|
BIG HEADLINE posted:When the warranty on the external WD runs out, I'll shuck it out of its enclosure and RAID 0 the drives. As it stands, though - I like having the ability to 'mirror' the drives manually via manual backups. Do you mean RAID 1? Or do you actually have a use for 16TB of fast but unreliable storage?
|
#
¿
Sep 3, 2017 16:54
|
|
|
|
| # ¿ Apr 19, 2024 23:20 |
|
|
priznat posted:What happened was after about 4-5 years the device wouldn't boot anymore and then all the drive data was unusable because I couldn't plug it into a pc, had to be in one of the same enclosures. They were EOL'd at that point so it would have been real hard to find a new one. I've run into this multiple times both for myself and clients. The solution is UnRaid, and despite its quirks it's been rock solid and blazing fast, even with nine drives, four dockers and two VMs running on a wheezy old AMD quad core. I will never buy a proprietary NAS again.
|
|
#
¿
Sep 5, 2017 23:54
|
|
|
ufarn posted:I mainly want the new Thunderbolt since the Z390 doesn't looktoo useful except for transfer speed. It's a small thing but I (evidently) don't upgrade too often. You're really going to wait a whole year for a new chipset instead of buying a $50 PCI card?
|
|
#
¿
Sep 8, 2017 19:19
|
|
|
I never thought I'd be perfectly happy with this 4770K five years later. I've gone through 2 coolers and 2 motherboards, and each time I thought "I wonder if this is worth the effort considering there's a new generation coming out", but nope. It just sits there, humming away at an easy 4.6Ghz overclock at 1.265v on a 120mm rad, never goes above 80C, and does everything I ask of it. The one and only downside is there are only 2 lanes for the m.2 slot on the board, so I'm not getting everything I could out of my boot drive, but I can barely feel the difference between SATA and NVME anyway, and I think that goes for pretty much everyone else as well. It's the longest I've ever owned a CPU. I suppose someday it'll be retired to server duty and live out its days transcoding Plex streams and the occasional VM or two, but I just can't imagine that happening in the next two or three years. I can understand upgrading a Sandy Bridge or earlier machine for the creature comforts and better IO options, but I guess I don't "get" the 8770k for anyone with Haswell or newer.
|
|
#
¿
Nov 3, 2017 01:11
|
|
|
craig588 posted:Most Haswells are only 4 cores. The whole big thing about Coffeelake is 6 cores in the mainstream platform. I'm going to move to an 8 core Broadwell eventually since I'm already on an X99 motherboard, but for most people Coffeelake is their first opportunity to get a 6 core processor for a reasonable price. I get that. Having one chip that can render like a boss without sacrificing single threaded performance is exciting. I guess it's just not better enough for me, especially with a die shrink right around the corner and bonkers DDR4 prices.
|
|
#
¿
Nov 3, 2017 02:11
|
|
|
Three-Phase posted:I’ll just put one of those cold-packs from the freezer on it for a temporary power up. I hope this is a joke.
|
|
#
¿
Nov 16, 2017 16:55
|
|
|
Paul MaudDib posted:I mean, the truth is that in 6 months nobody will care about this patch. Fixups happen, as long as there is actually a fixup it's fine. 0-5% performance impact is nothing at all as these things go. It's a clear, obvious, provable ~30% hit for iops with no roadmap for a better solution. The charts are rolling in in this very thread. This is a giant problem for datacenters. Do you own a lot of intel stock or something?
|
|
#
¿
Jan 6, 2018 20:11
|
|
|
cool startup feel posted:Intel CPU and Platform Discussion: Nice Meltdown A thousand times yes.
|
|
#
¿
Jan 9, 2018 03:01
|
|
|
Potato Salad posted:If we see these patched out like normal bugs in a month or two, yeah nothing I don't even see the point in rushing to patch this stuff. All of these "exploits" require elevated privileges and/or physically flashing a vendor signed bios patch. It's like pointing out that if someone beats you up and steals your wallet they'll have access to your bank cards. No poo poo.
|
|
#
¿
Mar 14, 2018 22:51
|
|
|
Paul MaudDib posted:As for getting defensive, I also literally pointed out that Intel had had a similar exploit less than 6 months ago. No, they didn't. None of the exploits CTS found were usable from a virtualized environment, and they never even theorized a remote vector for any of them. Spectre / Meltdown are serious vulnerabilities, the CTS findings are absolutely not.
|
|
#
¿
Apr 8, 2018 19:13
|
|
|
Generic Monk posted:also grounding strap is pussy poo poo and I've never needed one. never broke a piece of computer hardware by handling it ESD is real, latent failure is real, it's not debatable, there's decades of research and proof. Ground yourself.
