|
TomR posted:Wasn't the real problem the thickness of the glue they used causing the heat spreader to be lifted from the die a tiny amount, and that the actual TIM was fine? Yes. Getting the cheapest Corning paste that comes in a 4 ounce tube and repasting with that but with the glue cut away still showed 10-15 degree drops. Reaching the peak of TIM will get you a couple more degrees, but in general the performance difference between the cheapest TIM and most expensive (short of those gallium based ones that have potential for detrimental chemical reactions, or really thick applications like in Intel's case) isn't worth changing whatever weird bottle of paste you might have laying around.
|
# ? Jun 5, 2017 23:56 |
|
|
# ? May 10, 2024 05:46 |
|
Has anyone ever damaged a core, by repasting the TIM (successfully), and then refitting it with the heatspreader? I'm just trying to figure out if the glue used is some kind of buffer to super high mounting force sinks, or if it's just rear end.
|
# ? Jun 6, 2017 00:09 |
|
GRINDCORE MEGGIDO posted:Has anyone ever damaged a core, by repasting the TIM (successfully), and then refitting it with the heatspreader? No
|
# ? Jun 6, 2017 00:46 |
|
craig588 posted:Yes. Getting the cheapest Corning paste that comes in a 4 ounce tube and repasting with that but with the glue cut away still showed 10-15 degree drops.
|
# ? Jun 6, 2017 03:56 |
|
Linus is pretty much calling a hard pass on Sky-X (and a 'what the gently caress were they thinking' on Kaby-X): https://www.youtube.com/watch?v=TWFzWRoVNnE
|
# ? Jun 6, 2017 04:35 |
|
Man, you have to have really hosed up your marketing *and* your technology if even Linus "lol i dropped my 8-drive RAID 5 of 3 TB SSDs aren't i the randomest PC nerd xD" Sebastian is going to pass on your HEDT platform.
|
# ? Jun 6, 2017 04:44 |
|
So much for "everything's going so well we're gonna release Coffee Lake early!" http://wccftech.com/intel-coffee-lake-delayed-2018-8th-gen-kaby-lake-refresh/amp/
|
# ? Jun 6, 2017 06:13 |
|
Now Jay has a video making GBS threads on it too lol https://www.youtube.com/watch?v=QJNRtGo5IMc
|
# ? Jun 6, 2017 06:25 |
|
Still wondering when Purley platforms are gonna show up. Need me some massive 3647 pin cpus.
|
# ? Jun 6, 2017 06:31 |
|
TomR posted:Wasn't the real problem the thickness of the glue they used causing the heat spreader to be lifted from the die a tiny amount, and that the actual TIM was fine? Yeah, but that's not as catchy in YouTube comments as: "Intel *STILL* using lovely TIM, lols"
|
# ? Jun 6, 2017 06:38 |
|
I corrected a bunch when kaby came out on FB, but they're jeering idiots.
|
# ? Jun 6, 2017 07:15 |
|
I'm going to do a Linux desktop with a Windows gaming VM for my next desktop so I've been paying attention to all the HEDT stuff and Intel seems to be doing its damnedest to make Skylake-X as uncompelling as possible. I don't particularly relish dropping $1000 on a CPU just for PCIe lanes when I'd rather have a higher clocked 8-core, and cutting off 4 lanes is just spiteful when partners are announcing x299 boards with 4 and 5 M.2 slots. It's me, I'm the guy wanting to run RAID NVMe SSDs (ZFS) on my desktop, but for resiliency not (just) performance. I'm shocked none of the x399 boards announced so far have four M.2 slots though, seems like an easy win for AMD. I'd love to RAID10 smaller SSDs directly from the CPU for maximum at roughly the same price. I'm waiting for Volta so at least the skylake-x vs threadripper slap fight will be relatively settled by the time I build.
|
# ? Jun 6, 2017 08:57 |
|
Well, there is a bright side to all of this. Intel's Retail Edge program is going to have a TON of surplus stock to dangle. They'll probably sell the 18C/36T for $999 to "Rock Stars" just to clear their unsold inventory in Spring 2018. But yeah, GDDR6 isn't due until early 2018, which means we probably won't see Volta until late Spring 2018 or even early Autumn.
|
# ? Jun 6, 2017 09:11 |
|
BIG HEADLINE posted:So much for "everything's going so well we're gonna release Coffee Lake early!" Early 2018 for the consumer hexacore and 170/270 boards are not compatible, ouchies They might be going straight up against a 200$ second generation Zen if the stars align.
