|
Cygni posted:i was talkin 14th gen only, my bad. yeah 13th gen is an absolute mess with ECC support. bolds support it, non-bolds do not: Ha, nice chart, I was just digging through ARK getting more and more bewildered every time. -T have it, none of the -F or -KF have it, and xx500 or above have it, but xx400 or below do not. Except for apparently the 13490F, which does, because ?????
|
# ? Dec 21, 2023 01:11 |
|
|
# ? May 30, 2024 02:56 |
My recollection matches that of Cygni, that you need a W680 board to actually use DIMM-wide ECC and subsequently also get Machine Check Events, or whatever the newer API is, sent to the OS for it to decide what to do with any particular program (assuming it can crash it, and not have to crash itself because it's somewhere in kernel-space).
|
|
# ? Dec 21, 2023 09:33 |
|
Cygni posted:Rumors are that production costs are much higher and yields much lower than the N5-family, and that the promised power and density targets were missed. And it was late. I think you're paying way too much attention to the rumor mill, a lot of which is just completely fabricated clickbait nonsense. Most of this "M3 is a disappointment" narrative seems to come from sources who first tried to attract attention by talking up how giant a leap forward M3 would be, then when it inevitably failed to meet ridiculously lofty expectations, declared it a disappointment and began trying to make drama about TSMC N3 being bad. Meanwhile back in the real world M3 is a substantial upgrade over M2, especially GPU performance. Production costs are undoubtedly higher, but that's a trend that started a long time ago and is expected. You will never get to know the real yield numbers because fabs tend to keep that poo poo deeply secret, often even from their immediate customers (though Apple's a big enough account which gets special-enough treatment that Apple may actually know).
|
# ? Dec 21, 2023 11:15 |
|
BlankSystemDaemon posted:My recollection matches that of Cygni, that you need a W680 board to actually use DIMM-wide ECC and subsequently also get Machine Check Events, or whatever the newer API is, sent to the OS for it to decide what to do with any particular program (assuming it can crash it, and not have to crash itself because it's somewhere in kernel-space). Why do you need W680 specifically, and not any LGA 1700 board / chipset with ECC support? Does Intel lock out support to entirely prevent Core processors from working in C262 / C266 boards? It looks like maybe, look at the CPU support list here: https://www.asrockrack.com/general/productdetail.asp?Model=EC266D4U-2L2T quote:Supports Intel Xeon E-2400 series and Intel® Pentium® Gold G7400/G7400T processors
|
# ? Dec 21, 2023 15:23 |
Twerk from Home posted:Why do you need W680 specifically, and not any LGA 1700 board / chipset with ECC support? Does Intel lock out support to entirely prevent Core processors from working in C262 / C266 boards? You've linked a DDR5 board, and as I'm sure you know DDR5 has per-chip ECC with both motherboard and memory manufacturers doing an absolute piss-poor job of making it clear when they distinguish between meaning per-chip ECC and DIMM-wide ECC - to the point that I don't even expect ASRock to get it right, unless you're talking to one of their engineers. For what it's worth, Tom's Hardware does report it (although they bury the lede a bit, by not including it in the headline), and WCCFTech also published it just like Anandtech did.
|
|
# ? Dec 21, 2023 16:08 |
|
BlankSystemDaemon posted:The why is obvious: Market segmentation. Intel has been doing it for decades. Sure, but those articles are all from a year before C262/C266 launched and W680 was the only option for LGA1700 ECC. If Intel really did lock Core processors out of their server chipsets this time, it's the first time that has happened in years!
|
# ? Dec 21, 2023 16:11 |
|
Intel's Client and Datacenter groups are like two different companies.
|
# ? Dec 21, 2023 18:25 |
|
BobHoward posted:I think you're paying way too much attention to the rumor mill, a lot of which is just completely fabricated clickbait nonsense. Most of this "M3 is a disappointment" narrative seems to come from sources who first tried to attract attention by talking up how giant a leap forward M3 would be, then when it inevitably failed to meet ridiculously lofty expectations, declared it a disappointment and began trying to make drama about TSMC N3 being bad. Thats a fair point, a lot of tech media is now driven by the histrionic "leak" race, which is insanely annoying. Some of N3's struggles have also been in slightly more trustworthy sources like DigiTimes, though. Ultimately I was using N3B as a reference point for the concept that intel's rapid planned node changes might make current products obsolete more quickly than in the recent past. Using the M2 -> M3 improvements as a benchmark and putting all the other rumored portions like cost and yield aside, I don't think the planned node shrinks are going to quickly invalidate current products like they did in older generations. The performance improvements from node shrinks are just smaller now than they used to be. Twerk from Home posted:Sure, but those articles are all from a year before C262/C266 launched and W680 was the only option for LGA1700 ECC. I don't see Core support on any of the C262/C266 board data sheets, but most are just launching now. I would say someone might check in a review, but honestly I don't think anyone (except maybe STH?) is really gonna care enough to review these products. This "mid power/mid performance" workstation niche is insanely small now to the point of near irrelevance.