|
|
#
¿
Apr 23, 2018 17:54
|
|
|
fishmech posted:Macs aren't top end laptops except in price though. What on earth does that mean?
|
|
#
¿
Jul 18, 2018 18:54
|
|
|
Cygni posted:Gskill formally announced those weird double height DIMMs only compatible with a few ASUS boards. It makes a bit more sense now as ASUS is putting 2 memory slots on their high end overclocking boards now to hit higher speeds, so you need the double height dimms to get to the 64gig maximums the memory controller supports. Those things are going to be absurdly expensive and have a 99% return rate from people who don't realize it's non-standard and only works in a few boards (and some people who have the right board but suddenly find they can't fit an air cooler on it). Manufacturing enthusiast boards with only 2 RAM slots is dumb in the first place. Creating special non-standard, double-height RAM to "solve" the problem you engineered in the first place is loving astoundingly dumb.
|
|
#
¿
Oct 11, 2018 18:48
|
|
|
karoshi posted:Now with 32GB of HBM! Yeah but gamers don't really need 32GB, so what if we just did 8gb of HBM2, and since it's fast enough you could share it with the GPU and stick it in a supermini case and... we just invented the playstation 4
|
|
#
¿
Oct 11, 2018 21:52
|
|
|
BIG HEADLINE posted:Well, I'd advise against using Conductonaut on bare copper. The nickel plating is what insulates it from damage, even though it doesn't attack copper as badly as it does aluminum. Steve at GN did a video about Conductonaut's effects on different metals - and as I recall, it stains copper pretty readily, but I'd be worried more about long-term degradation. Gallium (or galinstan) will alloy with copper, but it shouldn't invade and destroy it like aluminum. You'll end up with an alloy layer on the copper that won't scrub off, but it shouldn't cause any problems. seriously though, keep it the gently caress away from aluminum. one drop will completely destroy any aluminum part. gallium don't give a poo poo. gallium nasty.
|
|
#
¿
Oct 23, 2018 17:02
|
|
|
il serpente cosmico posted:I haven't been able to find a clear answer on this, but what is the likely final for shrink for silicon? 3nm? Those numbers haven't really meant much since the switch to finfet. In terms of actual channel length and width between fins and such, TSMC's "7nm" and intels "10nm" were the same thing. It's more a branding term at this point.
|
|
#
¿
Oct 24, 2018 04:30
|
|
|
il serpente cosmico posted:Huh, I wonder why the 8700k generally beats or matches the 9900k when clocked the same? Still waiting on that DF review. They should be very close at the same clock. It's all still Skylake on 14nm+++++++++++++, the big differences are packaging, binning, cache and hardware meltdown fixes.
|
|
#
¿
Oct 24, 2018 21:50
|
|
|
My de-lidded 4770k has been sitting at 4.6ghz on a 120mm AIO for like 6 years now at 1.28 volts or so. Never goes over 76c. De-lidding felt dicey at the time, but I re-did the liquid metal a year ago and it hadn't changed much at all. The paste was in far worse shape.
|
|
#
¿
Oct 30, 2018 22:35
|
|
|
Yudo posted:Nehalem is part of the reason I never really bothered with OCing Haswell: OCing Nehalem was fun and more necessary due to still not quite being "good enough" for some applications. Plus you could get away with crazy poo poo and not set your PC afire. Between the process shrink, the FIVR, and TIM, the thrill was gone with Haswell. Haswell is definitely one of the trickier overclocks, but it wasn't actually much of a hassle after the de-lid process. FIVR is weird, especially on Haswell where it'll happily let you blow up your chip with seemingly reasonable settings, but it was the first time I was able to get a VID offset OC working properly, which is nice because it keeps the power saving features somewhat intact. It's a good thing to learn though because it's basically modern OC technique without the training wheels.