|
# ? Jun 6, 2017 10:04 |
|
I wonder if the August rumor was false all along or if they did this because Coffee Lake would have eaten into SL-X sales.
|
# ? Jun 6, 2017 10:07 |
|
eames posted:I wonder if the August rumor was false all along or if they did this because Coffee Lake would have eaten into SL-X sales. Well, now there's a rumor that Intel will sit on the 6C/12T i7s until later, again, being spread by WCCF, so more than a little salt is required.
|
# ? Jun 6, 2017 11:15 |
|
BIG HEADLINE posted:Well, there is a bright side to all of this. Intel's Retail Edge program is going to have a TON of surplus stock to raid dongle.
|
# ? Jun 6, 2017 11:29 |
|
Desuwa posted:and cutting off 4 lanes is just spiteful
|
# ? Jun 6, 2017 12:41 |
|
This is some serious bullshit, and it does feel like stringing people like me along. Maybe I’ll just go AMD, what a mess.
|
# ? Jun 6, 2017 12:42 |
|
Does Intel not have a backup plan for Zen performing the way it does at its respective pricepoints? 6C Coffee Lake delayed, Kaby Lake-X exists (?!), 14-18 core HEDT wasn't on the roadmap shown jus tdays before the launch, the Asus rep stealth-editing his posting from "we won't see 14-18 core parts until next year" to "later this year", etc.
|
# ? Jun 6, 2017 13:11 |
|
ufarn posted:This is some serious bullshit, and it does feel like stringing people like me along. What're you after? What's your budget? If the answer is "heavily threaded loads" and "not a ton of money", take a peek at the fact that you can get an 8C/16T 1700X with motherboard for $340 total at Microcenter. These threads need to get merged worse than Intel needs >4 cores on a mainstream platform. Edit: At the very least as a stopgap i5s need to pick up hyperthreading ASAP. Twerk from Home fucked around with this message at 15:08 on Jun 6, 2017 |
# ? Jun 6, 2017 15:01 |
|
eames posted:Does Intel not have a backup plan for Zen performing the way it does at its respective pricepoints? After the last three times they might have just gone "ehhh let's just assume AMD will continue to suck and not waste any man hours developing a plan for the increasingly unlikely situation that they stop sucking before going bust".
|
# ? Jun 6, 2017 15:47 |
|
Twerk from Home posted:What're you after? What's your budget? If the answer is "heavily threaded loads" and "not a ton of money", take a peek at the fact that you can get an 8C/16T 1700X with motherboard for $340 total at Microcenter. These threads need to get merged worse than Intel needs >4 cores on a mainstream platform.
|
# ? Jun 6, 2017 15:57 |
|
What do you mean Coffee Lake delayed? Wasn't it always 2018?
|
# ? Jun 6, 2017 15:58 |
|
Combat Pretzel posted:What do you mean Coffee Lake delayed? Wasn't it always 2018? There were rumors it was being pushed forwards to 3rd/4th quarter 2017 because "oh poo poo Zen"
|
# ? Jun 6, 2017 15:59 |
|
Apparently the 10-core Skylake-X overclocked to 4.3GHz doesn't have any significant gain over a 10-core Broadwell-E at the same clock. At least not in Cinebench (2364 vs almost 2300). https://www.hardocp.com/news/2017/06/05/intel_core_i9_skylake_e_clocks_no_higher_than_broadwell/ I guess raytracing isn't that cache sensitive for the new layout to prove itself.
|
# ? Jun 6, 2017 16:02 |
Combat Pretzel posted:Apparently the 10-core Skylake-X overclocked to 4.3GHz doesn't have any significant gain over a 10-core Broadwell-E at the same clock. At least not in Cinebench (2364 vs almost 2300). That is about what I would expect, 2%-3% is what we have been seeing for a while as far as IPC gains on Intel's CPUs.
|
|
# ? Jun 6, 2017 16:57 |
|
Skylake-X should be more impressive under AVX-512 workloads, but lol who gives a poo poo. Intel have been shipping full-rate AVX since Haswell and it's still barely utilized outside of HPC, it's going to be at least another 5 years from now before AVX-512 starts seeing non-HPC adoption.
