|
fishmech posted:Nah, it was really a bad bet on what would be the smart way to handle multiple cores. Microsoft got it right by going with symmetric core design. It continued the fine hardware tradition of "well the software guys can just write code for this right?" and the fine software tradition of "well the compilers will get better, right" and then the fine compiler author tradition of laughing uproariously, some grad students suffering through creating some new compiler techniques that remain untouched / unread for about 5 years. e: Actually I could add in the extra steps of the silicon / architecture guys being "this will fit on the die, right" and "you can remove the heat from the chip, right?"
|
# ¿ Apr 17, 2019 19:38 |
|
|
# ¿ May 2, 2024 14:36 |
|
My last AMD was an X2 ages ago for a low-power build. I think I want a 3900X to toss together a build-server / Linux workhorse at the office now. 12C/24T for $500 is loving nuts. Asus put up a nice article detailing the differences between all their families of motherboards which I appreciated. Looks like they are planning a workstation board on X570 which is awesome. If it comes with a lovely little on-board GPU that would be the tits. I feel dirty cheating after 20 years of running Intel hardware but can't justify the $$$ for performance here.
|
# ¿ Jun 7, 2019 19:31 |
|
apropos man posted:What are the use cases for a system like that? Joke responses are also welcome. Weather is big, CFD, and in the case of US national labs, nuclear weapons stockpile stewardship as we can’t actually blow up real nukes anymore so we have to use modeling to see if they will go boom when we want them too and not degrade unsafely.
|
# ¿ Jun 8, 2019 17:06 |
|
Crunchy Black posted:Yeah its important to note that it's probably going to take a little while for this all to actually shake out. Remember the hell that was early Skylake-S days when Intel made massive changes to their scheduler? Some of this stuff comes down to optimization in the wild. I’m keeping an eye out for kernel patches for Zen 2 (if they haven’t been pushed already) because I’m building a build server / dev box at work with a 3900X. Stocked up on every part now I think except mobo, CPU and case. Not a chance the stock CentOS 7 kernel will be optimized out of the box. DDR4 is so goddamned cheap, it’s ridiculous. I picked up some DDR4-3600 sticks for now. The last AMD rig I built was 10 years ago; how are the stock coolers or should I just get another Thermalright Ultra120 with stupid expensive adapter brackets, or a Hyper212 EVO? Planning on little to mild overclocking.
|
# ¿ Jul 6, 2019 17:17 |
|
Am I a weirdo for not being phased by these prices for mobos and CPUs at all? Granted I had no idea Sandy Bridge was going to last me 10+ years, but this box (which admittedly is for work) I expect to have long long legs for compiling / FPGA builds / general headless crunching. I’m also hoping the WS X570-ACE isn’t super popular so I can snag one. When did mobo makers actually start advertising specific PMICs? I twitched a little when they advertised IR3555s.
|
# ¿ Jul 7, 2019 17:50 |
|
Anyone preordered / queued up at B&H? Newegg has the Asus WS X570 In stock so I snagged that at least.
|
# ¿ Jul 7, 2019 23:46 |
|
DeadFatDuckFat posted:I have no clue about x570 boards, how did you decide on that one? I haven’t touched anything outside Asus for 10+ years, my current board is a P8P67 PRO that I’ve been very happy with. I went with the WS because it’s one of the least obnoxious models (Prime isn’t too bad, the Asus article explaining all their models was a good read), ECC is a nice to have in the future and the basic out-of-band from the Realtek NIC is good to have as well as I’m throwing together this box to be a work mule in an engineering lab. I wish it had more PCIe slots for various I/O adapters and FPGA dev boards but I’ll probably be able to live with it.
|
# ¿ Jul 8, 2019 15:28 |
|
Lambert posted:This isn't about an unfinished BIOS, it's just one of the usual tricks mainboard manufacturers use to make their high-end boards seem better. AMD would have to throw all manufacturers under the bus for this. Yeah this is usual ads-on crap; I’ve always first thing gone in and clicked every single “optimization” off (on Asus boards, that’s also the physical off switches for EPUs and TPUs), and most things that claim to be “automatic”. I nudge up DRAM voltage very slightly when using 4 sticks (think like 25mV-50mV or so on DDR3) and leave most besides the CPU and PLL voltages alone.
