|
Panty Saluter posted:I don't know what I'd do with all that cache but I want it Have your system crash from cache bitflips most likely. I wonder how long it will be before ECC pushes down to the consumer space since the cost difference is fairly small these days and the large cache/dimm sizes are just asking for problems.
|
#
¿
Apr 4, 2016 21:45
|
|
|
|
| # ¿ May 9, 2024 02:06 |
|
|
I thought only on the Xeon line, but I could be wrong on that.
|
|
#
¿
Apr 4, 2016 21:57
|
|
|
Captain Hair posted:I would be curious to see how my system would hold up with a 970 gtx fitted though. Do games ever pass stuff to cpu if the gpu is maxed out? Just a thought as that would flip the whole situation on its head really but I'm not sure any games actually do that? There are some game engines that, when you take the graphics settings down to low, certain things that would have been done on the GPU are pulled back to a CPU thread which can actually result in reduced performance. Don't have a comprehensive list, but it absolutely is a thing that exists in certain situations.
|
|
#
¿
Apr 6, 2016 01:50
|
|
|
Sidesaddle Cavalry posted:No it wasn't I mean, if you're going for the ultimate NUC, might as well get the SSD of SSDs Does that thing have some kind of ribbon cable that connects to a PCIe slot? I've only seen the M.2 and the conventional expansion card ssd's, that one is new to me.
|
|
#
¿
Apr 11, 2016 14:54
|
|
|
priznat posted:This form factor (sff-8639 or U.2 as Intel is calling it) is mostly for enterprise backplanes etc.. The cables mechanically identical to the somewhat defunct SATA-Express so we might see some motherboards with ports but I think it'll be rare, slot and m.2 are gonna be the vast majority of consumer grade for a while. Nice. One of the issues I've bumped in to with VM hosts it how to handle SSD caching drives, either you put them on the SAS/SATA controller and eat the throughput limitations or you stick them in the PCIe expansion slots and cross your fingers that you don't need to add any additional expansion cards for NICs or whatever. Glad to see they're working to make the disk backplanes usable again.
|
|
#
¿
Apr 11, 2016 16:40
|
|
|
This gives a pretty good rundown on how to connect a U.2 drive to current systems, if you were ignorant like me. http://www.tomshardware.com/news/sff-8639-u.2-pcie-ssd-nvme,29321.html
|
|
#
¿
Apr 11, 2016 16:57
|
|
|
Subjunctive posted:And the NVMe controller chips run hotter? They're supporting a 32gbit bus instead of an 6/12gbit one and you're pulling more of the logic up from the system board directly to the drive.
|
|
#
¿
Apr 11, 2016 19:39
|
|
|
HMS Boromir posted:They probably mean this. Unfortunate coincidence as far as the name goes. The top benchmarks on their website are all Xeons, followed by the 5960X, then the 5930K and 5820K. If a Xeon isn't an option the 5820K seems like the obvious choice. If its anything like raytracing in revit, they're CPU bottlenecked but can scale to as many cores as you can throw at it so Xeons usually fare well because you're getting 8+ of them or whatever. the problem generally is that lots of this software when you're moving through a 3d model normally its bottlenecked on a single CPU thread, so you have to balance CPU clock for normal usage and massive parallelism for render jobs. this could all be sorted out if they just dumped the render jobs to OpenCL/CUDA where it belongs but: LOL that industry is so hosed I'm not holding my breath. What we're currently doing is issuing a normal i5 system with a high clock speed/4 cores for normal design/review usage and then we keep a pool of 2 or 3 high-end xeon desktops with a bunch of cores that people RDP in to when they need to fire off the render job because that is less common.
|
|
#
¿
Apr 19, 2016 17:54
|
|
|
RH goes down as temps go up so just turn off all the power management features and leave that sucker on 24/7.
|
|
#
¿
May 6, 2016 15:05
|
|
|
canyoneer posted:https://www.reddit.com/r/IAmA/comments/4j15bs/were_the_intel_team_behind_the_new_skull_canyon/ What those idiots want is a Steam Link, which already exists.
|
|
#
¿
May 15, 2016 02:34
|
|
|
Bullshit. http://www.cpubenchmark.net/laptop.html Even your bottom of the barrel Intel mobile chips (your i3's or m5's, 15W TDP) can spank the high-end stuff from five years prior. They are insanely more efficient today and still improving and you don't get to make poo poo up whole cloth.
|
|
#
¿
May 18, 2016 17:54
|
|
|
Paul MaudDib posted:Top Skylake i3 was missing, next highest (6100H): 3820 And how many cores do the processors you are quoting have to achieve those benchmark scores you are citing? I was intentionally comparing 2 core processors but you're throwing 4 core ones in which is a complete bullshit comparison unless you account for it.
