|
Thats... a lot for a home NAS. Most home gamers will be well served with like a cheap 2600k in a box with a couple HDs and use Storage Spaces tbh.
|
#
?
Feb 24, 2022 23:53
|
|
|
|
| # ? May 8, 2024 11:41 |
|
|
It's definitely overkill but that's no reason not to covet it for a badass NAS/home server 
|
|
#
?
Feb 25, 2022 00:03
|
|
|
priznat posted:I just realized from that table that they don't have any efficiency cores, which is kind of a shame for something that would be great in a NAS. HMP isn't nearly as smart as a lot of people think it is, because fundamentally you're relying on the CPU scheduler to make assumptions about whether a quantum of work (a generic term meaning the minimum unit that the scheduler divides processing tasks into) will spend more energy finishing faster on a performance core, or whether the efficiency cores can finish the quantum of work quickly enough that it'll have been cheaper. The scheduler has to guess this without being given any information about what the program thinks is most efficient, because while the compiler might be able to figure something like that perhaps based on input by the programmer, no codebase or toolchain has been modified to make these kinds of decisions, and I'm not aware of any serious efforts to that end. This is further complicated by scientific testing that reveal that you can spend years of work-hours writing tens of thousands of lines of scheduler code, with very fine-tuned heuristics, and still get scores that are indistinguishable or in some cases slightly (and in one case significantly) worse than a scheduler that's written in only a few thousand lines of code. Even if HMP is a thing that's existed in the form of ARM bigLITTLE for years, the technology is very much still in its infancy, and I'm not convinced it won't be found to be a flop before all of the issues have been figured out - assuming they do, which itself is an open question. Cygni posted:Thats... a lot for a home NAS. Most home gamers will be well served with like a cheap 2600k in a box with a couple HDs and use Storage Spaces tbh. priznat posted:It's definitely overkill but that's no reason not to covet it for a badass NAS/home server 
BlankSystemDaemon fucked around with this message at 00:41 on Feb 25, 2022 |
|
#
?
Feb 25, 2022 00:30
|
|
|
BlankSystemDaemon posted:They're Ice Lake, not Alder Lake. Xeons are always quite a bit behind, unless we're talking about the very high-end Xeons that the hyperscalers buy every manufactured unit of.  right, forgot those just came along post Ice Lake. Those 4 core ones at 45W TDP would be excellent for a quiet home server regardless.
|
|
#
?
Feb 25, 2022 00:34
|
|
|
BlankSystemDaemon posted:If you've got lots of what I'm sure are home movies you've definitely recorded yourself, and want to archive them with the best possible image quality, the only solution is to use h264 or h265 with CRF in slow-mode, which is the only way to makes use of all of the threads while doing video encoding. The vast majority of people are downloading media, not ripping it themselves any more.
|
|
#
?
Feb 25, 2022 05:07
|
|
|
Cygni posted:Thats... a lot for a home NAS. Most home gamers will be well served with like a cheap 2600k in a box with a couple HDs and use Storage Spaces tbh. Don't the newer CPUs and chipsets idle a lot more efficiently than Sandy? Or no?
|
|
#
?
Feb 25, 2022 06:11
|
|
|
K8.0 posted:Don't the newer CPUs and chipsets idle a lot more efficiently than Sandy? Or no? It kinda depends on the use case. Sandy will go down to single digit package power at idle. And for most home use cases, the CPU isn't going to be hit very hard most of the time anyway. The fancy Synology 4 bay home NAS use 2 core/2 thread, 4 year old Atom CPUs for example. For my personal data point, I used a 2c/2t 35w Sandy Celeron for a number of years, but eventually dropped in a 3770S due to them getting cheap. It spent a majority of its life at sub 10% usage, running as a seed box with VPN software, Plex server, and 4 spinning rust in Storage Spaces. The HDs used more power than the CPU most of the time. Honestly the only time I saw high CPU usage were weird instances where the VPN software was having trouble with all the torrent traffic for some odd reason. Only happened a handful of times though. I recently got a lot deal on ebay with a 4790S + ITX board + 16gb of ram for $160, ostensibly to upgrade the NAS... but really just cause i wanted to play with Haswell. I just VNCed in and its running at 2% CPU usage at 1.2ghz and 12w package power, ha. So yeah, if you are doing heavy lift stuff or constant CPU decoding/encoding on the NAS, more powerful parts might make more sense. But for most home gamers, you can probably go cheap and used and be totally sufficient. And as an aside, in general i highly recommend rolling NAS/HTPC/seed box/home security/home server with cheapo hardware. Really opens a lot of doors in nerdom. Cygni fucked around with this message at 07:53 on Feb 25, 2022 |
|
#
?