|
# ¿ Jul 8, 2019 15:31 |
|
I can’t find any of the Ryzens on Amazon FWIW, maybe the mobile app doesn’t show it?
|
# ¿ Jul 8, 2019 18:57 |
|
So on the Linux side of things...CentOS 7/7.5 a horrible idea on a 3900X? Looks like there's that systemd and/or RdRand thing, but as far as I can tell BIOS / uCode updates have triaged that problem for now. More concerned about kernel 3.10 and a brand new uArch, and don't want to really recompile and run my own. Also kind of considering actual RHEL 8 since CentOS 8 isn't out yet and this is a work expense.
|
# ¿ Aug 1, 2019 18:58 |
|
BangersInMyKnickers posted:There's some errata fixes for Zen2 in the latest kernel patch that came out this week, that should hopefully address the most egregious stuff for the moment. But I would go RHEL 8 if I had the choice. Like mainline mainline kernel, or actually for CentOS 7.x series right now? Never setup RHEL before but would definitely give it a try.
|
# ¿ Aug 1, 2019 22:21 |
|
BangersInMyKnickers posted:3.10 kernel in 7.6 in RHEL, I assume its available on CentOS at this point since its been a few days. Awesome — I tried poking around the CentOS mailing lists, do you have a link to that patch / the mailing list it would be on per chance?
|
# ¿ Aug 1, 2019 23:13 |
|
BangersInMyKnickers posted:Kernel 3.10.0-957.27.2 is the one with some zen2 fixes in it bundled in with CVE stuff. It's all errata fixes so don't expect any documentation for it. Have a support case open with right now dealing with some zen stuff which is how I found out Ahh, got it — thanks for the inside scoop! Excited to see how fast Buildroot turns around Linux builds and Vivado builds FPGA bitstreams with 24 threads.
|
# ¿ Aug 2, 2019 00:45 |
|
Statutory Ape posted:Could I get a gigabit connection and become a video game server host or something Google: someone employ this man at Stadia, stat
|
# ¿ Dec 19, 2019 19:02 |
|
As one of my profs used to say, there isn’t analog and digital, there’s just analog and really really fast analog. The design of high-performance caches and other SRAMs are fascinating. I wonder if anyone was foreseeing the tight integration of power management and basically playing the game around a control loop to keep voltage above BOR levels as a significant area of investment. movax fucked around with this message at 21:04 on Dec 27, 2019 |
# ¿ Dec 27, 2019 21:01 |
|
Combat Pretzel posted:I'm pretty sure he meant the electrical and signalling bullshit involved in implementing SRAM caches. Yeah, this is what I was primarily talking about but the post about temporal and spatial locality being fundamental to a cache doing its job is certainly correct. There are increasing amounts of EDAC (error detection and correction) being found in caches how as transistors shrink; ECC on L2 caches have been offered a la carte to silicon integrators from IP vendors for some time now.
|
# ¿ Dec 28, 2019 16:43 |
|
VorpalFish posted:I mean a 2080ti right now is fine on 8 lanes of pcie3, let alone 16. I don't see pcie4 being useful for gpus for awhile. Better use case is like 10/100gig ethernet or feeding nvme drives with fewer lanes so you can provide more connectivity from the chipset, or maybe some future thunderbolt spec. I think it’s more appealing for the first reasons there, using fewer lanes / physical I/O where possible. I always liked doing bandwidth bridging in designs with PCIe Gen 2 or 3 feeding into a switch that would fan out to slower devices that didn’t need all the bandwidth. Getting 10 GbE in x1 would be great for density.
|
# ¿ Dec 29, 2019 21:11 |
|
I forget which site it was, but they did a giant comparison of pastes / TIMs and included toothpaste and chocolate, among a few others.
|
# ¿ Jan 1, 2020 23:08 |
|
So does Intel just pay off the OEMs again this time to cockblock AMD, or ... ?
|
# ¿ Jan 7, 2020 05:39 |
|
Ok, I’m going crazy, what the gently caress am I doing wrong where all my Boot Options disappear on an ASUS TRX40? All the devices are there (2x NVMe drives), but the BIOS just lists zero options for booting. I think it’s from me flipping the CSM on/off, but never seen this before.
