|
gently caress. I might have to dig up that effort post from the NAS thread / my notes where I evaluated every single mATX option (Node 804 case) that would give me the peRfeCt ESXi host to do everything I wanted. My current system would be perfect if 1) I stopped working long enough to bring it up and 2) Supermicro didn't gently caress me on the X11SSL and not power the iGPU, meaning I can never QuickSync video. I just want an rear end-ton of PCIe lanes (NVMe hello!), on-board SAS3008 / or enough spare PCIe slots to get my Chelsio NICs / HBA / etc in. But, insert argument here now about how there's enough CPU power to transcode a pile of streams and I only realistically have like... 2 users, and a 45 Mbit upstream and it's all really a moot point.
|
# ? Sep 9, 2021 19:56 |
|
|
# ? Jun 2, 2024 19:10 |
movax posted:gently caress. I might have to dig up that effort post from the NAS thread / my notes where I evaluated every single mATX option (Node 804 case) that would give me the peRfeCt ESXi host to do everything I wanted. My current system would be perfect if 1) I stopped working long enough to bring it up and 2) Supermicro didn't gently caress me on the X11SSL and not power the iGPU, meaning I can never QuickSync video. I just want an rear end-ton of PCIe lanes (NVMe hello!), on-board SAS3008 / or enough spare PCIe slots to get my Chelsio NICs / HBA / etc in.
|
|
# ? Sep 10, 2021 12:44 |
|
I'm on a Node 804 until I get my 3U DAS SAS machine setup and let me tell you that I dearly miss my hot swap bays when I'm trying to change out a bad hard drive. I am also definitely regretting not labeling my hard drives before putting them in. I'm hoping that I won't have any problems with it until I get my new array worked out. The U-NAS case I had before was decommissioned in a frantic week after enough mechanical issues, my PSU failing for the _third_ time, and the backplane getting wonky after almost a decade of crappy closets and basements. There was literally rust on some of these parts. I'm going to cannibalize my current X570 based workstation to be my new home server when I do a refresh in a year or so. I'm also a madman and have managed to successfully use my old NUC with the Thunderbolt eGPU enclosure to run with my SAS HBA to the 3U DAS. I'm really impressed that it worked without any hitches in my Ubuntu 20.04 LTS setup. I will probably regret this but I'm going to see how far this Rube Goldberg NAS topology can go, dammit.
|
# ? Sep 10, 2021 15:52 |
|
Another Alder Lake leak: https://sakhtafzarmag.com/%D9%82%DB...86%D8%AA%D9%84/ Core i9 12900K - 8C+8c - $599 Core i9 12900 - 8C+8c - $509 Core i7 12700K - 8C+4c - $429 Core i7 12700 - 8C+4c - $359 Core i5 12600K - 6C+4c - $279 Core i5 12600 - 6C+0c - $249 Core i5 12400 - 6C+0c - $209 Supposed Cinebench results:
|
# ? Sep 10, 2021 21:43 |
|
That graph header formatting seems pretty suspicious, but it'll be cool if true.
|
# ? Sep 10, 2021 22:20 |
|
The graphs in that article are hosed, the y-axes are all over the place. Still, looks like an 18% single core improvement.
|
# ? Sep 11, 2021 11:44 |
|
It's good that the 12600K will be less than $300 but I mourn the death of the sub-$200 CPU.
|
# ? Sep 11, 2021 12:14 |
|
oh ffs the 12400 isn't going to have little cores? succ also, those Cinebench numbers are for R23 I looked up techpowerup's review of the i5-11400 and they have a single-core score of 1,543, and a multi-core score of 11,653 for the Ryzen 5 5600X, which seems to match that leak's chart roughly looking at that chart for the i5-12600K: 1,900 for single-core over 17,250; under 18,500 for multi-core at 1,900 single-core that's completely past the scale: TPU has the i9-10900K at 1,371; the Ryzen 9 5950X at 1,658; the i9-11900K at 1,687 at 17,250 multi-core that's higher than the i9-10900K (14,319), the Ryzen 7 5800X (15,730), and the i9-11900K (16,190) only the Ryzen 9 3900X (18,556), the Ryzen 9 5900X (21,707), and the Ryzen 9 5950X (25,916) post higher that's... dang that's something, if true https://www.techpowerup.com/review/intel-core-i5-11400f/6.html https://www.techpowerup.com/review/intel-core-i9-11900k/7.html gradenko_2000 fucked around with this message at 14:11 on Sep 11, 2021 |
# ? Sep 11, 2021 14:08 |
|
I can't get over the -4000 in that 2nd chart.
