 20:36 |
|
craig588 posted:I was assuming 1700s did 3.9 and 1800xes did 4GHz on average, but it's still not much of a difference. 1700s doing 3.8GHz and 1800xes doing 3.9GHz is still less than a 5% improvement for a massive jump in cost. Even at only 3.8GHz I'd still recommend a 1700 if someone actually could use 8 cores. With the embargo lifting soon on Coffeelake that might be about to change, but for now it's a good cheap option for 8 real cores even if each core is slower. Yeah, I'm just saying that you are definitely getting a better bin with 1700X/1800X so it's not a total waste of money. 1700s are fully-functional but the ASIC quality is not real good.
|
# ? Sep 29, 2017 20:52 |
|
Paul MaudDib posted:Yeah, I'm just saying that you are definitely getting a better bin with 1700X/1800X so it's not a total waste of money. 1700s are fully-functional but the ASIC quality is not real good. The bins are still "capped" because the best chips go into TR, so you don't even have the chance at a perfect 1800X sample. Another romanian site published a Coffee Lake review. 5.0 Ghz at 92°C with a Corsair H115 AIO at full fanspeed but cooler than a 7700K at stock frequencies which is impressive considering +2 cores and -100 Mhz. http://translate.google.com/translate?u=https%3A//wasd.ro/hardware-reviews/mobo-cpu/intel-core-i7-8700k-review/%23verdict&hl=en
|
# ? Sep 30, 2017 14:50 |
|
eames posted:The bins are still "capped" because the best chips go into TR, so you don't even have the chance at a perfect 1800X sample. I've gotten worse Cinbench scores with 5.0 @ 90c vs 4.8 @ 80c so I'm guessing 92c is gonna be hitting the throttle limit somewhat. Maybe we'll start to see more 360 AIO's like the EK Predator 360? [edit] Huh. I would have thought the EK's 360mm gave better temps than Corsair's 280mm. Ak Gara fucked around with this message at 15:46 on Sep 30, 2017 |
# ? Sep 30, 2017 15:38 |
Ak Gara posted:I've gotten worse Cinbench scores with 5.0 @ 90c vs 4.8 @ 80c so I'm guessing 92c is gonna be hitting the throttle limit somewhat. The surface area difference between a 360 rad and a 280 rad is actually quite small, and the fans Corsair uses on the H110i GT can move a ton of air but get very loud doing so, so good top end performance at the cost of noise.
|
|
# ? Sep 30, 2017 15:53 |
|
Sidesaddle Cavalry posted:Oh definitely. Here's a few craig588 posted:I think this is a cool one covering parts that even as a professional I have no direct experience with. (I skipped the youtuber introing the video, the actual video is all one of Nvidia's engineers giving a tour) Dang, these are great especially the Intel Sand to Silicon one. I'd love for them to release an HD version of these videos. Really fascinating stuff! Thanks!
|
# ? Sep 30, 2017 18:00 |
|
I'm conflicted about reviews that use the stock fans for all of the coolers, I have an H100i and the fans maxed out push a ton more air than my D15 fans but they're also terribly loud where I'd never use them like that. In contrast the maxxed out D15 fans are so quiet you can still barely hear them. On one hand how many people are going to replace whatever the stock fans that come with the cooler, but also how many people are going to run their cooler at 40+ dB 24/7? I don't have any EK coolers but based on how Corsair's stock fans are among the loudest fans I have I bet the EK fans are somewhat slower and much quieter considering EK is a more "premium" brand while Corsair is more "mainstream".
|
# ? Sep 30, 2017 18:05 |
craig588 posted:I'm conflicted about reviews that use the stock fans for all of the coolers, I have an H100i and the fans maxed out push a ton more air than my D15 fans but they're also terribly loud where I'd never use them like that. In contrast the maxxed out D15 fans are so quiet you can still barely hear them. On one hand how many people are going to replace whatever the stock fans that come with the cooler, but also how many people are going to run their cooler at 40+ dB 24/7? I don't have any EK coolers but based on how Corsair's stock fans are among the loudest fans I have I bet the EK fans are somewhat slower and much quieter considering EK is a more "premium" brand while Corsair is more "mainstream". IIRC EK uses Vardar fans which are quieter than the ones Corsair uses. The Corsair ones are still good fans, they didn't cheap out on them but with a max RPM of 2500 they are going to be loud at full tilt, the Vardars max out at something anywhere from 1150 to 2200 RPM so they will be quieter at max speed no matter what. The problem with replacing the fans is that almost all of these AIOs have very similar radiators, pumps and so on, so a lot of the differentiation is via the fans and other features like lighting, so if you replace the fans you have also gotten rid of most of the performance difference. Replacing the fans is also problematic because good fans are expensive, for instance a couple Corsair MLs will cost you $40-$50, so cheaper AIOs would look a lot better and the more expensive ones that come with nice fans would suffer, motivating the companies to include very cheap fans mostly to the detriment of the consumer.