|
# ¿ Feb 21, 2020 18:41 |
|
Klyith posted:Did you also turn on secure boot or something? Don’t think so. BIOS says Secure Boot disabled, set to “Other OS”. Only have two NVMe drives, and I’m installing RHEL 8. With the CSM off, my USB drive doesn’t even appear as an installation option for RHEL 8. There’s no way the TRX40 in TYOOL 2020 needs the CSM to boot via NVMe, does it? Flipping it back on and clearing RTC brought back the options.
|
# ¿ Feb 21, 2020 18:55 |
|
Never mind, it’s me, I can’t read / didn’t read the BIOS warning. Disabled CSM, told it I was installing Windows, rebooted and launched RHEL setup in UEFI mode and I think it’s good now. I left BIOS development right when UEFI was becoming a thing / Secure Boot was starting to rear its head...failure to RTFM. E: nope, never mind, installed off UEFI mode but still no joy being able to boot NVMe. movax fucked around with this message at 19:14 on Feb 21, 2020 |
# ¿ Feb 21, 2020 19:09 |
|
orcane posted:60°C for heat transfer through plastic is pretty good? Don’t forget mayonnaise and toothpaste...
|
# ¿ Mar 8, 2020 11:02 |
|
Evil Robot posted:Fractal Design Define Mini C which has excellent air flow. It's not really low profile; I just have only 400mm of clearance under my desk to place the case so ATX cases don't work. This makes me very curious to see a picture of your desk setup... do you have like a standing desk with an under-slung case mount?
|
# ¿ Mar 23, 2020 18:36 |
|
Evil Robot posted:Nope! Just a shelf above my case. That explains it! I've used the Mini C for workstations at the office supporting engineering testbeds — just cheap, clean and very simple cases with thoughtful design features. I love Fractal.
|
# ¿ Mar 24, 2020 01:59 |
|
Crunchy Black posted:Considering the curve that VR brings I'd hope y'all aren't too broken up about running rigs for...10? years without an upgrade. When has that ever happened like, ever in computing? Not for awhile. Building the 3960X Threadripper rig for myself at work has finally shown me how long in the tooth my 2600K has gotten especially considering my hideous levels of multi-tasking / open browser tabs. Granted some of my issues are likely stemming from a broken Win10 install, but starting to think about upgrading at home. The Intel fanboy in me is still willing to hold out for their (perhaps) return to the top of the crown, but I think I'm doing a HEDT rig next.
|
# ¿ Mar 25, 2020 00:38 |
|
karoshi posted:There was a 'github leak' where some intern doing testing of (possibly) ps5, xBoneSx and navi10 IP made the repository public. But it was just test results. It leaked some clocks and CU counts, but that's surely not the loving source code to the GPUs. I mean, if testbenches or similar verification stuff leaked, that would certainly not be ideal but holy poo poo if they are (really, any company with IP) using loving GitHub to store their poo poo, even on their enterprise plans. Semiconductor IP is valuable and I've heard stories of the security measures guarding their GDS-II / mask sets when it gets to that point. Literally the sauce.
|
# ¿ Mar 25, 2020 19:21 |
|
So... I'm still confused as to what actually got "leaked". Is it actual Verilog/VHDL/RTL for their GPUs, by way of OVM/UVM testbenches? Or a bunch of code that reveals the innards of their drivers and therefore probably internals (register maps/command interfaces/etc.) to the GPU? This isn't AMD, but here is a fascinating read / description of NVIDIA's Falcon: http://download.nvidia.com/open-gpu-doc/Falcon-Security/1/Falcon-Security.html And slides that detail more of it: https://riscv.org/wp-content/uploads/2017/05/Tue1345pm-NVIDIA-Sijstermans.pdf If it shows details of how AMD's equivalent of the deeply embedded controllers work, that could be a bad thing(TM).
|
# ¿ Mar 27, 2020 20:17 |
|
Methylethylaldehyde posted:We really just need more pcie add in boards that let us put 4 or 8 m2 drives in one 4x or 8x slot. Agreed — all it takes a PCIe switch! Also — obviously the forums are on fire to some degree, if you want to post about it, check this out here: https://forums.somethingawful.com/showthread.php?threadid=3929022
|
# ¿ Jun 24, 2020 19:47 |
|
Now we just need the PEX8896 (or w/e they call it) 96 lane PCIe 4.0 switch to bandwidth bridge to PCIe 3.0 NVMe drives because that's still going to be so loving fast.