Feb 25, 2022 07:50
|
|
|
package power doesn't include the chipset, which has also come down quite a bit. remember the push for Intel to get chipsets onto 14nm to satisfy a Cali requirement, during their shortages of 2017-2018? new systems definitely do idle quite a bit better than older ones. There's a lot more gated off and idled down and better nodes in general. I bet the average beige box probably idled at 30-35w system power and now it's maybe 15-20W. I had a Conroe 2c optiplex that I got surplus for $5 that I remember was an absolute pig, it may have been even more like 40W or 45W. I had a Thinkpad W510 and that sucker does drain its battery even if you're just surfing. And it still has a discrete chipset. I would guess part of the idea of the ATX12VO is to get rid of some of the power overhead from maintaining those legacy 3v and 5v rails and get the idle power down on a beige box system. M.2 SATA or mSATA is a pretty low ask these days and you can have a small converter for that, pretty much everything else runs at 12v. Much better to have a small point-of-use converter for that than to have a big rail in your PSU. There are some mini-ITX boards (or thin-ITX?) that just have a DC barrel jack, I kinda like that idea in principle but it might be a bit much in terms of current.
|
|
#
?
Feb 25, 2022 08:12
|
|
|
My Ivy Bridge 3470 optiplex draws like 65w at idle which isn't great, though it's with a GTX 1070 so who knows how much of it is from that.
|
|
#
?
Feb 25, 2022 09:07
|
|
|
The sandy bridge based mac mini pulls 10 watts idle, so it's not impossible for an old system to be low power, but a lot of them weren't designed with that in mind.
|
|
#
?
Feb 26, 2022 01:15
|
|
|
Do we have any good goss on the Intel 12th gen laptop parts, especially the H class/45W?
|
|
#
?
Feb 26, 2022 13:59
|
|
|
Lol yeah never mind, I've got the classes confused vvv mobby_6kl fucked around with this message at 21:12 on Feb 26, 2022 |
|
#
?
Feb 26, 2022 18:03
|
|
|
Am I missing something? H series parts are already on sale and reviewed? It's the lower power P/U parts we haven't seen yet. https://www.techspot.com/review/2410-intel-core-i7-12700h/
|
|
#
?
Feb 26, 2022 18:09
|
|
|
A little out of the loop when it comes to 12th gen Intel and DDR5 but I'm helping a buddy price out PC parts for a new build. Mainly gaming (primarily esports titles) and podcasting but as far as I can tell, he'd be fine perfomance wise with a DDR4 platform to save some money right?
|
|
#
?
Feb 26, 2022 23:01
|
|
|
jisforjosh posted:A little out of the loop when it comes to 12th gen Intel and DDR5 but I'm helping a buddy price out PC parts for a new build. Mainly gaming (primarily esports titles) and podcasting but as far as I can tell, he'd be fine perfomance wise with a DDR4 platform to save some money right? Absolutely yes, no need to go for DDR5 at the moment
|
|
#
?
Feb 26, 2022 23:10
|
|
|
Is it a bad idea to benchmark CPU changes with XTU without doing a system reset. I changed the core ratio from 4.6 to 5.1 and got about a 10% boost with Intel's own benchmark. Then I reverted to stock settings to get a baseline in 3DMark. But 3D Mark crashed after 5 runs of the CPU profile benchmark (again this was at the default 4.6 ratio). Temperatures were well below 60C.
|
|
#
?