|
# ? Dec 21, 2023 18:56 |
|
Cygni posted:I don't see Core support on any of the C262/C266 board data sheets, but most are just launching now. I would say someone might check in a review, but honestly I don't think anyone (except maybe STH?) is really gonna care enough to review these products. This "mid power/mid performance" workstation niche is insanely small now to the point of near irrelevance. I realize they're niche, I'm planning a 1U/1A colo build and they seem to be the best option for that still. The listed supported Pentiums are Core processors, right?
|
# ? Dec 21, 2023 19:16 |
|
Twerk from Home posted:The listed supported Pentiums are Core processors, right? Correct, the Pentium Gold G7400 is a dual-core, hyperthreaded Alder Lake part with no E-cores. It will be the very last Pentium-branded chip ever. SpaceDrake fucked around with this message at 19:30 on Dec 21, 2023 |
# ? Dec 21, 2023 19:22 |
|
until the reboot in a few years
|
# ? Dec 21, 2023 19:31 |
|
Twerk from Home posted:The listed supported Pentiums are Core processors, right? Yeah, i was using the term "Core" in the brand sense. Just the E-series Xeons and Alder Lake Pentiums were on the support lists for the boards i could find.
|
# ? Dec 21, 2023 19:58 |
|
SpaceDrake posted:Correct, the Pentium Gold G7400 is a dual-core, hyperthreaded Alder Lake part with no E-cores. It will be the very last Pentium-branded chip ever. I thought the mobile Pentium G8500 was a neat chip in theory. 1P+4E, 6 threads, low power. Would have made a lot of Chromebook users very happy but I think I only ever saw one machine that used it. Maybe we'll see some cool ultra low power tile stuff. Something that could be passively cooled perhaps. I really like fanless Chromebooks but they seem to be the domain of ARM stuff.
|
# ? Dec 21, 2023 20:06 |
|
Cygni posted:Thats a fair point, a lot of tech media is now driven by the histrionic "leak" race, which is insanely annoying. Some of N3's struggles have also been in slightly more trustworthy sources like DigiTimes, though. Ultimately I was using N3B as a reference point for the concept that intel's rapid planned node changes might make current products obsolete more quickly than in the recent past. Using the M2 -> M3 improvements as a benchmark and putting all the other rumored portions like cost and yield aside, I don't think the planned node shrinks are going to quickly invalidate current products like they did in older generations. The performance improvements from node shrinks are just smaller now than they used to be. Yeah that's fair. Node shrinks don't deliver the same raw gains they used to, so expecting new nodes to immediately make old products obsolete isn't how things work any more. Although Intel specifically might well be able to obsolete a lot of their old products if they ever manage to get off their own runaway train and design a new P core from scratch. They've boxed themselves into a P core optimized to win single thread benchmarks at any cost, and the downsides don't help them make well rounded products which are actually good in the real world, especially in any power constrained application. But this wouldn't have anything to do with process node, of course.
|
# ? Dec 22, 2023 02:00 |
|
BobHoward posted:Yeah that's fair. Node shrinks don't deliver the same raw gains they used to, so expecting new nodes to immediately make old products obsolete isn't how things work any more. I've been speculating that the Atoms, the E cores, are Intel's future. I cannot wait for Sierra Forest to get a glimpse of it. I realize it only has 128-bit vector units and not much memory bandwidth per core, but it's still going to be the most interesting part in ages. poo poo, I have no idea why Intel still put QPI on them. I guess technically UPI, but still isn't that a decent chunk of extra I/O that goes wasted in a single socket configuration, which I'd bet is relatively common for core counts that high? Why didn't the Xeon-D line get as much attention? If I recall correctly Xeon-D was designed explicitly for both performance per watt and value for total compute, which made it really effective to scale up to a gazillion nodes: https://engineering.fb.com/2016/03/...cking-up-power/ Xeon-D is still around but it doesn't seem to be a huge or thriving segment from what I can tell. I could be wrong!