|
|
# ? Sep 30, 2017 19:33 |
|
craig588 posted:I'm conflicted about reviews that use the stock fans for all of the coolers, I have an H100i and the fans maxed out push a ton more air than my D15 fans but they're also terribly loud where I'd never use them like that. In contrast the maxxed out D15 fans are so quiet you can still barely hear them. On one hand how many people are going to replace whatever the stock fans that come with the cooler, but also how many people are going to run their cooler at 40+ dB 24/7? I don't have any EK coolers but based on how Corsair's stock fans are among the loudest fans I have I bet the EK fans are somewhat slower and much quieter considering EK is a more "premium" brand while Corsair is more "mainstream". Corsair makes lots of fans, most of them are bad. However they do make the absolutely best perf/noise 120mm fans on the market, the ML120s. The ML140s are great but are merely as good as the Vardar 140s. The Pros and LED colored are just as good as the standards.
|
# ? Sep 30, 2017 19:57 |
|
I plan on getting the 8700k and maybe an AIO cooler to go with it if it runs hot like the 7700k. What is the go to Noctua fan for these AIO coolers if money is no issue
|
# ? Sep 30, 2017 20:54 |
|
Rabid Snake posted:I plan on getting the 8700k and maybe an AIO cooler to go with it if it runs hot like the 7700k. What is the go to Noctua fan for these AIO coolers if money is no issue Corsair ML120 or ML140, better in every single way as proven through rigorous testing, check thermalbench.com they are the only ones really doing proper fan/cooling testing right now. Noctuas not had anything competitive in the 120mm or 140mm space for a while.
|
# ? Sep 30, 2017 20:59 |
|
Rabid Snake posted:I plan on getting the 8700k and maybe an AIO cooler to go with it if it runs hot like the 7700k. What is the go to Noctua fan for these AIO coolers if money is no issue Noctua F12s and P12s are good. I think the F12s are louder but both models come with low speed adapters. They aren't new and innovative at this point but they were ahead of the game for years.
|
# ? Sep 30, 2017 21:32 |
|
rage-saq posted:Corsair ML120 or ML140, better in every single way as proven through rigorous testing, check thermalbench.com they are the only ones really doing proper fan/cooling testing right now. Noctuas not had anything competitive in the 120mm or 140mm space for a while. Is there a non-pro ML120? The dual pack seems affordable but can't tell if there's a difference from the singles labelled "Pro"
|
# ? Sep 30, 2017 21:41 |
|
The review I saw says that the only difference is the color and the non LED ones actually perform marginally better. The review also said the Corsair ML and the Noctua PPC 3000 fans perform within the margin of error of each other. I was almost ready to replace 7 Noctua fans with them, but it seems like the modern Noctuas still perform just as well. http://thermalbench.com/2016/07/29/corsair-ml140-pro-140-mm-fan/
|
# ? Sep 30, 2017 21:51 |
|
mewse posted:Is there a non-pro ML120? The dual pack seems affordable but can't tell if there's a difference from the singles labelled "Pro" The dual pack is non pro. No differences. [quote="“craig588”" post="“476924272”"] The review I saw says that the only difference is the color and the non LED ones actually perform marginally better. The review also said the Corsair ML and the Noctua PPC 3000 fans perform within the margin of error of each other. I was almost ready to replace 7 Noctua fans with them, but it seems like the modern Noctuas still perform just as well. http://thermalbench.com/2016/07/29/corsair-ml140-pro-140-mm-fan/ [/quote] Sure enough, I misremembered the ML140 review where the top fans are all kind of equal. The ML120 is a different story though.