|
# ? Sep 11, 2021 14:30 |
|
Dr. Video Games 0031 posted:It's good that the 12600K will be less than $300 but I mourn the death of the sub-$200 CPU. Yeah, the 4C/8T i3s were an outstanding value that got completely overlooked by the Rocket Lake update. Do we know that Alder Lake is not launching lower than i5? Is i3 and below going to be 14nm skylake / kaby / comet lake refreshes for all eternity?
|
# ? Sep 11, 2021 14:47 |
Actuarial Fables posted:I can't get over the -4000 in that 2nd chart.
|
|
# ? Sep 11, 2021 14:50 |
|
Twerk from Home posted:Yeah, the 4C/8T i3s were an outstanding value that got completely overlooked by the Rocket Lake update. the i9, i7, and i5 K models are launching this year everything else, namely i3 models, non-K models, Pentiums (and presumably Celerons) are launching Q1 2022 there's still going to be an i3, but it's also only going to be a 4c8t model, just with Alder Lake's Golden Cove big cores. Obviously that's a significance performance bump over the last/current Comet Lake i3, but it also is kinda intriguing to me because an i3 is exactly the kind of thing that would show up on a lot of office/non-high-performance contexts where a little core might be useful, and yet those are the ones that are lacking it. I'm guessing it has to do with die space/die design.
|
# ? Sep 11, 2021 15:06 |
|
Is there anything concrete on thermals/power consumption with alder lake? I'm not a fan of how much juice the current intels take, and how much heat they put out.
|
# ? Sep 11, 2021 16:23 |
|
Zedsdeadbaby posted:Is there anything concrete on thermals/power consumption with alder lake? I'm not a fan of how much juice the current intels take, and how much heat they put out. I would expect substantial efficiency gains as not only is there an IPC uplift but these will also be the first mainstream desktop chips made on 10 ESF / Intel 7. How much juice they take just like the current generation will likely be down to the user, at least in DIY space. Long term power limit is exposed even on low end motherboards. If you have enough cooling and want to you'll probably be able to run them at 200+ watts. If you want a chip that stays at 65 watts you can probably do that too at the cost of some clock speed. Modern processors are all pretty much choose your own adventure with both Intel and AMD. Flip side is because of the denser process they may be harder to cool at a given heatload than the 14nm chips. Thermal density and all that.
|
# ? Sep 11, 2021 16:43 |
Zedsdeadbaby posted:Is there anything concrete on thermals/power consumption with alder lake? I'm not a fan of how much juice the current intels take, and how much heat they put out.
|
|
# ? Sep 11, 2021 16:44 |
|
Zedsdeadbaby posted:Is there anything concrete on thermals/power consumption with alder lake? I'm not a fan of how much juice the current intels take, and how much heat they put out. The leak suggested a PL2 of 228W (vs 251 for rocket lake) and the same PL1 as the last several generations, 125W. There are still the 65W parts available as always, but the clocks will be lower. If you want cool/silent, those are the parts. Both AMD and Intel run their parts much further along the voltage/frequency curve than they used to, because they believe most DIY/gaming computer users don’t care about an extra 30-40W when they have a GPU pulling 300W+, and aren’t running them at max all core boost all day… and they are likely correct about that.
|
# ? Sep 11, 2021 20:10 |
NUMBER BIGGER!