|
# ? Dec 22, 2023 02:26 |
|
Yeah I love the Xeon-D stuff and it seems criminally undermarketed or something, it’s just so neat!
|
# ? Dec 22, 2023 02:34 |
Twerk from Home posted:Why didn't the Xeon-D line get as much attention? If I recall correctly Xeon-D was designed explicitly for both performance per watt and value for total compute, which made it really effective to scale up to a gazillion nodes: https://engineering.fb.com/2016/03/...cking-up-power/ Unfortunately, Intel is the only company to offer something like the Xeon-D. BlankSystemDaemon fucked around with this message at 02:39 on Dec 22, 2023 |
|
# ? Dec 22, 2023 02:34 |
|
Twerk from Home posted:I've been speculating that the Atoms, the E cores, are Intel's future. I cannot wait for Sierra Forest to get a glimpse of it. I realize it only has 128-bit vector units and not much memory bandwidth per core, but it's still going to be the most interesting part in ages. Xeon-Ds are for stuff like your rack mounted appliances - NAS, networking gear, etc. In other words, a “boring”, lower margin market you won’t hear too much about. edit: based on the responses I assume there’s another type of Xeon-D?
|
# ? Dec 22, 2023 02:41 |
|
hobbesmaster posted:Xeon-Ds are for stuff like your rack mounted appliances - NAS, networking gear, etc. In other words, a “boring”, lower margin market you won’t hear too much about. Apparently Facebook used multiple generations of them for a lot of their front-end web serving, which is indeed boring but I'd think that they chose them for a reason. Both the link in my post and https://www.servethehome.com/twin-lakes-intel-xeon-d-2100-platform-powering-facebook/ for the next generation.
|
# ? Dec 22, 2023 02:45 |
|
I looked at them for a project or two when the first few generations came out but the economics never really worked out over a smaller quantity of regular Xeons.
|
# ? Dec 22, 2023 03:03 |
|
I remember when the first gen came out thinking it was perfect for NAS usage -- built in high speed networking, low TDP, lots of PCIe lanes. The newer ones you can get on boxes from Supermicro and co. now... they are intended for 'edge' usage, so I guess if you want to drop a storage appliance at the base of a telecom tower?
|
# ? Dec 22, 2023 23:12 |
|
movax posted:I remember when the first gen came out thinking it was perfect for NAS usage -- built in high speed networking, low TDP, lots of PCIe lanes. The newer ones you can get on boxes from Supermicro and co. now... they are intended for 'edge' usage, so I guess if you want to drop a storage appliance at the base of a telecom tower? Edge is such a terrible marketing word, but you can absolutely imagine what say these things are for: https://www.advantech.com/en/products/ece3ad39-aace-4ee0-ac2d-14526d12eaa0/fwa-5072/mod_9cabe57d-9c69-4194-b22d-f0383f643e95
|
# ? Dec 22, 2023 23:16 |
|
everybody is edging these days, the word is losing its meaning
|
# ? Dec 22, 2023 23:36 |
|
Twerk from Home posted:What's the point of these things? Just use i5s and i7s, they seem better in every way. the desktop xeons are really not for server workloads at all these are for huge pools of memory and enormous numbers of pcie lanes for workloads that'll actually get saturated if behind an oversubscribed switch on the consumer desktop devices edit: gonna emphasize that no sane person is going to run virtualization that actually matters on these. virtualization is for chips with consistent and linear uplift of feature set generation to generation. Potato Salad fucked around with this message at 23:54 on Dec 22, 2023 |
# ? Dec 22, 2023 23:50 |
|
Potato Salad posted:the desktop xeons are really not for server workloads at all the E-2400 series he was talkin about has the same 20 lanes as the desktop parts, and same max of 4 dimms. you might be thinkin of the W-2400 Sapphire Rapids parts Cygni fucked around with this message at 00:03 on Dec 23, 2023 |
# ? Dec 23, 2023 00:01 |
|
Potato Salad posted:the desktop xeons are really not for server workloads at all And this is why “edge” is a word instead of server - apparently every 5g tower need VMs for a vRAN doesn’t count
|
# ? Dec 23, 2023 06:19 |
|
I have a question for anyone with a Raptor Lake system, do you experience stutters when autoscrolling in Chromium browsers (Chrome, Edge, Opera), specifically in image heavy sites and while adjusting autoscroll speed? I built two RPL systems, one with an ASUS board and another with an MSI board and I get a stutter when autoscrolling in Chrome and Edge that can go from being subtle to downright choppy. This happens in both Win11 22H2 and 23H2. I found a good way to check this is to load macrumors.com on both Chrome and Edge side by side and autoscroll through each browser while moving the mouse to adjust the scroll speed. The only thing that seems to mitigate it is to set the power plan to high performance mode, but its odd that its required to make browsing not feel like crap. I have an ancient 2500K system that runs Chromium browsers smooth as butter, so its frustrating to have stutters on powerful new hardware.
