|
eames posted:Yeah, stock clocks and PC-2400/PC-2666 RAM with pretty loose CL17 timings all around. I am curious what the benchmarks would look like with PC-3200 CL14. I'd expect the ryzen parts to make up some ground in that case, no? what with infinity fabric and all.
|
# ? Jul 10, 2017 19:27 |
|
|
# ? Apr 27, 2024 18:17 |
|
Toms Hardware also looked into the whole VRM situation with a 7900X. They used a custom watercooling block + water chiller for constant 20°C coolant and the core temperature still went up to 94°C at 300W before the VRMs throttled. Actual power consumption at stock frequencies is ~230W with "enhanced turbo" and AVX workloads. http://www.tomshardware.com/reviews/-intel-skylake-x-overclocking-thermal-issues,5117.html
|
# ? Jul 10, 2017 20:06 |
|
eames posted:Actual power consumption at stock frequencies is ~230W with "enhanced turbo" and AVX workloads. "Enhanced turbo" here means 4.3 GHz all-core and also an increase in voltage. Yeah, that does push your power consumption up versus the specced 4.0 GHz.
|
# ? Jul 10, 2017 20:18 |
|
At $1000 a chip Intel should include a water cooling setup with the radiator sitting in a small fridge you'd put next to your case. Call it a Gaming Fridge - Gamer Cool - with a Gaming paint scheme and RGB, it would also store your Gamer Fuel and Gamer Snacks so you don't have to get up whilst sitting in your Gaming Chair at your Gaming Desk, gaming.
|
# ? Jul 10, 2017 20:54 |
|
3peat posted:At $1000 a chip Intel should include a water cooling setup with the radiator sitting in a small fridge you'd put next to your case. Call it a Gaming Fridge - Gamer Cool - with a Gaming paint scheme and RGB, it would also store your Gamer Fuel and Gamer Snacks so you don't have to get up whilst sitting in your Gaming Chair at your Gaming Desk, gaming. Most minifridges have a rating of probably 10W. Your big house fridge has a rating of like 100W. They work by insulation to keep the cold in, they have very little actual cooling capacity. Also, probably aren't rated for anywhere near 24/7 operation. You can put an AIO radiator in an ice-water bath or even use a dedicated phase-change cooling system designed for PCs, but then you have to deal with condensation. Tom's Hardware was actually using a phase-change system for their overclocking results (Alphacool Eiszeit 2000). Frankly why bother? The problem is that heat/cold just aren't moving from the chip to the IHS, or from the IHS to the chip. The IHS is already under 30C in their example, while the core is hitting Tjunction, getting the IHS a little cooler isn't going to change anything because it's not making good thermal contact with the die. I realize you're joking but seriously, just accept that if you have any interest in SKL-X you need to delid. The thermal performance is just unacceptable without it, even going nuclear with a phase-change system. I think there's a pretty good argument for it with delidding though, 6-10C of Skylake at 5 GHz is going to be fantastic for gaming, which (like it or not) is a massive chunk of the market for these enthusiast chips. If all you're doing is CAD or video encoding then get a Xeon/Ryzen/Threadripper. Paul MaudDib fucked around with this message at 22:12 on Jul 10, 2017 |
# ? Jul 10, 2017 21:01 |
|
Paul MaudDib posted:Tom's Hardware was actually using a phase-change system for their overclocking results (Alphacool Eiszeit 2000). Frankly why bother? I think the point was exactly to highlight the rest of your conclusion: that it simply doesn't matter what sort of cooling solution you slap on Skylake-X, it will never be sufficient because it's not actually the problem. Why they didn't also throw in some test runs with a delided chip slapped on under their giggle-cooler remains a mystery.
|
# ? Jul 10, 2017 22:05 |
|
I do like that Intel probably forced derBauer to walk his initial comments back to say something to the effect of "well, this was MY experience in testing several different boards objectively, your experience might differ." Sounds like he got told "walk it back a bit or you'll never get another free evaluation chip from us ever again."
|
# ? Jul 10, 2017 22:09 |
|
If that's the case and he had some balls, he'd tell Intel to go gently caress themselves and air the dirty laundry. I doubt he'll still be relevant when the next bigger CPU release will be out (that alluded new architecture in a few years), so he might just go all out. Maybe I just missed out on things, but I can't remember the last few releases to have been treated as such big deals as Skylake-X has been made out to be.
