|
lol do not use intel networking
|
#
¿
Apr 9, 2019 20:03
|
|
|
|
| # ¿ May 9, 2024 02:37 |
|
|
Do like nics that start randomly rebroadcast traffic to each other while in S3 for no reason, effectively looping your network if two or more systems start doing that behavior on a segment at a time? Do you like wildly unstable linux drivers that will drop link EVERY loving TEN SECONDS that they refuse to fix and instead just wallpaper over the problem by reinitializing the driver as quickly as possible? WELL HAVE I GOT A PRODUCT FOR YOU. https://www.youtube.com/watch?v=zoNtXVMOutE
|
|
#
¿
Apr 9, 2019 20:08
|
|
|
Dell had to eat a full swap in our VMware cluster from X710's to whatever Qlogic was selling because of how poo poo they were. Spent over 6 months fighting with them and trying every conceivable firmware and driver combo we could think of and they were still horrible. Incredibly fast, could do 10gig line speed on 64 byte frames, but the firmware and drivers were atrocious. The qlogics could only hit 4-5gig on standard 1500 byte frames and would only hit 10gig on jumbos but at least they could hold a drat link.
|
|
#
¿
Apr 9, 2019 21:47
|
|
|
Wild EEPROM posted:Would you rather use an intel nic or a Realtek one though I would honestly prefer the realteks because they don't cause broadcast storms in the middle of the night on systems in S3.
|
|
#
¿
Apr 10, 2019 13:59
|
|
|
movax posted:When did the Intel NICs all go to poo poo? I remember making very conscious choices in my home hardware to make sure I had all Intel NICs. Though, they are all 82xxx series GigE NICs, not 10GbE. We've been having issues on our optiplex fleet with embedded intel nics for at least the last five years requiring firmware updates to correct.
|
|
#
¿
Apr 11, 2019 02:54
|
|
|
PCjr sidecar posted:All that Ethernet overhead? Gross. we got super-jumbo 16 and 64k frames now, its not too bad. even my 5 year old Force10 gear would support up to 12k.
|
|
#
¿
Apr 12, 2019 14:59
|
|
|
PCjr sidecar posted:Packet size doesn’t necessarily help with small message latency though. You're talking about microseconds of overhead, it doesn't matter in the vast majority of use cases and you should be using Optane if it does.
|
|
#
¿
Apr 12, 2019 18:12
|
|
|
priznat posted:Was there a leaked roadmap for server stuff too? just glue two or more of the desktop roadmaps together
|
|
#
¿
Apr 24, 2019 20:08
|
|
|
Twerk from Home posted:This is hardly true because they remain the default option in large parts of the high-margin market. Top laptop SKUs and the 4 socket or more market are going to remain entirely Intel, and AMD isn't even competing in those spaces. You're going to be able to do a quad socket intel workload on a dual-socket epyc 2 in the very near future. I'm sure intel can glue some more chips together to get higher core density but they're charging so much loving money at this point is more cost effective to buy more 2U dual-socket AMD servers even if you need twice as many as 4U quad socket intels.
|
|
#
¿
Apr 25, 2019 02:21
|
|
|
Intel has a lot of yield issues with these monolithic designs, I seriously doubt they will be able to beat AMD on price with the ability to do much more favorable binning on the chiplet architecture.
|
|
#
¿
Apr 25, 2019 04:14
|
|
|
Twerk from Home posted:Is hyperthreading more of a performance benefit on bigger Xeons than on the smaller client chips and Xeon E? How big of a deal is it to turn off hyperthreading on the big boys, Xeon Golds and Platinums? HT performance benefit is completely dependent on your workload being one that benefits from it. With that said, if you need a Gold/Plat Xeon with that many cores odds are your workload parallelizes well and will benefit from HT.
|
|
#
¿
May 14, 2019 20:26
|
|
|
Khorne posted:Most of the recent exploits are not a threat to consumers on their own hardware. They still allow leaking sandboxed process or kernel data to a non-privileged user processes. There's not usually much of payoff of executing that kind of attack against consumers but there's money to be made by stealing bank account creds or whatever someone is bound to try it.
|
|
#
¿
May 16, 2019 21:36
|
|
|
DrDork posted:Except, as has already been noted, consumers aren't really much at risk, and the mitigations only minorly impact performance. For big data, where disabling HT might actually be painful, the cost of throwing out and replacing all your poo poo with enough AMD processors to get back to the same level of performance you had with Intel would be hilariously more expensive than just disabling HT / patching and buying a couple extra Intel boxes to make up the difference. If your workload parallelizes well, the higher end Epyc SKUs (7551 in particular) will already give you 2-3x more ops for your dollar over a similar Platinum Xeon. We're already moving and Zen2 only makes the advantage even greater with doubling the core density.
|
|
#
¿
May 17, 2019 16:29
|
|
|
Sorry, are you fabbing your own systems from scratch? The whole point of an OEM is to offload those to someone else.