|
# ? Dec 30, 2023 21:02 |
|
By "auto scroll" do you mean the window scrolls down when the mouse reaches the bottom?
|
# ? Dec 31, 2023 02:37 |
|
Rocko Bonaparte posted:By "auto scroll" do you mean the window scrolls down when the mouse reaches the bottom? I mean middle clicking to bring up the circle thing and moving the whole mouse to scroll.
|
# ? Dec 31, 2023 05:10 |
|
I might be able to ask around this upcoming work week post-New-Years, but don't expect much.
|
# ? Dec 31, 2023 19:17 |
|
Have you upgraded to LogoFail patched bioses yet? My asus mobo claimed "3. This update includes the patch for the LogoFAIL vulnerabilities" in patch notes released 2 days ago. It's good to check, because it kind of allows BIOS rootkits on level which the running OS can't simply detect.
|
# ? Dec 31, 2023 19:33 |
|
LegitMaan posted:I have a question for anyone with a Raptor Lake system, do you experience stutters when autoscrolling in Chromium browsers (Chrome, Edge, Opera), specifically in image heavy sites and while adjusting autoscroll speed? I built two RPL systems, one with an ASUS board and another with an MSI board and I get a stutter when autoscrolling in Chrome and Edge that can go from being subtle to downright choppy. This happens in both Win11 22H2 and 23H2. I have a 13600KF and macrumors scrolled smoothly on chrome using the middle mouse click.
|
# ? Dec 31, 2023 19:51 |
|
Looks like my NUC 9 Extreme finally punched out for good. I think the power supply died.
|
# ? Jan 2, 2024 19:11 |
|
I haven't seen a single review of Meteor Lake or anyone really digging into the new features anywhere, what's up with that? Meanwhile, Emerald Rapids availability is immediate and easy, impressive launch there.
|
# ? Jan 4, 2024 16:17 |
|
Twerk from Home posted:I haven't seen a single review of Meteor Lake or anyone really digging into the new features anywhere, what's up with that? There have been a couple of reviews: https://www.youtube.com/watch?v=Obtc24lwbrw https://www.youtube.com/watch?v=oGAcGnBFfHk It seems to be some kind of weird limited launch though so I don't know what's going on.
|
# ? Jan 4, 2024 16:34 |
|
Twerk from Home posted:I haven't seen a single review of Meteor Lake or anyone really digging into the new features anywhere, what's up with that? Because it's mobile only, you can't really find reviews of just the chip, you have to review a whole platform. There are some out there, though. CPU performance seems unexciting, the GPU is apparently real good, and some are finding very good battery life, but battery life is gonna be down to so much more than the just the soc. https://hothardware.com/reviews/intel-meteor-lake-review?page=1
|
# ? Jan 4, 2024 16:35 |
|
mobby_6kl posted:It seems to be some kind of weird limited launch though so I don't know what's going on. Reviewing mobile parts is not as fast or straightforward as socketable desktop parts. This isn’t unexpected.
|
# ? Jan 4, 2024 16:37 |
|
mobby_6kl posted:It seems to be some kind of weird limited launch though so I don't know what's going on. feels like a "we told the stock holders that we would launch by the end of the year, and we did!* *two devices that are out of stock" sort of situation. unfortunately similar in launch to the similar in performance 7840U.
|
# ? Jan 4, 2024 16:58 |
|
The NUC is back, now with an aesthetic that is somehow even less appealing to non-enthusiasts/gamers. Goodbye weird skull, hello weird RGB gamer bullshit: https://videocardz.com/newz/asus-rog-nuc-confirmed-launching-at-ces-2024
|
# ? Jan 5, 2024 23:22 |
|
|
# ? May 30, 2024 02:56 |
|
Hats off to the guy on the NUC team that found a Skull Canyon on the map to satisfy the geographic naming convention guidelines.
|
# ? Jan 6, 2024 00:07 |