|
# ? Jul 10, 2017 23:36 |
|
guys, i know this is crazy but... i'm beginning to suspect that this whole skylake-x release is a bit rocky!
|
# ? Jul 10, 2017 23:38 |
|
Combat Pretzel posted:Maybe I just missed out on things, but I can't remember the last few releases to have been treated as such big deals as Skylake-X has been made out to be. That's because the last few releases happened in a vacuum. Things changed when Zen — against all odds — turned out to be not completely a terrible architecture.
|
# ? Jul 10, 2017 23:43 |
|
Skylake-SP/Cascade Lake (Xeon Platinum/Gold/Silver/Bronze) is formally gonna launch tomorrow, with XCC up to 28 cores and 8 sockets on Intel's new mesh interconnect. Videocardz posted the slides before the NDA: https://videocardz.com/70874/intels-epyc-response-xeon-scalable-processor-skylake-sp
|
# ? Jul 11, 2017 05:23 |
|
"Architected" is only a valid word if your name is Ivanka Trump.
|
# ? Jul 11, 2017 05:27 |
|
I don't get why they need to do the naming change. It seems confusing. The skylake/kabylake E3s are unfucked with and are e3-1225 v6 etc.
|
# ? Jul 11, 2017 05:28 |
|
BIG HEADLINE posted:"Architected" is only a valid word if your name is Ivanka Trump.
|
# ? Jul 11, 2017 05:33 |
|
It's entirely possible to have grammatically valid words which are nonetheless virtually never used in common English because there are better words to fill the space. In this case, "design" works a good bit better and doesn't make people wonder whether you're semi-illiterate, or just in love with highly questionable conjugations.
|
# ? Jul 11, 2017 13:31 |
|
Maybe it's just me, but 'architect' has a connotation of finality that 'design' doesn't have. Something that was architected has more involved planning because it's going to be very difficult to change later, like moving load bearing walls versus changing the page layout in a magazine. I also never questioned the literacy of the speaker the first time I encountered the phrase.
|
# ? Jul 11, 2017 17:06 |
|
The amusing bit is that actual architects and such do not use the term. It's almost exclusively used in IT/computers for whatever reason.
|
# ? Jul 11, 2017 17:49 |
|
Well, put two pots of coffee on. AnandTech just posted a 23 page deep dive on the new Xeons vs Epyc. Thankfully, the pages are pretty short by AT standards. http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
|
# ? Jul 11, 2017 18:05 |
|
My friend is having a field day with this.friend posted:• Intel introduces AVX-512 friend again posted:Imagine Intel makes an Iris Pro CPU with Skylake-SP
|
# ? Jul 11, 2017 18:49 |
|
Cygni posted:Well, put two pots of coffee on. AnandTech just posted a 23 page deep dive on the new Xeons vs Epyc. Thankfully, the pages are pretty short by AT standards. My assertion has been that TR is probably going to be basically a NUMA processor and probably won't scale well with games. Are there any relevant benchmarks here to point to on that, like applications that tend to be analogous to gaming performance? The whole "dual-socket Epyc is basically a virtual octal socket" closer does pretty much confirm that as far as I'm concerned. Paul MaudDib fucked around with this message at 19:09 on Jul 11, 2017 |
# ? Jul 11, 2017 19:05 |
|
DrDork posted:It's entirely possible to have grammatically valid words which are nonetheless virtually never used in common English because there are better words to fill the space. In this case, "design" works a good bit better and doesn't make people wonder whether you're semi-illiterate, or just in love with highly questionable conjugations. Architecture and design can be two different types of roles within a hardware company. https://en.wikipedia.org/wiki/Hardware_architect WhyteRyce fucked around with this message at 19:13 on Jul 11, 2017 |
# ? Jul 11, 2017 19:10 |
|
DrDork posted:It's entirely possible to have grammatically valid words which are nonetheless virtually never used in common English because there are better words to fill the space. In this case, "design" works a good bit better and doesn't make people wonder whether you're semi-illiterate, or just in love with highly questionable conjugations. quote:The term “architecture” in computer literature can be traced to the work of Lyle R. Johnson, Frederick P. Brooks, Jr., and Mohammad Usman Khan. All of which were members of the Machine Organization department in IBM’s main research center in 1959. As forums poster WhyteRyce mentioned, 'design' in this context would be specific to logic design. Do we talk about ISD's and microdesign features?