|
# ¿ Jun 24, 2020 20:03 |
|
Paul MaudDib posted:It's basically an overclockable (presumably?) Epyc. This is the fully enabled part, runs UDIMM/RDIMM/LRDIMM, unlocked memory controller clocks, 128 PCIe lanes, 8 memory channels, the full shebang. Single socket only but who cares. I feel dirty saying it but since I basically amortized the cost of my 2600K over ten years...well I'm ready to jump ship after getting a taste of 3960X at work. I just hope everything will fit into my Meshify C. Fuckload of money but if paradigms continue... should just be a GPU upgrade every 3 years still.
|
# ¿ Jul 11, 2020 19:45 |
|
LRADIKAL posted:Except your server goes down every time you system update, or game instability crashes your server, or games interfere with your transcoder, or if you want your PC to be quiet, but it had tons of spinning rust. I wanted to toss everything into one UBER BOX (not quite gaming + serving, but more like network stuff), but decided against it because I didn't want my network to go down for exactly those reasons.
|
# ¿ Jul 18, 2020 07:20 |
|
sean10mm posted:The important thing before you buy is to read the fine print on the board specs, because there is no hard and fast rule for "Use X slot, Y is disabled" across board makers. They all juggle it differently, unless its X570 in which case go hog wild. Assuming the manuals aren’t complete poo poo, the block diagram in there is the best way to check, followed by having to actually read the manual to confirm. Don’t trust the website specs if you want to be 100% sure, especially with mobos with a ton of SKUs — marketing people probably copy pasted and may have missed something.
|
# ¿ Jul 20, 2020 22:04 |
|
gradenko_2000 posted:oh my god lmao I feel like I’m reading a post from an audiophile forum...
|
# ¿ Aug 16, 2020 07:09 |
|
NewFatMike posted:Ha, Arctic Silver? Good luck with your CPU cooler creating micro cavities in the thermal compound in the 285kHz range. You're really limiting your 0.001% lows. Brb, inventing computophile grade motherboard standoffs to isolate harmful mains noise from the motherboard and brighten the bits whilst removing jitter and noise from RAM slots, therefor solving RAM compatibility’s problems.
|
# ¿ Aug 16, 2020 22:13 |
|
From Hot Chips, AnandTech covering some of the APU stuff coming out in more detail now. X-post GPU thread for obvious reasons. Interesting situation with the new consoles being more or less contemporaneous with the state-of-the-art for PC gaming — Zen2 cores, plus fixed-function accelerators for audio, storage and other functions. Wondering how games will map to PC levels of performance offered; would a current-gen Threadripper with its extra cores make up for the HW accelerated audio processing and storage decompression?
|
# ¿ Aug 18, 2020 03:35 |
|
I'm curious, what multiple game scenarios are these? Some MMO type stuff?
|
# ¿ Aug 21, 2020 21:55 |
|
ConanTheLibrarian posted:https://twitter.com/chiakokhua/status/1298433876058988544 I regret not buying into more TSMC when it was down in the $30s.
|
# ¿ Aug 27, 2020 02:24 |
|
Cygni posted:Nobody knows for sure, but AMDs CEO has confirmed the desktop parts will launch before the end of the year. Probably a good bet that they launch after the server versions, and probably a good bet the server versions launch in September so... soon™? I'm still stupidly holding out hope for Zen 3 Threadrippers this year, but realistically it's got to be Q1/Q2 next year.
|
# ¿ Aug 28, 2020 00:18 |
|
|
# ¿ May 2, 2024 14:36 |
|
punk rebel ecks posted:I see. Should it be something to worry about if I'm trying to "future proof" a PC build for next-gen a year from now? Some things might load a few seconds slower / you might miss out on some Konami Kojima gimmick. This is just “weird” because this generation, consoles are contemporaneous with PCs (CPU, GPU, RAM, storage, all are not only the same as PCs, but they’re even likely the same exact IP blocks) and have the advantage of some custom hardware. I think the optimized NVMe thing is actually pretty cool, and an interesting solution to their constraints. PCs can spend more $$$ / thermal budget / whatever to achieve similar goals.
|
# ¿ Aug 28, 2020 06:55 |