|
|
#
¿
Nov 1, 2018 20:55
|
|
|
Siets posted:Can anyone link to an idiot-proof Haswell OC guide? I’m in the same boat where I’m waiting on AMD 7nm to see what happens and I would like to give OC’ing a try to squeeze some extra FPS out of my system until then. Was planning to just pick up a Hyper 212 Evo or similar tier air cooling solution. Luckily I bought the 4670K so I can actually OC it, I think because the no-OC proc wasn’t in stock at my Micro Center at the time. Sure. Here: 1. Get a decent cooler and apply it properly. The 212 evo is a bit long in the tooth, consider either an AIO (they're all basically the same, whatever you can fit in your case, the bigger the better), a Noctua DH15 or a Dark Rock Pro 4. Yes, these are all sort of expensive options, yes they're worth it because you'll probably keep them far longer than you will your CPU. 2. Set your VCore voltage in your BIOS to 1.25v. Do not use "adaptive" or "offset" settings, those are bad, bad danger settings for haswell. Stay out. AVX2 loads can and will blow up your chip. No use. No. Manual only. 3. Set your multiplier to 4ghz, boot your machine and run the AIDA64 stress test. Use Intel XTU or whatever utility you want to keep an eye on your core temps and voltage. If you see your voltage changing, stop immediately because something ain't right, go back to step 2. If your temps are climbing past 78C, ya goofed again, go back to step 1. 4. Reboot and bump your multiplier up to 4.2. Run the test again for 30 minutes. Repeat this cycle, bumping up by 100mhz at a time until either you crash or your core temp is over 80C. Once you find this spot, drop back down 100mhz and burn it in for 12 hours. If it's stable, you're good to go. Congrats, you did it hooray. Obviously you can spend more time tweaking the balance between voltage and temperature. Anything consistently under 80C is fine for a 24/7 overclock, as is anything under 1.3v vcore. And you can get into the hairy terrain of adaptive vcore offsets and ring bus voltage and memory overclocking, but that's all "if you're bored" territory.
|
|
#
¿
Nov 1, 2018 21:51
|
|
|
craig588 posted:Huh? I've been using offset with a Haswell for like 3 years, at idle I have like 1 volt and loaded it's like 1.3V The old wisdom used to be that fixed voltages were better for stability testing, but unwise for long term use. What changed? I have too, but you have to be careful. Haswell's FIVR has a big problem with AVX2 loads like the ones used in stress tests. It'll shoot the voltage up to dangerous levels with what seems like a reasonable offset otherwise. Dude asked for a foolproof guide, so I didn't touch on anything that could cause actual damage to the CPU. You can usually switch to offset once you find your sweet spot, but I'd still stay away from adaptive entirely. There's nothing wrong with fixed voltage long term as long as temps stay under 80, other than wasted power. Also, absolutely none of this applies to Skylake and later, just to be clear. edit: Siets posted:Honestly I was all set to go with trying my first OC this weekend but this post scared me. see above, if you're nervous just make sure your voltage stays put under 1.3 and you'll never have to think about it. Dr. Fishopolis fucked around with this message at 22:49 on Nov 1, 2018 |
|
#
¿
Nov 1, 2018 22:43
|
|
|
Palladium posted:Looks like the "OC is quick and easy" is still so as long as you don't give a poo poo about thermals and longevity. We're talking about a specific problem with Haswell chips and adaptive voltage, and I posted how to do it safely and easily. If you keep your temps under 80c, there's zero longevity concern. Anything Skylake and newer is even easier, and modern XMP means you barely have to think about it. I don't really know where your comment is coming from but it isn't true.
|
|
#
¿
Nov 2, 2018 19:53
|
|
|
BobHoward posted:everything goes through power planes in the organic substrate. excuse me the what now?
|
|
#
¿
Jan 7, 2019 08:32
|
|
|
BobHoward posted:The substrate material in many chip packages is an organic material in the chemistry sense, meaning it's largely composed of carbon. Organic substrates are several sheets of a glass filled polymer material laminated with epoxy. That's fantastic. I always assumed that was regular old PCB, but of course that makes sense. Thanks for explaining!
|
|
#
¿
Jan 8, 2019 08:33
|
|
|
craig588 posted:Doubling the core count can do nothing if your apps are not able to use 8 cores. If they can though, doubling the core count will double the performance. Not really, software doesn't scale like that. You'll get a benefit in (properly) multithreaded applications, but double the core count will never actually double the performance. Also, while things have been slow since Haswell, a new 14nm chip is still going to give you a very significant performance boost over Sandy Bridge at the same core count just from the DDR4, better power handling and OC headroom, infinitely better motherboard OC tools and a decent IPC advantage. All that said, a well overclocked 3570K is still enough power for most modern applications and games, but that's not going to be true in a year or two, and an upgrade would be noticeable even at the same core count.
|
|
#
¿
Jan 10, 2019 19:10
|
|
|
K8.0 posted:Yes, but the answer to your original question comes in two parts. It's also reasonably likely that at least one of Zen 2 and Ice Lake will bring a bigger performance bump than we've seen in years, so if you don't need the additional performance right now, it's also not a bad time to wait. Yeah, that's the necessary caveat I forgot to put in my post. We're months away from 7nm and there's still a massive shortage of 14nm parts, so if you can wait, then wait.