Mar 3, 2022 20:17
|
|
|
Benchmarking should always done from the same initial state, unless you're developing caching mechanisms where knowing the difference between numbers in cold and hot state is the only way you can know if your code is better or worse. If a benchmarking methodology doesn't include restarting, you can just go ahead and ignore every single result posted by whoever's making those benchmarks. If people don't report their benchmarking methodology, they're probably trying to deceive. BlankSystemDaemon fucked around with this message at 21:33 on Mar 3, 2022 |
|
#
?
Mar 3, 2022 21:31
|
|
|
Thanks, though I was more worried about instability than accuracy. I think the software's buggy too. After I changed the cache ratio (just out of curiosity, I read it probably wouldn't matter much or be detrimental), I got a lower score, so I changed it back to default. But I couldn't get the old score, and the CPU was acting strange even if I reverted everything to default. It would get much hotter than it did before at stock. This was all after multiple reboots. The fix that worked was forcing a crash by overclocking too much, which on the next boot actually restored the default settings, and everything's fine now. Could have been my fault, but idk how since I was using Basic Tuning, which only let's you control 2 settings (core ratio and cache ratio).
|
|
#
?
Mar 3, 2022 22:18
|
|
|
Is there a memory timing/sub-timing calculator for Alder Lake, like the Ryzen one?
|
|
#
?
Mar 3, 2022 23:06
|
|
|
https://www.anandtech.com/show/17295/asrock-industrials-nuc1200-box-series-brings-alder-lake-to-ucff-systems Dual 2.5Gbe NICs instead of the previous 1 Gbe + 2.5Gbe. Entirely overkill for anything homelab, but I definitely appreciate the push towards a new baseline of tech.
|
|
#
?
Mar 5, 2022 13:20
|
|
|
Finally. 1gbe's been the default for at least a decade, it's ridiculous.
|
|
#
?
Mar 5, 2022 15:41
|
|
|
mobby_6kl posted:Finally. 1gbe's been the default for at least a decade, it's ridiculous. It does feel ridiculous, but I also haven't seen a good cheap home switch with anything faster than 1gbe.
|
|
#
?
Mar 5, 2022 15:51
|
|
|
Twerk from Home posted:It does feel ridiculous, but I also haven't seen a good cheap home switch with anything faster than 1gbe. Mikrotik has a 8+1 port 10GbE SFP switch that is like $250, also there are some 8-12 port 2.5GbE switches (base-T so they don’t need SFP modules for every port) springing up around the $200 price point. Netgear has an interesting switch (MS510TXM) with 4x2.5gbe, 4x 10gbe (with 2.5/5 support), and two SFP 10gbe for $500, and QNAP has one with 12 ports where 8 are switchable between either SFP or base-T around the $550 price point. But that’s still a bit on the pricey side imo. Even most nerds won’t stomach more than $200 and the general public probably needs to fall into the <$100 range before they can possibly be upsold.
|
|
#
?
Mar 6, 2022 00:11
|
|
|
The general public is on 100 or 200mbit internet 99.99% of their traffic is coming down that pipe. It's just an unfortunate situation where enthusiasts want a better feature set and the market isn't there to drive it yet.
|
|
#
?
Mar 6, 2022 00:53
|
|
|
https://twitter.com/VideoCardz/status/1500446127375765506?t=oTueTesjsqg7kutG6Gr9EA&s=19 It's just one stick of either, so this isn't anything more than an office PC curiosity, but it's still notable that someone did it
|
|
#
?
Mar 6, 2022 13:22
|
|
|
COVID and semiconductor shortages mean that the mad scientists over at ASRock haven't gotten their usual shipments of cocaine.
|
|
#
?