|
|
#
¿
May 17, 2019 16:38
|
|
|
Twerk from Home posted:They already are, the pricing is far more aggressive on the 62xx CPUs, and 42xx silvers are priced to be able to fight 2S vs AMD 1 socket. It's not far more aggressive, you get about 20% more CPU cycles for your dollar compared to the 1st gen scalable xeon. Intel Gold Xeons are somewhat competitive on pricing against the 7501 and 7601, but the 7551 is the obvious choice as its 5% less CPU for half the cost. For raw CPU cycles, you need Xeon platinum chips to compete with the huge quantity of cycles an Eypc socket gives you and you're paying a 4x price premium to do it. Yeah, your all core turbo frequency on Intel is going to be 15% higher but that doesn't mean much when you're getting 50% more cores on the AMD side.
|
|
#
¿
May 17, 2019 17:14
|
|
|
Shipon posted:Kind of hard not to patch when software and hardware vendors cram patches down your throat by mandatory device updates. So read the documentation and adjust accordingly. Every single one of the mitigations that can have a performance impact gives you the ability to override and disable it on desktops, and they’re opt-in on servers.
|
|
#
¿
May 18, 2019 14:20
|
|
|
Khorne posted:That's actually pretty cool. I mean, it clearly benefits Intel but it also benefits researchers. No, it sucks because they're trying to manipulate the research and the amount they were offering to an entire research team is absolute peanuts compared to what they're going to be able to pull down in salary by doing the full disclosure and graduating with this under their belt.
|
|
#
¿
May 23, 2019 15:57
|
|
|
karthun posted:Usually these things come with a 3-6 month NDA so the supplier can fix the problem in the public before the issue is made public. They still get the full disclosure and credit. Intel was already getting that, its par the course for responsible disclosure. Intel was trying to bribe them to make the impact look less severe. It was a poo poo move by a company caught with its pants down
|
|
#
¿
May 23, 2019 17:19
|
|
|
surf rock posted:Oh, great, sounds like it'll be a short wait then. One question: Zen2 is AMD, right? The last time I did a computer purchase, folks seemed to think that AMD was just a total joke. Have they managed to catch up or even leapfrog Intel over the last half-decade? Buldozer was an absolute joke, but Zen is pretty good if you're will to trade off clock speed for more cores and Zen2 looks like its going to close the gap real fast
|
|
#
¿
May 23, 2019 17:54
|
|
|
KillHour posted:Maybe if you're just looking at the server hardware, but once you factor in the extra switches, PDUs, software licenses, storage, hardware failures and management overhead? Assuming your workload parallelizes well, a Epyc 7551 32c@2.55ghz is going to run you ~$2250/socket while the most comparable Xeon is the Platinum 8168 24c@3.4ghz for $5,928/socket. So yeah, 40U of 7551's are going to run you $90k in processor costs while the Xeon would be $240k. Even with the potential cost overhead you're talking about this isn't even close. Maybe if you're doing blades or FX2 with high CPU density the embedded switches and shared PSUs will make it break in Intel's favor but if you're just looking for conventional U pancake servers Intel's offerings are poo poo. And if AMD can get traction I would expect to start seeing the OEMs packaging Epyc platforms in to blade or other modular chassis configs. BangersInMyKnickers fucked around with this message at 16:34 on May 28, 2019 |
|
#
¿
May 28, 2019 16:30
|
|
|
Twerk from Home posted:List pricing is not relevant at all for high end Xeons. There's enormous discounting that happens at the vendor level, supported by direct deals the vendors do with Intel. I'd also say that in general the Platinums are for 8-socket configs, because you can run quad socket with Golds without paying the 8S support price premium. How much are discount are you getting on those Platinum chips? Best we could squeeze Dell on was ~20% and it was still a terrible proposition.
|
|
#
¿
May 28, 2019 16:52
|
|
|
redeyes posted:When was this? I've never seen a Dell like that I believe it was the clam-shell gens. GX150 up to maybe the GX600 series. This was Pentium 3-4 days, long while back.
|
|
#
¿
Jun 25, 2019 16:50
|
|
|
Sandy bridge was just coming out when SC2 released. Most people's processors were older dual-core systems at best. It made zero sense for blizzard to code their engine to take advantage of systems with 4 or 8 cores when that hardware barely existed in the consumer space.
|
|
#
¿
Jul 3, 2019 16:17
|
|
|
Agreed posted:Got my 4770k to 4.4GHz at 1.28V right now, input voltage of 1.9V, and it seems quite stable. Using an Asus Z87 Sabertooth board, it has been straightforward enough to overclock. Factory sticker says 1.4vcore is safe and plenty of people are driving theirs upwards of 1.5v for years without ill effect. You should be well within tolerance.