|
|
# ? Sep 11, 2021 20:14 |
|
Cygni posted:Both AMD and Intel run their parts much further along the voltage/frequency curve than they used to, because they believe most DIY/gaming computer users don’t care about an extra 30-40W when they have a GPU pulling 300W+, and aren’t running them at max all core boost all day… and they are likely correct about that. I don't know how it is in the States, but here in the UK wholesale energy prices are going up through the roof this year (you can thank the most incompetent government since Eden for this, plus many many many more highly urgent & immediate problems facing our society right now) and it's something we think about more often than we typically like to
|
# ? Sep 11, 2021 21:38 |
|
Zedsdeadbaby posted:Is there anything concrete on thermals/power consumption with alder lake? I'm not a fan of how much juice the current intels take, and how much heat they put out. Bad enough that they're rumored to be making the boxed coolers better, finally: https://twitter.com/VideoCardz/status/1436372862215106561 I'm the asterisk next to 65W
|
# ? Sep 11, 2021 23:13 |
|
Twerk from Home posted:Yeah, the 4C/8T i3s were an outstanding value that got completely overlooked by the Rocket Lake update. The 11400's MSRP is less than $200, in theory, making it a really great value too. Too bad only OEMs seem to have decent access to it. All X400 CPUs have been less than $200, until Alder Lake. It's just how it's going to be from now on. AMD already rose their prices for Zen 3 and Intel would be foolish not to follow suit to some degree (the 12600K seems like a potentially good value still, though).
|
# ? Sep 12, 2021 11:24 |
|
FuturePastNow posted:Bad enough that they're rumored to be making the boxed coolers better, finally: You know they're doing this only to put leds on their coolers lmao
|
# ? Sep 12, 2021 13:17 |
|
A lot of articles popping up in my feed saying alder lake with ddr5 memory is worse for gaming than existing setups 😬 Apparently the bandwidth is very high but so is the latency, similar to the issues zen 1 had
|
# ? Sep 15, 2021 19:44 |
|
I havent seen any leaks about gaming specifically (links if ya got em?), but lots of stuff about this very likely being like every other DDR transition: at least at first, the cost/performance probably wont be worth it. There are lots of DDR4 boards leaked for Alder Lake already though, so at least theres that!
|
# ? Sep 15, 2021 20:39 |
|
The question is, are any of them based on data, or are they just the obvious predictions based on previous DDR transitions? To some extent the memory controller will matter, but a lot of it is down to what the manufacturers are speccing and what the module manufacturers are willing to validate/warranty. Early on, it's not likely to be so hot.
|
# ? Sep 15, 2021 20:42 |
|
K8.0 posted:Early on, it's not likely to be so hot. well it might be hot in the other sense https://www.digitaltrends.com/computing/corsair-ddr5-ram-requires-much-better-cooling/
|
# ? Sep 15, 2021 20:45 |
|
People complaining about chipset fans will seem quaint with DDR6s 12mm ram fans (one per chip).
|
# ? Sep 15, 2021 20:48 |
|
K8.0 posted:The question is, are any of them based on data, or are they just the obvious predictions based on previous DDR transitions? https://www.tomshardware.com/news/gear-4-mode-tested-on-alder-lake I didn't look at it too closely but it seems like the memory controller can only clock so high. To facilitate RAM speeds beyond what it can manage, there are different gears. To run your RAM at extremely high speeds you would use gear 4, meaning the memory controller's clock ticks only once for every 4 RAM ticks. This seems to be the source of latency. The article notes that gear 2 lead to worse game performance on existing chips.
|
# ? Sep 15, 2021 21:06 |
|
repiv posted:well it might be hot in the other sense I just hope there will be temp read outs now.
|
# ? Sep 18, 2021 18:08 |
|
https://videocardz.com/newz/specifications-of-intel-12th-gen-core-t-series-alder-lake-s-processors-have-been-leaked There's been a leak of T-model Alder Lake CPUs. These are low-power variants designed to run at a TDP of just 35w. This seemingly confirms that non-K i5 and i3 CPUs aren't going to have little cores
|
# ? Sep 19, 2021 15:44 |
|
6c/12t at 35W tdp is not bad at all and a long time coming
|
# ? Sep 19, 2021 16:13 |
|
Zedsdeadbaby posted:6c/12t at 35W tdp is not bad at all and a long time coming Haven't those been around forever? 8700T, 10700T was 8C/16T at 35W. These are what's in boring corporate micro form factor desktops. My employer gets them in bulk for under $600 for complete machines, nvme drive and all in the 10600T config. https://ark.intel.com/content/www/us/en/ark/products/129948/intel-core-i7-8700t-processor-12m-cache-up-to-4-00-ghz.html
|
# ? Sep 19, 2021 16:49 |
|
https://videocardz.com/newz/intel-core-i9-12900k-shows-up-on-ashes-of-the-singularity-benchmark-website Looks like someone was goofing with the clocks/timings on their new toy.
