|
PC LOAD LETTER posted:Apparently requires lots of luck or really good binned parts and stability is still touch and go most of the time. Epycs only allow one DIMM per channel and the number of people who need 4 DIMMs is a drop in the ocean so AMD don't have a lot of motivation to make a memory controller that can handle 4 DIMMs well.
|
#
?
Aug 10, 2023 14:02
|
|
|
|
| # ? May 30, 2024 12:12 |
|
|
ConanTheLibrarian posted:Epycs only allow one DIMM per channel and the number of people who need 4 DIMMs is a drop in the ocean so AMD don't have a lot of motivation to make a memory controller that can handle 4 DIMMs well. All Epycs have been designed to support 2 DIMMs per channel, and AMD says that they're able to fix their latest generations memory problem with just a BIOS update. https://www.tomshardware.com/news/amd-responds-to-claims-of-epyc-genoa-memory-bug-says-update-on-track 2DPC gets Epyc to 48 DIMMs in a 2 socket server.
|
|
#
?
Aug 10, 2023 14:55
|
|
|
Good luck fitting that number on a server motherboard without even more physical compromises than 12 DIMMs/socket takes. That article even says they think it's more likely to be used in a single socket setup. Not to mention that memory speed drops from 4800 to 4000 when going to 2DPC on Genoa. To address my original point, the Tom's article also says "market insiders have even predicted that support for 2DPC could end with the DDR6 standard". Gigabyte showcased a board with 2DPC, but I'm assuming it was a prototype because it's not listed on their site 9 months after the photo was taken:
|
|
#
?
Aug 11, 2023 11:09
|
|
|
When are we finally going to do proper quad-channel with 1dpc for consumer chips? What exactly is preventing that from being a thing? Is it just a segmentation thing, to make HEDT more appealing?
|
|
#
?
Aug 11, 2023 11:22
|
|
|
Pins and traces.
|
|
#
?
Aug 11, 2023 11:39
|
|
|
For the moment there's still a lot of scope to boost memory bandwidth by increasing memory clocks. Once those gains are exhausted, we'll be even deeper into the mismatch between logic and PHY scaling. Adding an extra two channels may be prohibitively expensive since (a) the CPU manufacturer has to produce a larger die, (b) the mobo manufacturer needs a more complicated MB, and (c) the consumer needs to buy 4 DIMMs to get the most out of their system. Therefore,,,ConanTheLibrarian posted:Stack everything. A compute die on top of a cache die on top of a PHY die on top of a HBM die. Let the 3D revolution begin!
|
|
#
?
Aug 11, 2023 12:21
|
|
|
LGA makes it easy to add tons of contacts though. They're already at over 1700 for AM5. LGA2011-v3 and LGA2066 have a bit over 2000 and were 4 channel memory platforms. There are mobos that support those sockets brand new that sell for around $160 these days so its not THAT expensive or onerous to do. Especially vs what new AM5 mobos can cost. They're probably worried about appeasing OEM's would be my guess. OEM's are already so drat cheap they still often ship modern systems with a single DIMM of the lowest clocked stuff they can get to save a buck or 2. They'd probably be pissed if they were forced to buy 4 DIMMs every time for AMD systems. That and until recently I don't think most cared about iGPU performance which is where that extra bandwidth would really matter. FWIW technically DDR5 is already quad channel buuuut its 4x 32bit channels instead of 4x 64bit channels like with LGA2011-v3 and LGA2066 so we'd reaaaaallllly wanting here are 8x 32bit channel memory systems!
|
|
#
?
Aug 11, 2023 12:38
|
|
|
Dr. Video Games 0031 posted:When are we finally going to do proper quad-channel with 1dpc for consumer chips? What exactly is preventing that from being a thing? Is it just a segmentation thing, to make HEDT more appealing? Cost. The requirement for the signal path are really strict these days, which means that routing more lines makes every existing line more expensive. I think there is very little chance that the same platform that needs to support normal, low-end desktop will also ever support 4 channels with dimms. It's a bit more possible if Dell CAMM becomes a thing even on desktop, as those were designed to make routing cheaper. But even with them, I think the odds are pretty remote. Honestly, I think consumer CPU memory buses will only grow wider once memory is soldered on the same substrate as the CPU. Because that's the only way to do it cheaply. PC LOAD LETTER posted:LGA makes it easy to add tons of contacts though. The socket is a lot less of a limit these days than what's under the socket.
|
|
#
?