|
|
#
¿
Jul 3, 2019 16:20
|
|
|
My processor keeps throwing sugma faults
|
|
#
¿
Jul 11, 2019 15:27
|
|
|
D. Ebdrup posted:I'm a little bit excited about is getting a quad-core non-SMT 7W CPU that has SHA512 as part of its supported crypto-primitives, and the higher bandwidth on the platforms. Why the hell are you trying to use sha512 for checksumming??
|
|
#
¿
Aug 5, 2019 18:04
|
|
|
I mean, yeah, if you've got QA cards to offload to then by all means go nuts but there just aren't enough permutations on a 128kb block to make a hash collision probable over sha256 compared to some other corruption event. https://blogs.oracle.com/bonwick/zfs-deduplication-v2 quote:When using a secure hash like SHA256, the probability of a
|
|
#
¿
Aug 5, 2019 19:02
|
|
|
lole
|
|
#
¿
Aug 22, 2019 16:36
|
|
|
Shipon posted:Does this mean software devs are actually going to have to put work in now? absolutely not, if they get faster systems then they can keep slacking off and being poo poo because Endless Abundance
|
|
#
¿
Sep 18, 2019 14:31
|
|
|
ECC without NMI is unacceptable for server applications, but I do not agree that is matters that much for home/gaming use. Even perfectly healthy dimms can bitflip occasionally and this will stop that from impacting you, as well as generally increasing overall reliability. Yeah, if you are losing a dimm then troubleshooting it will be a pain in the rear end but a memtest will generally turn it up eventually while reducing the likelihood of system problems being caused by ram.
|
|
#
¿
Nov 15, 2019 23:05
|
|
|
FuturePastNow posted:This isn't even as bad as P4 vs. Athlon 64. In the consumer space, yes the gap isn't that great. In the server market, Intel is only staying afloat with institutional momentum. Epyc chips beat them out on performance and cost pretty much across the board.
|
|
#
¿
Nov 27, 2019 15:56
|
|
|
DrDork posted:Most of the Spectre varients that could be applied through JS have been effectively neutered by the timing degradation applied by Chrome and Firefox (and probably Edge? Does anyone even use Edge?). There is a software solution, but its ugly. You basically have to manage what cores important threads execute on and keep everything else the hell away of it. Important thread executing? Get everything else off that core and force the cache to evacuate once its done. On bare metal, this is probably doable now with a 4+ core system, but it will hurt. Try to do this on a virtual hypervisor with a heavy vm load? lol no. At that point if you're trying to protect important "enterprise things" you're far better off either dividing your VMs in to dedicated clusters based on risk of arbitrary code execution, locking down code execution so nothing untrusted can run (this includes stopping things like browsers/python/java/etc that have some intermediate code interpretation layer), or just saying gently caress it and turning off HT taking the performance hit.
|
|
#
¿
Dec 4, 2019 22:41
|
|
|
Yeah, having a reasonable chance at uplifting a 4/6 core budget system to 16 or possibly more is something Intel can't really touch. The platform has more longevity, especially if you are the kind of person buying in the $/performance sweet spot and refreshing every few years to save some cash.
|
|
#
¿
Dec 9, 2019 19:37
|
|
|
The processor is going to be what dictates being able to physically support ddrl since that's where the memory controllers are. But the motherboard/firmware from the OEM can override the compatibility matrix and send any voltage they see fit to the dimm, so if Dell is locking you out of that then tough luck. With that said, a 1.35v dimm will probably run at 1.5v well enough.
|
|
#
¿
Jan 6, 2020 20:42
|
|
|
Those run ~250W PSUs right? 1.5 vs 1.35v on the dimms isn't going to be an appreciable difference on a desktop system, maybe a watt or two. It's really more for battery savings where you're trying to shave every last watt. The GPU and CPU are going to be your big draws, if you're seeing voltage dips with the config you're so far past the capacity of your PSU that ddr3l isn't going to help anything.
|
|
#
¿
Jan 6, 2020 21:14
|
|
|
FRINGE posted:If youre building out sets of office machines that need some power for loads of parallel tasks, but the only display power they need is at the level of browsers, Office, and maybe VS/sql, the integrated stuff with a mediocre motherboard are fine. it was a big gamechanger when the intel integrated stuff started supporting dual-displays. there was a $100/box line item that everyone could drop overnight
|
|
#
¿
Jan 24, 2020 14:21
|
|
|
I never realized Delta's logo was a loudspeaker which feels very fitting
|
|
#
¿
Jan 29, 2020 00:42
|
|
|
|
| # ¿ May 9, 2024 02:37 |
|
|
future ghost posted:I sold all my Deltas awhile back for about $30 each.. No idea who wanted them or why. I had one of these fat fuckers and the static pressure was enough to move it along tables like a tethered hovercraft: http://www.sidewindercomputers.com/df12gf.html Oh yeah, I'm not complaining about the performance. These are the things that were shipping in my servers for years and so long as you had thermal management and a proper fan curve they're good for the job. Absolutely sucks having to bench a system in your office with one of these things for any amount of time
|
|
#
¿
Jan 29, 2020 15:35
|
|