Mar 6, 2022 15:32
|
|
|
11900K for $350 at MC https://www.microcenter.com/product/633421/intel-core-i9-11900k-rocket-lake-35ghz-eight-core-lga-1200-boxed-processor-heatsink-not-included
|
|
#
?
Mar 7, 2022 03:52
|
|
|
My 12700k just will not run my 3600 DDR4 RAM at 3600 in Gear 1. 3600+ Gear 2 works fine, and so does 3500 Gear 1 at standard voltages. But 3600 Gear 1 will not pass Memtest even at 1.4V DDR4 and 1.25V system agent. Running a Gigabyte board on the latest BIOS. CPU seems fine otherwise. Stable at 52 (3-core)/51(6-core)/50(8-core) with E-cores at 40 and ring at 42 with a -0.050 Vcore offset. (Memory was tested before overclocking the CPU. Currently running 3500 Gear 1.) Can the integrated memory controller be a dud even though the rest of the chip is great?
|
|
#
?
Mar 7, 2022 17:37
|
|
|
Boat Stuck posted:My 12700k just will not run my 3600 DDR4 RAM at 3600 in Gear 1. 3600+ Gear 2 works fine, and so does 3500 Gear 1 at standard voltages. But 3600 Gear 1 will not pass Memtest even at 1.4V DDR4 and 1.25V system agent. I believe gigabyte has the worst DDR4 memory compatibility out of all the board manufacturers so that might have something to do with it. Maybe a newer bios revision will help in the future otherwise the difference between 3500 and 3600 should be negligible.
|
|
#
?
Mar 7, 2022 21:26
|
|
|
Also if you really care you can try loving around with changing subtimings manually and seeing if that's the issue. It probably isn't but maybe it's worth trying, you might give up less performance that way.
|
|
#
?
Mar 7, 2022 21:50
|
|
|
My understanding is the difference between like a 3200 and 3600 kit with roughly equivalent subtimings is normally less than 5% even in most production softwares. Just don't worry about it
|
|
#
?
Mar 9, 2022 04:25
|
|
|
What timings you have? I use "standard" 1.35V with 2 kits of 2x8GB 3600CL15 g.skill on adus z690 strix-a. So that is 4x8GB single rank modules with 12700K. They did not function at cl15 reliably, so I put 16-16 etc. timings instead of pushing voltage. I think the vccsa/io voltage is maybe 1.1V? I should probably look again. I think it was like 1.25V with XMP settings, but I was really afraid of so high voltages so I dialed it down. Do you use one kit or two kits? How many dimms? Single or dual rank sticks? Ihmemies fucked around with this message at 08:15 on Mar 9, 2022 |
|
#
?
Mar 9, 2022 08:12
|
|
|
Ihmemies posted:What timings you have? I use "standard" 1.35V with 2 kits of 2x8GB 3600CL15 g.skill on adus z690 strix-a. 3600C15 will be bdie, which ships with lifetime warranties at 1.6v. You don't need to worry about going above 1.35v. Four sticks is always harder to run than two, which is taken into account when binning 4 stick kits. So you typically won't get full speed out of two 2 stick kits versus a single 4 stick kit rated at the same speed, because the 4 stick kit needs to be a better bin. DRAM voltage is very low impact on the CPU itself, up to around 1.65v is fine for daily usage if the memory DIMMs can run with high voltage but it'll likely need active cooling at those voltages. 1.5v is typical for daily without extra cooling. VCCSA/VCCIO affect CPU longevity, but 1.25v is not overly high. 1.35v is more like the daily limit for the high end, and at 1.1v you're laughing. I'd try 1.4v VDIMM with your xmp settings given its four sticks.
|
|
#
?