|
# ? Oct 1, 2017 00:30 |
|
mewse posted:Is there a non-pro ML120? The dual pack seems affordable but can't tell if there's a difference from the singles labelled "Pro" The non-Pro MLs don't have rubberized sound/vibration dampeners built into the mount points. That's the only difference, other than the Pros that have LED lighting. Both the Pro and non-Pro carry a five year warranty. I bought five ML140 Pros for my case, but honestly the non-Pros are designed for case mounting (since most newer cases have some form of sound dampening built into the mounts) and the Pros are typically suggested for use on radiators, since anyone using watercooling obviously wants some semblance of quietude, and putting hard plastic against harder (and hollow) metal usually isn't a good way to get that. You can also buy rubber/silicon washers that can act as vibration dampeners for way cheaper than going with all Pros like I did. BIG HEADLINE fucked around with this message at 00:39 on Oct 1, 2017 |
# ? Oct 1, 2017 00:36 |
|
BIG HEADLINE posted:So evidently Intel was so jealous that Chris Roberts partnered with AMD way back when that they decided to get on board the Vapor Train: http://www.guru3d.com/news-story/new-intel-slides-reveal-product-roadmap-up-to-2018.html i so badly want this to be a portent of amd partnering with derek smart will never happen as that would require derek smart games to actually launch in order to be demoed but one can dream
|
# ? Oct 1, 2017 01:37 |
|
Generic Monk posted:i so badly want this to be a portent of amd partnering with derek smart At this point I suspect if Star Citizen ever launches, you'll have the option of playing it on your system or a ~specially tailored CLOSED system that's been specifically designed for MAXIMUM OVERDRIVE~ for a mere starting price of $2999.
|
# ? Oct 1, 2017 01:40 |
|
We'll probably be well into cloud gaming by the time something tangible from Star Citizen is available. Main expense will be their dumb flight controllers.
|
# ? Oct 1, 2017 01:44 |
|
ufarn posted:We'll probably be well into cloud gaming by the time something tangible from Star Citizen is available. Yes...but I'm sure the $4999 Ultra Mega Supreme with Guacamole Nebula Controller Box will come with a free 'pocket carrier' or some other capital ship that can't be gotten any other way.
|
# ? Oct 1, 2017 01:47 |
|
BIG HEADLINE posted:At this point I suspect if Star Citizen ever launches, you'll have the option of playing it on your system or a ~specially tailored CLOSED system that's been specifically designed for MAXIMUM OVERDRIVE~ for a mere starting price of $2999. are you implying that there will be an option other than the bespoke system? the game is going to crash on launch on anything that isn't croberts' jewel encrusted acer predator 21x. pimp chalice optional but highly encouraged
|
# ? Oct 1, 2017 02:08 |
|
Generic Monk posted:are you implying that there will be an option other than the bespoke system? the game is going to crash on launch on anything that isn't croberts' jewel encrusted acer predator 21x. pimp chalice optional but highly encouraged Probably. And seeing how I started this dangerous segue, let me suggest we end it before the thread becomes about ~that game~ instead of Intel managing to succeed in the consumer space despite repeatedly punching themselves in the balls whenever given the opportunity to do literally anything else.
|
# ? Oct 1, 2017 02:18 |
|
Sandybridge-E still holding up well in 1080p compared to SL-X' new cache layout and mesh architecture https://www.youtube.com/watch?v=wqv_6qhbHxw
|
# ? Oct 1, 2017 08:50 |
|
eames posted:Sandybridge-E still holding up well in 1080p compared to SL-X' new cache layout and mesh architecture God, Skylake-X is shockingly bad.
|
# ? Oct 1, 2017 12:42 |
|
BIG HEADLINE posted:Yes...but I'm sure the $4999 Ultra Mega Supreme with Guacamole Nebula Controller Box will come with a free 'pocket carrier' or some other capital ship that can't be gotten any other way. Maybe they'll do their own ICO while they're at it. God these people. I'm still not sure how you screw up a game this much.
|
# ? Oct 1, 2017 12:56 |
|
HalloKitty posted:God, Skylake-X is shockingly bad. I think the takeaway is Intel hasn't done much multi-core innovation in like 5 years.