|
# ? Sep 21, 2021 22:34 |
|
Someone also posted a 30k+ cinebench with 12900k. So it tops 5950x. We'll see how the 3D cache version from AMD works out.
|
# ? Sep 22, 2021 18:37 |
|
AMD was claiming 15% increased performance in gaming, so my guess is actually around 4-5% in reality. Zen 4 itself is the one that's actually taking on alder-lake, I am not expecting zen 3+ to be competitive with it.
|
# ? Sep 23, 2021 22:51 |
|
Zedsdeadbaby posted:AMD was claiming 15% increased performance in gaming, so my guess is actually around 4-5% in reality. I wouldn't actually be surprised at a full 15%, most games loving love giant caches. iirc Intel's old Broadwell i7-5775-R punched way above it's weight for a long time because it had the 128mb L4 eDRAM on board to support the iris pro iGPU. I think it had dramatically better minimum frame times so things were generally smoother feeling, even if the maximums weren't way higher.
|
# ? Sep 24, 2021 14:49 |
|
Digital Foundry did a comparison of the 5775 and a similar modern chip (same cores and L3 that is): https://www.youtube.com/watch?v=XJaSCnVcCdQ With clocks equalised, there isn't a lot of improvement considering the chips came out 5 years apart. As a bonus, there are also results for where they equalised the number of active cores on better 10th gen Intels, and the extra cache really did help games. The issue I'd see with zen 3+ is the extra layer of cache means around 50% more silicon for only 15% more perf. The higher cost may not be worth it.
|
# ? Sep 24, 2021 23:01 |
|
ConanTheLibrarian posted:Digital Foundry did a comparison of the 5775 and a similar modern chip (same cores and L3 that is): Anandtech's Ian Cuttress seems to imply that the "cache" on Zen3+ appears to play a more important part in terms of topology and forming interconnects from core-to-core, which would imply that the actual "cache" is more a fringe benefit or camouflage than anything else. https://www.youtube.com/watch?v=8teWvMXK99I SwissArmyDruid fucked around with this message at 04:20 on Sep 25, 2021 |
# ? Sep 25, 2021 00:14 |
|
ConanTheLibrarian posted:The issue I'd see with zen 3+ is the extra layer of cache means around 50% more silicon for only 15% more perf. The higher cost may not be worth it. Might they utilize an older process for the cache portion to help save on costs? Unsure if they've said if the extra cache is going to be on-chip or on-die.
|
# ? Sep 25, 2021 00:52 |
|
DrDork posted:Might they utilize an older process for the cache portion to help save on costs? Unsure if they've said if the extra cache is going to be on-chip or on-die. I don't think they use an older process. The extra layer of cache is 64MB compared to the 32MB under it because the masks are totally optimised for SRAM cells. I don't think they'd double the density if they used a bigger node. TSMC's stacking tech can have multiple layers, so you could imagine a solution where they use a stack of say 12nm cache instead of one 7nm layer. But I'd say power/heat becomes an issue with that approach.
|
# ? Sep 25, 2021 12:18 |
|
|
# ? Jun 2, 2024 19:10 |
|
That makes sense, there are already libraries of sram cell designs for different purposes and tradeoffs. If the entire die is for cache, they can even use completely different processes, like what is the case for DRAM. 3D stacking had been around for a while in the pre- competitive research part of the industry. Cooling and power delivery always seemed like the big issues. It seems that AMD are the first to find a way to make it profitable. For some reason Intel wasn't.
|
# ? Sep 25, 2021 22:49 |