|
|
#
¿
May 18, 2016 18:21
|
|
|
Boiled Water posted:Rule of thumb: if the curvature exceeds that of a medium sized plate your fibers are gonna bust. bend radius for 50micron is like 5cm so not nearly as bad as it used to be. a fairly stiff pvc shield is probably sufficient for consumer use but I don't really see the point when copper could handle those short runs fine
|
|
#
¿
Jun 2, 2017 18:21
|
|
|
priznat posted:The IBM POWER9 has PCIe Gen 4 slots apparently, but good luck getting one of those if you're not a massive hyperscale or enterprise customer. PCIe 3 tops out at what, 30W per slot without additional connectors? Should be nice for some of the absurdly dense NVMe devices that I assume are coming up.
|
|
#
¿
Jun 8, 2017 21:18
|
|
|
"Go buy an 18core i9 if you really want to Game" Go gently caress yourself. Even the host couldn't hide his revulsion.
|
|
#
¿
Jun 9, 2017 17:12
|
|
|
Eh, it's going to take a latency-sensitive process that can effectively leverage >8 threads or memory intensive enough to have to reach through the crossbar to hit pages on a different memory controller before any of that on Ryzen is going to have a meaningful impact. Those things are few and far between, mostly the OS resource scheduler will avoid splitting the process on the crossbar and it won't matter.
|
|
#
¿
Jun 19, 2017 20:59
|
|
|
ArgumentatumE.C.T. posted:If the core is boiling, shouldn't the coolant be a little more... enthusiastic about cooling? Actually grabbing onto some of that heat and doing something with it? You're talking about a temp reading on a tiny amount of silicon which has very little mass and thermal conductivity which interfaces with a hunk of metal that has many times more mass. The core is going to hold on to a lot of heat under load, and those numbers are useless unless we have other chips to compare against.
|
|
#
¿
Jun 19, 2017 23:11
|
|
|
I've been running BF1 well enough on a 2500k so I am inclined to say that is a load of crap.
|
|
#
¿
Jun 21, 2017 19:04
|
|
|
TomR posted:Are the clock speeds locked down on those types of chips, or do the motherboards from servers just not offer those types of adjustments anyway? Xeon multipliers are locked, so you get the P-states as they ship. Better cooler = more time available in a given P-state. Servers are going to prioritize stability over raw performance.
|
|
#
¿
Jun 23, 2017 16:01
|
|
|
It's going to be a while before most things scale well beyond 4 threads, I'd say Intel still has an edge in the near term. Get the K variant so you have the option for overclocking and stretching it down the road.
|
|
#
¿
Jun 23, 2017 22:35
|
|
|
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
|
|
|
Malloc Voidstar posted:only fools use water cooling, intellectuals (me) use the more convenient noctua nh-d15s What do they think that second fan is really going to be doing for them?
|
|
#
¿
Jul 14, 2017 21:49
|
|
|
craig588 posted:It's big enough that you don't need a 3rd fan because it lines up with your existing exhaust fan (You are running a x38 Panaflo, right?). You can run all 3 at 600RPM and still overclock to 4.5GHz or whatever. Taking out a fan means the others have to run faster to meet the same cooling performance. More and bigger fans mean less noise. Putting fans in front of them isn't additive to their airflow, you're creating back pressure by shoving the air in to the blades of the next fan spinning at the same rate. There's not nearly enough material obstruction or distance to justify any of that and its idiotic.
|
|
#
¿
Jul 14, 2017 22:29
|
|
|
B-Mac posted:Anyone have experience delidding a 6700k? I have one core that gets 13C hotter than the others when playing some games, up to 86C spikes at times. Fairly certain the cryorig h7 is mounted properly with max-4 paste and I have no desire to go water. My idles temps are between 25-30C. There's quite a few people on HardOCP with good results for delidding that gen of processor https://hardforum.com/threads/6700k-delid-before-and-after-results.1878870/ Both temp and fan speed way down under the same load on a heavy overclock. Keep in mind that most games favor one of their cores more heavily than others and you will see a temp differential because that one will be hitting turbo clocks more than the rest. But it does seem like a good candidate regardless.
|
|
#
¿
Jul 19, 2017 19:30
|
|
|
Cygni posted:I would also check the PSU if you got a spare around. Weird power related crashes are due to dying PSUs a lot of the time. I would suspect degradation of caps for the for the CPU voltage regulators with that age.