|
# ? Jun 6, 2017 17:56 |
|
I was hoping for 5% but I guess I'll take 3%. If Intel can get it to OC close to 5 GHz (17% improvement over Broadwell-E's 4.3 GHz) then that would put it ~19% ahead of BW-E in total, which puts it 24-29% ahead of Ryzen in single-thread performance. Which is really close to the magic "6-core Intel matches 8-core AMD" target that would mean price-to-performance parity between the two. (Side note but lately I've been feeling that Cinebench is an increasingly irrelevant benchmark. It seems to bear no resemblance to real x264/x265/Premiere/Blender anymore - the 1800X blows out the 6900K in the synthetic Cinebench but the 6900K wins all the real-world benchmarks. Maybe it's OK for comparing within an architectural family (eg Broadwell vs Skylake) but I don't think it's the least bit useful anymore for comparing (say) Ryzen and Broadwell.) Paul MaudDib fucked around with this message at 18:16 on Jun 6, 2017 |
# ? Jun 6, 2017 18:03 |
|
Paul MaudDib posted:(Side note but lately I've been feeling that Cinebench is an increasingly irrelevant benchmark. It seems to bear no resemblance to real x264, x265, or Premiere anymore - the 1800X blows out the 6900K in the synthetic Cinebench but the 6900K wins all the real-world benchmarks. Maybe it's OK for comparing within an architectural family (eg Broadwell vs Skylake) but I don't think it's the least bit useful anymore for comparing (say) Ryzen and Broadwell.) Technically Cinebench isn't a synthetic benchmark, it's just a standalone/crippled version of Cinema 4Ds internal renderer. Although I'll grant you that (1) it usually lags several versions behind C4D proper and (2) from my experience it seems like nobody actually uses C4D internal anymore. Nearly all of the C4D artists I encounter have moved on to third party GPU renderers like Redshift and Octane.
|
# ? Jun 6, 2017 18:10 |
|
Paul MaudDib posted:
Uhh, Cinebench is offline 3D renderer from Cinema 4D that uses either ray or path-tracing, not sure which. Video rendering/encoding never had any resemblance to 3D rendering.
|
# ? Jun 6, 2017 18:16 |
|
SinineSiil posted:Uhh, Cinebench is offline 3D renderer from Cinema 4D that uses either ray or path-tracing, not sure which. Video rendering/encoding never had any resemblance to 3D rendering. It doesn't match up to Blender's performance either. The 6900K is ahead by 14% and 10% in the two standard benchmarks, while the 1800X is ahead by 9% in Cinebench R15 MT. So that means Cinebench is skewed ~20-25% from real-world benchmarks.
|
# ? Jun 6, 2017 18:21 |
|
Paul MaudDib posted:It doesn't match up to Blender's performance either. The 6900K is ahead by 14% and 10% in the two standard benchmarks, while the 1800X is ahead by 9% in Cinebench R15 MT. I misunderstood what you were saying there.
|
# ? Jun 6, 2017 18:24 |
|
SinineSiil posted:I misunderstood what you were saying there. I thought Cinebench did include some video encoding too but I guess not. (but I would expect it to correlate fairly well to other multithread-friendly media workloads like video encoding anyway) But like I said, it doesn't match other rendering benchmarks either, so it doesn't seem to correlate well to anything anymore.
|
# ? Jun 6, 2017 18:30 |
|
Paul MaudDib posted:I was hoping for 5% but I guess I'll take 3%. If Intel can get it to OC close to 5 GHz (17% improvement over Broadwell-E's 4.3 GHz) then that would put it ~19% ahead of BW-E in total, which puts it 24-29% ahead of Ryzen in single-thread performance. Which is really close to the magic "6-core Intel matches 8-core AMD" target that would mean price-to-performance parity between the two. e: Also I don't know why they even bother with TechArp's x264 benchmark when they're already testing using HandBrake (where Ryzen wins) Malloc Voidstar fucked around with this message at 19:04 on Jun 6, 2017 |
# ? Jun 6, 2017 18:44 |
|
I think people claiming Ryzen has similar IPC to Intel are looking at cherry-picked benchmarks, or things that don't fully scale with CPU speed (ie games). Looking at http://www.anandtech.com/bench/product/1853?vs=1729 It seems like the 6900k and 1800x are basically tied at stock clocks. Except the former is clocked 400mhz lower, which means Broadwell-E has about a 15% IPC advantage. Looking at the same benchmarks using the 1700x indicates that these scores roughly scale with frequency. Ryzen will reliably OC to 4ghz (+11%) versus broadwell E at 4.3ghz (+34%). So at max OC Intel is about 20% faster. If skylake-x gets a 5% avg IPC gain over Broadwell-E, and clocks at 4.5ghz it will be 35% faster than Ryzen. That makes the Skylake 6 core ($390) competitive against the 1700 non-X ($330). Especially when you consider that if total speed is equal then lower core count CPUs are strictly superior for single threaded applications. X299 boards cost more than Ryzen ones, but the former has HEDT features; Ryzen is a mainstream platform. I don't really get the rips on the 7740k and the NVMe raid stuff. They're lame but...they are extras; if they didn't exist at all what complaint would you have? Just ignore them. Like, Ryzen is cool, I think stitching dies together is cool and novel, and I'm super glad AMD has forced Intel to lower prices, but I don't see a reason to declare Intel dead because their competitor finally released a product that would have been state of the art in the year 2012.