|
# ? Oct 1, 2017 18:44 |
|
HalloKitty posted:God, Skylake-X is shockingly bad. If you look at gaming alone, sure, but that's a waste of a $1000 CPU (or even a $600 CPU). Tech Report recently added streaming benchmarks to their suite and the Skylake-X parts look awesome for single-machine streaming at 1080p60 (especially in the case of the i7-7820X versus Threadrippers). At least in their benchmarks, productivity and power efficiency are competitive with, if not superior to, AMD's stuff, too. "Shockingly bad" is for stuff like Vega 10. Anybody who cares about gaming performance exclusively doesn't have to spend more than the cost of an i7-7700K or i7-8700K to get good results. That's been the case at least since Haswell and the i7-4770K/4790K.
|
# ? Oct 1, 2017 22:49 |
|
Everyone pooh-poohs AVX512 as some niche thing but a surprising amount of the actual compute-intensive stuff (i.e the stuff you're buying a 10-18 core CPU for) actually can use it. x264, x265, and Blender all see huge IPC gains from AVX512 - to the extent that SKL-X is pushing 50-100% more performance-per-core in these tasks. It's basically just that games don't like the cache layout very much, so you can no longer have the ideal work-hard play-hard workstation anymore, you have to choose one or the other. It's the same situation Threadripper is in, performance usually isn't catastrophically bad but it's not what you expect from a $1000-2000 HEDT processor either (after all you can build a very nice second rig for gaming with $2000). Right now the only real answer there is "wait for Coffee Lake". The other thing is that AVX512 slurps power down like crazy, which is made even worse since a lot of sites are in the habit of benchmarking power using Prime95 (which is essentially as pure an AVX synthetic as it's possible to create). But you're also churning out a lot more throughput at those levels, and given a fixed task size it's not like you're going to see less total power consumption from running the task on general-purpose units. So it's pretty funny to see people be surprised by Skylake-X having equal or greater efficiency to Ryzen/TR products here. No poo poo, there's two halves to the efficiency equation, and you're ignoring half of it when you just look at power consumption between processors with such a drastic difference in IPC. Cinebench-per-watt (higher is better): Blender joules-vs-time (closer to origin is better): Paul MaudDib fucked around with this message at 23:38 on Oct 1, 2017 |
# ? Oct 1, 2017 23:21 |
|
They limited AVX 512 to only the 1000 dollar and up processors. I can't use 10 cores, but I might have been able to use 8 AVX 512 capable cores.
|
# ? Oct 1, 2017 23:33 |
|
craig588 posted:They limited AVX 512 to only the 1000 dollar and up processors. I can't use 10 cores, but I might have been able to use 8 AVX 512 capable cores. I don't think that's right. The 7800X lists AVX512 capability in the Ark, and Wikipedia agrees.
|
# ? Oct 1, 2017 23:37 |
|
Oh, that's interesting. I remember when they launched reviews made a big deal that anything below the 7900K didn't have AVX 512 support besides emulation that cut the speed in half.
|
# ? Oct 1, 2017 23:38 |
|
craig588 posted:They limited AVX 512 to only the 1000 dollar and up processors. I can't use 10 cores, but I might have been able to use 8 AVX 512 capable cores. This is incorrect; all Skylake-X chips can perform AVX-512 operations within the bounds of whatever arcane subsets of the extension they support. The difference is that Core i7s (and some Xeon SPs) can only perform AVX-512 operations by fusing their two AVX2 units; Core i9s (and other Xeon SPs) can both fuse their AVX2 units and execute AVX-512 instructions on a dedicated functional unit all at once. In theory, you get double the AVX-512 throughput on Core i9s as i7s but all SKL-X CPUs are capable of executing AVX-512. e;fb
|
# ? Oct 1, 2017 23:38 |
|
Paul MaudDib posted:Everyone pooh-poohs AVX512 as some niche thing but a surprising amount of the actual compute-intensive stuff (i.e the stuff you're buying a 10-18 core CPU for) actually can use it. x264, x265, and Blender all see huge IPC gains from AVX512 - to the extent that SKL-X is pushing 50-100% more performance-per-core in these tasks. Reviewers need to stop using loving Handbrake. The x264 devs added AVX512 paths back in June, long before reviewers got their Skylake-X samples, but because Handbrake is still pinned to an x264 commit from November of last year SKL-X reviews were largely using the old AVX2 paths.