|
|
#
¿
Jul 24, 2017 18:32
|
|
|
Paul MaudDib posted:Ah, a connoisseur of the "slap it back on there" thermal paste application method. I had a service tech one time who thought you had to apply the entire tube for every socket. That was a great mess to scrape out of an expensive dual-socket blade board.
|
|
#
¿
Jul 27, 2017 10:52
|
|
|
Did he Dremel that poo poo out to make it fit or just force it until the fiberglass gave way?
|
|
#
¿
Aug 28, 2017 17:43
|
|
|
There's almost no justification for a full mini-tower for standard office machines and lots of places have moved over to some kind of SFF model because who needs a big box of air, and those have less space to work with on the MB headers, but you still get full size DP there for no reason. Only seem to get used on laptops.
|
|
#
¿
Sep 26, 2017 22:07
|
|
|
What is "hot"? Idle temps are probably going to be somewhere around 40c because its reducing fan speed to minimize noise. Assuming it isn't intended behavior, it sounds like the heatsink is mounted wrong.
|
|
#
¿
Sep 26, 2017 23:34
|
|
|
Not to mention Kaby lake hardware will be getting liquidated out the OEM dell/hp/whoever channels for at least another year or two
|
|
#
¿
Oct 11, 2017 00:06
|
|
|
MaxxBot posted:The 2500k/2600k is still a decent chip for gaming and that's gonna be 7 years old soon. Yeah, I'm still on mine and the performance gains are only now getting compelling enough to make me consider an upgrade
|
|
#
¿
Oct 11, 2017 16:36
|
|
|
BIG HEADLINE posted:Overclockable and already-overclocked quad cores without hyperthreading have ~12-18 months of comfortable use left. Hyperthreaded OCed quads probably have 24-36 months of headroom. 67/7700K owners shouldn't be sweating much. Hyper threading adds MAYBE a 5% performance improvement under very specific workloads that typically aren't games. You're really overselling its value.
|
|
#
¿
Oct 17, 2017 20:38
|
|
|
craig588 posted:I'm a HT decrier, I run my 5820K with HT disabled because it hurts performance in stuff I do, but I think the idea is once stuff starts really using 6 cores the 4C8T variants of processors will be able to hold on a little longer than the straight 4C4T versions. HT doesn't add cycles, it just allows you to jam instructions through the gaps in the execution pipeline. In the best case, you're pushing instructions through there that have no latency requirements (for some reason) so they won't negatively impact anything but in real world scenarios you're just incurring additional latency for that little bit of extra output. This is great for large computational loads that run in batches like you get in business environments where throughput trumps transactional latency, but its generally poo poo for games and overcommit past your physical cores and in to the logical HT ones is a recipe for poo poo performance and latency/concurrency problems.
|
|
#
¿
Oct 17, 2017 20:51
|
|
|
crazypenguin posted:Curious what you're basing this on? Digital Foundry's tests for i5 vs i7, for example, show 30% for games that multithread well. Which is consistent with the gains for many other (non-gaming) parallel workloads. We've done extensive testing of it with business workloads and the rub basically is that if you are seeing large HT gains, it because your worker threads are full of holes because your code is poorly optimized. Once the load is back up because something is busy, those holes in the pipeline disappear and now that second thread you were trying to jam through a single core is forced to queue along side the primary thread and the latency goes to poo poo. Yeah, your max FPS or whatever during periods were you aren't CPU bottlenecked is going to improve but the low FPS periods where it is CPU bottlenecked is going to be worse than if HT wasn't being utilized because of the additional latency/queuing/contention. Avg numbers doesn't mean much when your floor is down and standard deviation is up in both directions. HT is a blunt, dumb tool and optimization inside the existing worker threads on 1 thread per core is better bang for the buck because of the increased consistency.
|
|
#
¿
Oct 17, 2017 21:03
|
|
|
Sure get out the dremel and hack off the interposer and you're good to drop it in
|
|
#
¿
Nov 9, 2017 19:36
|
|
|
I hate Linus so much
|
|
#
¿
Nov 17, 2017 15:03
|
|
|
How common were chipped dies back in the day when Athlon shipped like that? I went the P3 to P4 route so there was always an IHS sitting in there.
|
|
#
¿
Nov 20, 2017 21:39
|
|
|
Don Lapre posted:p3 had a bare die I had Slot 1, baby
|
|
#
¿
Nov 20, 2017 22:52
|
|
|
|
| # ¿ May 9, 2024 02:06 |
|
|
In more fun news, don't forget to apply security patches to your processors: https://security-center.intel.com/advisory.aspx?intelid=INTEL-SA-00086&languageid=en-fr
|
|
#
¿
Nov 20, 2017 22:53
|
|