|
|
#
¿
Jan 10, 2019 19:47
|
|
|
WattsvilleBlues posted:So is Plex video streaming one of the things that would benefit from a newer CPU or more RAM? If you're transcoding multiple streams over the internet or to low-powered devices, yes. For in-home use, it barely matters. My plex server is some old 4 core bulldozer thing with 8gb ram and it's vast overkill for the 8 people who use it, since it's always sending to either a Roku, an nvidia Shield or someone's computer or phone, all of which gets locally decoded. edit: for reference, it used to be a 2011 mac mini with a 2-core mobile processor and I never ran out of headroom with that either. Dr. Fishopolis fucked around with this message at 21:31 on Jan 10, 2019 |
|
#
¿
Jan 10, 2019 21:29
|
|
|
I switched to unRaid ages ago and never looked back. It's not RAID at all, it's basically JBOD with parity, which is excellent for a shitload of reasons.
|
|
#
¿
Sep 1, 2019 16:32
|
|
|
I don't really get how any of that is useful unless you need to play around with that specific platform and would rather take the time to deal with an incredibly sketchy motherboard than spend money.
|
|
#
¿
Sep 6, 2019 16:50
|
|
|
SwissArmyDruid posted:Originally linked that, but I kind of felt like it was preaching to the choir. Ain't nobody here under the delusion that Intel somehow looks good doing this, right? paging paul maudib to the intel thread, paul maudib to the intel thread.
|
|
#
¿
Nov 25, 2019 16:44
|
|
|
repiv posted:avx512 already has reduced precision ops meant for ML, and the next iteration is adding matrix FMA ops akin to nvidias tensor cores I actually don't get AVX512. Intel is never, ever going to catch up with GPUs for floating point math, why would you push your cpu to do a thing your GPU can do in half the time for half the power?
|
|
#
¿
Jul 25, 2020 16:35
|
|
|
repiv posted:CPU SIMD still has its place, offloading to the GPU is a high latency operation that only pays off if you have a sufficiently large batch of work to do all at once. If you have a lot of small batches then the CPU can probably finish them before the GPU would even start executing. That makes sense to a point but then wouldn't the thermal and silicon budget be better spent on including actual tensor cores (or equivalent?) Doesn't even need to be a full blown igpu. It just feels like intel is trying to halfass a problem that's already solved. e: assuming they're doing this in good faith and it isn't just for benchmarks Dr. Fishopolis fucked around with this message at 19:15 on Jul 25, 2020 |
|
#
¿
Jul 25, 2020 19:10
|
|
|
Stubb Dogg posted:Having CPU that had proper tensor cores with shared memory with CPU would be a killer and we’d switch our inference workloads there immediately. I guess that answers my question from a page ago. I wonder who will do it first.
|
|
#
¿
Jul 26, 2020 16:12
|
|
|
the market is definitely dumb, it's literally a reactive graph of rich people feelings about things they barely grasp. however, right now the market feels like AMD and Nvidia are gonna make a bunch of cash in the next few years while intel... isn't, which seems like the world's most obvious take.
|
|
#
¿
Jul 29, 2020 15:23
|
|
|
I'm still happily gaming on a delidded and thoroughly overclocked 4770k. I'm just getting to the point where frametimes are starting to get out of whack in newer engines (i.e. Decima) but i'm still not gonna bother upgrading that rig until zen 3 at least. edit: vvv 
Dr. Fishopolis fucked around with this message at 23:37 on Jul 29, 2020 |
|
#
¿
Jul 29, 2020 23:24
|
|
|
Of course they're overvalued, but so is literally the entire market. It's a bit silly to even talk about with the whole thing due to collapse at any moment.
|
|
#
¿
Jul 30, 2020 01:36
|
|
|
Saying AMD "only" has 10% server market share is a bit of a weird point to make in favor of intel, considering it was around 1% a little over three years ago.
|
|
#
¿
Jul 30, 2020 03:27
|
|
|
A lot of what you're describing is just regulatory capture and institutional rot leading to forever contracts, intel could live on for eons that way. Maybe they'll get real weird with it like Microsoft does. Or they could just become AOL, quietly reaping billions from forgotten dialup subscriptions. I don't think AMD is going to leap into the government / education space that easily, but they're definitely creeping into AI and cloud compute and there be growth in them thar sectors.
|
|
#
¿
Jul 30, 2020 05:12
|
|
|
and mac os
|
|
#
¿
Jul 31, 2020 14:48
|
|
|
|
| # ¿ Apr 19, 2024 23:20 |
|
|
NewFatMike posted:They're not even going to bring over the fixed keyboard? Jesus. That design is hosed, they need to throw it out and start over. It's the 2008 macbook air all over again, except this time they can't admit their mistake.
|
|
#
¿
Aug 3, 2020 19:24
|
|