Mar 9, 2022 09:18
|
|
|
B-Mac posted:I believe gigabyte has the worst DDR4 memory compatibility out of all the board manufacturers so that might have something to do with it. Maybe a newer bios revision will help in the future otherwise the difference between 3500 and 3600 should be negligible. Ihmemies posted:What timings you have? I use "standard" 1.35V with 2 kits of 2x8GB 3600CL15 g.skill on adus z690 strix-a. I'm using 4x16GB Crucial Ballistix DDR4-3600, rated 16-18-18-38, Micron E-die, dual-rank I believe. I'm currently running them at the XMP voltages and timings, just 3500 instead of 3600, Gear 1. Runs at 3600 and higher without an issue in Gear 2. I'm leaning more and more towards it being a Gigabyte issue, since the CPU is great otherwise. I may theoretically be able to solve things through subtiming fuckery since the board does POST at 3600 Gear 1 (it's Memtest86 that fails), which means I can play around with subtimings manually until it works, but I think I'll just accept 3500 for now, and see if Gigabyte ever comes out with a better BIOS. I do understand that 4 DR sticks can be hard on the memory controller. It's just interesting how everything is completely fine at 3500 at stock voltages and just goes nuts at 3600 even with much higher voltages. The frequency wall is real. Boat Stuck fucked around with this message at 21:07 on Mar 9, 2022 |
|
#
?
Mar 9, 2022 21:05
|
|
|
https://www.anandtech.com/show/17308/the-intel-w680-chipset-overview-ecc-for-alder-lake-workstations They're speculating that all 12th gen Core CPUs will support (unbuffered) ECC with W680 instead of just the lower-rung SKUs, which is really interesting. If this means Xeon E is dead it could only really be a good thing.
|
|
#
?
Mar 10, 2022 09:01
|
|
|
Well 4 dual rank sticks are the hardest to drive. And those memory timings are already quite slow already. All I can say is that the starting point is not great 😐
|
|
#
?
Mar 10, 2022 09:13
|
|
|
Anime Schoolgirl posted:https://www.anandtech.com/show/17308/the-intel-w680-chipset-overview-ecc-for-alder-lake-workstations I am very  right now regarding the AVX512 section.it looks like they hammered out the power spike --> throttling issue in Sapphire Rapids, and I'm very upset that I'm going to have to wait another couple years to see this in person because one client just refreshed a little niche model workload onto past gen W. Some workloads are still very near and dear to my heart and I would have loved to see that bad boy purr on this gen.
|
|
#
?
Mar 14, 2022 20:01
|
|
|
I honestly don’t quite know what to make of the whole AVX-512 saga. The whole thing feels very last-minute, like iirc it’s mentioned in some programming guides, and obviously it wasn’t comprehensively disabled. If it was last minute, why? Intel didn’t think of the problems of heterogeneous instruction set? Their solution didn’t work? Some thing in the kernel they needed didn’t happen? Silicon bug that is problematic enough to cancel support over? Literally putting the design on the silicon, paying for a pretty huge unit (it’s probably >10% of the total area) on every unit you produce that is wasted space, not even turning it on for some limited enterprise skus? It feels like a silicon bug, but I guess they’re committing to it for next gen too? The whole thing is wack, Intel totes has money and the scale to do Skylake and Skylake-512 parallel designs if they want. I guess it does make a good silicon heatsink at least? Paul MaudDib fucked around with this message at 05:46 on Mar 15, 2022 |
|
#
?
Mar 15, 2022 05:39
|
|
|
I guess there could still be E-series Xeons coming with the little cores disabled so AVX-512 can run. That would be my guess as to why it was designed the way it was. I get that them dropping the ECC lock on Alder Lake hints that there wont be an E-series Xeon, but it could still happen. It is still bizarre to pack AVX-512 units into the desktop SKUs given how many masks of Alder Lake that Intel already makes. At least 3-4 already? Would think they would cut em out for one of the masks that will absolutely never touch it, like the mobile parts. Unless they want to reuse that for embedded?
|
|
#
?
Mar 15, 2022 07:10
|
|
|
|
| # ? May 8, 2024 11:41 |
|
|
Paul MaudDib posted:I guess it does make a good silicon heatsink at least? Dark silicon in action.
|
|
#
?
Mar 15, 2022 09:49
|
|