|
# ? Jun 6, 2017 20:07 |
|
Eyes Only posted:If skylake-x gets a 5% avg IPC gain over Broadwell-E, and clocks at 4.5ghz it will be 35% faster than Ryzen. That makes the Skylake 6 core ($390) competitive against the 1700 non-X ($330). Especially when you consider that if total speed is equal then lower core count CPUs are strictly superior for single threaded applications. X299 boards cost more than Ryzen ones, but the former has HEDT features; Ryzen is a mainstream platform. This is my basic argument as well. I think your numbers may be a little optimistic here, I'm still expecting the 6C SKL-X to land slightly behind the 8C R7s (~5%), but Intel's goal here is to put a 6C up against the 8C with roughly equivalent multi-threaded performance (i.e. a 25%+ lead in single-threaded performance). My rule of thumb (guesstimate) is that the HEDT chips are ~100 MHz slower than the equivalent "Small" version of the architecture. For example, Small Haswell reliably goes to 4.7 GHz with extra voltage, Haswell-E reliably goes to 4.6 GHz (with extra voltage). Small Broadwell goes to 4.4 GHz, Broadwell-E goes to 4.3 GHz. Since a 6700K reliably goes to 4.7, I really think Skylake-X is going to land at no less than 4.6 GHz max overclock. Less than 4.5 would be abysmal. The other train of thought is that Kaby Lake is basically just Skylake on a revised process (plus iGPU stuff that won't be in the -E chips), so if Intel runs Skylake-X on the new Kaby Lake process then we could be looking at closer to 5 GHz right out of the gate. This is supported by the 4.5 GHz single-core turbo speeds on some of the 10C SKUs. Even cherrypicking the best core on the chip (Turbo Boost 3.0), if they can hit 4.5 GHz at stock-ish voltage then 5 GHz should be doable with extra voltage. (myself, if I could get 4.5 GHz all-core at near stock voltages, that would be plenty for me) quote:I don't really get the rips on the 7740k and the NVMe raid stuff. They're lame but...they are extras; if they didn't exist at all what complaint would you have? Just ignore them. Yeah, I mean, the 4C4T and 4C8T are dumb SKUs that nobody is going to buy. Who cares? The Intel RAID thing only really matters if you want a bootable RAID array, or soft-RAID is just too slow for you. Anyone in this use-case is already dropping tens of thousands of dollars on NVMe drives anyway, so another $100 for hardware RAID support is no big deal. Someone else suggested that this is really going after Dell's side business in hardware RAID controllers more than anything. The PCIe lanes thing is actually kinda dumb, I was hoping that Intel would put 44 lanes on all the HEDT chips (except the 4C models) just as a differentiator from Ryzen, but it's not the end of the world either. The TIM is definitely the dickpunch in this launch.
|
# ? Jun 6, 2017 20:54 |
|
gonna lol if they also TIM the xeons $4000 processors using goop with poor application tolerances for heat transfer
|
# ? Jun 6, 2017 21:09 |
|
Paul MaudDib posted:
I agree with all of this. But I do wonder about the TIM thing. It's been established that the problem is not the TIM specifically, but more the spacing because of the glue, right? Is there any chance that that has been corrected or will be better on these chips than it has been in the past?
|
# ? Jun 6, 2017 21:29 |
|
|
# ? May 10, 2024 05:46 |
|
Lowen SoDium posted:Is there any chance that that has been corrected or will be better on these chips than it has been in the past? Maybe, but they could have improved it at any point after Ivy, and they didn't.
|
# ? Jun 6, 2017 21:36 |