|
# ? Oct 1, 2017 23:41 |
|
TheJeffers posted:This is incorrect; all Skylake-X chips can perform AVX-512 operations within the bounds of whatever arcane subsets of the extension they support. The difference is that Core i7s (and some Xeon SPs) can only perform AVX-512 operations by fusing their two AVX2 units; Core i9s (and other Xeon SPs) can both fuse their AVX2 units and execute AVX-512 instructions on a dedicated functional unit all at once. In theory, you get double the AVX-512 throughput on Core i9s as i7s but all SKL-X CPUs are capable of executing AVX-512. Oh, so definitely gimped then. On the other hand that's the same tradeoff that Ryzen makes, so you're not going to gain anything by switching to TR there. A 7820X is still using substantially less energy in the Blender render than an 1800X. Intel still has a small technological lead but they're just so intent on segmenting their market as far as humanly possible that eventually they're going to squander it if they keep it up. Skylake-X makes little enough sense as it is (for anything except pure compute tasks) that they don't really have any room for that anymore. I don't really see any good reason to buy a SKL-X i7 over a low-end SKL-X Xeon (if you're going for memory channels/PCIe lanes) or Coffee Lake (for gaming performance). Or, y'know, Ryzen/Threadripper. Paul MaudDib fucked around with this message at 23:51 on Oct 1, 2017 |
# ? Oct 1, 2017 23:43 |
|
I wonder how long it'll take Intel to support ECC on their consumer/prosumer platforms. SL-X would be a good place to start but then they can't charge double for the otherwise identical Xeon parts.
|
# ? Oct 1, 2017 23:51 |
|
Paul MaudDib posted:Oh, so definitely gimped then. On the other hand that's the same tradeoff that Ryzen makes, so you're not going to gain anything by switching to TR there. Lesser Skylake-X is only "gimped" relative to greater Skylake-X. Ryzen CPUs can't perform AVX-512 operations at all so you still have twice the theoretical AVX throughput of Zen (1x fused 256-bit op per however many cycles) from AVX/AVX2/AVX-512 on Skylake-X even on Core i7s. I believe AVX-512 is not just about throughput but also being nicer to code for for certain things, so there's still value in even a lesser implementation of it for some, I imagine.
|
# ? Oct 1, 2017 23:51 |
|
TheJeffers posted:Lesser Skylake-X is only "gimped" relative to greater Skylake-X. Ryzen CPUs can't perform AVX-512 operations at all so you still have twice the theoretical AVX throughput of Zen (1x fused 256-bit op per however many cycles) from AVX/AVX2/AVX-512 on Skylake-X even on Core i7s. I believe AVX-512 is not just about throughput but also being nicer to code for for certain things, so there's still value in even a lesser implementation of it for some, I imagine. I keep hearing about AVX-512. Can someone give me an example app that takes advantage of this (other than goddamn Blender)
|
# ? Oct 2, 2017 00:23 |
|
redeyes posted:I keep hearing about AVX-512. Can someone give me an example app that takes advantage of this (other than goddamn Blender) There's not much outside of HPC so far. The two examples that come to mind are x264 and the Embree raytracing library, which powers UE4s lightmapper for example.
|
# ? Oct 2, 2017 00:40 |
|
|
# ? May 28, 2024 05:42 |
|
TheJeffers posted:Lesser Skylake-X is only "gimped" relative to greater Skylake-X. Ryzen CPUs can't perform AVX-512 operations at all so you still have twice the theoretical AVX throughput of Zen (1x fused 256-bit op per however many cycles) from AVX/AVX2/AVX-512 on Skylake-X even on Core i7s. I believe AVX-512 is not just about throughput but also being nicer to code for for certain things, so there's still value in even a lesser implementation of it for some, I imagine. It isn't gimped at all. Every skylake x cpu has the same full throughput AVX512 (64 single-precision FLOPs per cycle per core). Exactly double that of Haswell/Broadwell/consumer skylake. Tons of people have tested them on various boards. It's kind of concerning that every reviewer who commented about this got it wrong because I can't actually find any information from Intel to suggest that the i7 SKUs would be any different (how would they be anyway? They're all based on the same die) https://communities.intel.com/thread/116484
|
# ? Oct 2, 2017 06:15 |