|
# ? Jul 11, 2017 22:37 |
|
WAR DOGS OF SOCHI posted:guys, i know this is crazy but...
|
# ? Jul 11, 2017 23:03 |
|
Omnipath connector? The Purley has a secondary external bus? What the gently caress?
|
# ? Jul 11, 2017 23:09 |
|
Combat Pretzel posted:Omnipath connector? The Purley has a secondary external bus? What the gently caress? Optional, for HPC shops.
|
# ? Jul 11, 2017 23:13 |
|
Also, their new C620 supports RDMA over 10G? I mean what the gently caress? You can't find a recent PCIe adapter of theirs that does it, plenty of sources on the web suggesting they've dropped it, and now they're back on it with their drat server chipset? Good grief, Intel!
|
# ? Jul 11, 2017 23:20 |
|
whats going on here?
|
# ? Jul 11, 2017 23:24 |
|
3peat posted:whats going on here?
|
# ? Jul 11, 2017 23:26 |
|
Yeah it looks like they forgot to mark the watt symbol on that column
|
# ? Jul 11, 2017 23:28 |
|
Anime Schoolgirl posted:EPYC is much more efficient at POV-Ray per watt I meant why is the xeon drawing almost 50% more power for lower floating point performance? Is this an architecture issue, lack of optimization or bug? It's the one thing in that review I found weird, otherwise the chips are fairly close and trading wins even if the xeon is twice the price
|
# ? Jul 11, 2017 23:44 |
|
Anime Schoolgirl posted:I think Skylake-X might not be up to par woah, let's not go too far here!
|
# ? Jul 12, 2017 03:12 |
|
3peat posted:I meant why is the xeon drawing almost 50% more power for lower floating point performance? Is this an architecture issue, lack of optimization or bug? It's the one thing in that review I found weird, otherwise the chips are fairly close and trading wins even if the xeon is twice the price Anandtech's notes heavily implied that the really-wide AVX512 paths were sucking power down even when they weren't being used somehow. quote:The only conclusion that we can draw so far, is that the EPYC 7601 is likely to draw more power when running integer applications, while the rather wide FP units of the Intel CPUs are real power hogs even if they do not run heavy AVX applications
|
# ? Jul 12, 2017 03:20 |
|
WAR DOGS OF SOCHI posted:woah, let's not go too far here! Intel CPU and Platfrom Discussion: Skylake-X Might Be...Bad? *knocks on custom waterloop*
|
# ? Jul 12, 2017 10:24 |
|
Given the choice between 4 glued-together Desktop Dies and glue between the Die and the Heatspreader, I'd choose the former. https://www.techpowerup.com/235092/intel-says-amd-epyc-processors-glued-together-in-official-slide-deck
|
# ? Jul 12, 2017 13:06 |
|
They missed the fourth checkmark of "better able to bribe/influence CFOs & CTOs."
|
# ? Jul 12, 2017 13:25 |
|
Intel: Handbags at dawn.
|
# ? Jul 12, 2017 13:28 |
|
So what exactly is the difference between a server and desktop product these days anyway? Especially when the purported desktop product supports all relevant server things like ECC, high bus bandwidth and virtualization? That slide is pretty much a pants-making GBS threads admission.
|
# ? Jul 12, 2017 14:08 |
|
Combat Pretzel posted:Also, their new C620 supports RDMA over 10G? I mean what the gently caress? You can't find a recent PCIe adapter of theirs that does it, plenty of sources on the web suggesting they've dropped it, and now they're back on it with their drat server chipset? Good grief, Intel! It got introduced as a feature of SMB3, I bet Microsoft is driving that more than anything.
|
# ? Jul 12, 2017 14:09 |
|
eames posted:Given the choice between 4 glued-together Desktop Dies and glue between the Die and the Heatspreader, I'd choose the former. Yeah Intel only glued together TWO of their dies.
|
# ? Jul 12, 2017 14:11 |
|
|
# ? Apr 27, 2024 18:17 |
|
Combat Pretzel posted:So what exactly is the difference between a server and desktop product these days anyway? Especially when the purported desktop product supports all relevant server things like ECC, high bus bandwidth and virtualization? Well, you see, for Intel their desktop chips don't support ECC--that's what makes a server chip a server chip, after all!
|
# ? Jul 12, 2017 14:27 |