Aug 11, 2023 13:24
|
|
|
The other question to ask is: how many things are actually bandwidth limited by 2 channels of memory on a normal desktop CPU? You might be able to say something like "bumping from DDR4 4800 to 6000 has x% average increase in FPS across a bunch of games", but that speed boost is also decreasing latency. Meanwhile if you take a Threadripper with 8 goddamn memory channels and look at reviews with gaming performance, it's nothing special. Games aren't the only thing in the world, but I suspect that relatively few desktop tasks are being limited by memory bandwidth, particularly on the midrange chips like a 7800X or 13600K. Quite possibly a 7950X with 16 cores would see some improvement on some sort of serious-business task like rendering or whatever, but that's why HEDT / workstation exists.
|
|
#
?
Aug 11, 2023 15:58
|
|
|
Did memory training get slightly longer with ComboPI 1.0.0.7b?
|
|
#
?
Aug 11, 2023 17:27
|
|
|
Klyith posted:The other question to ask is: how many things are actually bandwidth limited by 2 channels of memory on a normal desktop CPU? Tbh where it really matters for consumers is igpu - where memory bandwidth is shared, and is a real constraint for how performant your GPU can be. But at least for now that's way less relevant on desktop where everyone who wants GPU performance is throwing in a pcie card anyways. There's been a lot of talk about igpu cannibalizing the low to midrange GPU market though, and that's going to need bandwidth if it's going to happen.
|
|
#
?
Aug 11, 2023 23:04
|
|
|
Tuna-Fish posted:The socket is a lot less of a limit these days than what's under the socket. Introducing the world's first 96-layer motherboard PCB, only from ASRock!
|
|
#
?
Aug 11, 2023 23:10
|
|
|
Kazinsal posted:Introducing the world's first 96-layer motherboard PCB, only from ASRock! Pffft, boring engineering safety margins to prevent problems? You must be thinking of asus or evga! The Asrock mobo with quad-channel DDR5 would still have an 8 layer PCB, but the ram slots are arranged in a rhombus around the CPU to avoid crosstalk. One of them is on the back of the mobo because it's colocated with the GPU PCIe slot. And every other USB jack on the IO pane is rotated 90 degrees because they found it cancels a small but significant EFI source. Also every setting in the BIOS has to be set to a multiple of seven, so that it will never operate at the same frequency as the DDR. And they always take 2 extra weeks to ship because they always get held up in transit when the Postal Inspector drug-detection dogs keep alerting on them, for some inexplicable reason.
|
|
#
?
Aug 12, 2023 00:49
|
|
|
VorpalFish posted:Tbh where it really matters for consumers is igpu - where memory bandwidth is shared, and is a real constraint for how performant your GPU can be. Yeah for consumers it might matter the most for igpu, but aside from Macs the computers that use igpus are for the most budget conscious buyers, where increasing the motherboard cost by $50 might be a real problem.
|
|
#
?
Aug 12, 2023 03:41
|
|
|
Klyith posted:The Asrock mobo with quad-channel DDR5 would still have an 8 layer PCB, but the ram slots are arranged in a rhombus around the CPU to avoid crosstalk. One of them is on the back of the mobo because it's colocated with the GPU PCIe slot. And every other USB jack on the IO pane is rotated 90 degrees because they found it cancels a small but significant EFI source. Also every setting in the BIOS has to be set to a multiple of seven, so that it will never operate at the same frequency as the DDR. 
|
|
#
?
Aug 12, 2023 03:53
|
|
|
Klyith confirmed working at AsRock.
|
|
#
?
Aug 12, 2023 10:52
|
|
|
I upgraded my ASUS ROG B650E-F to the newish 1636 BIOS that uses AGESA 1.0.0.7b and they changed the ROG logo on bootup to red lmao. I was able to run my G.Skill Flare 2x 16GB 6000MHz 36-36-36-96 memory at 6400MHz 30-36-36-76 but the 3DMark score went up less than 2% and the mobo automatically jacked up my CPU voltages so I pulled it back to stock EXPO.
|
|
#
?
Aug 16, 2023 07:29
|
|
|
new AMD chipset drivers for AM4/AM5 posted https://www.amd.com/en/support/kb/release-notes/rn-ryzen-chipset-5-08-02-027
|
|
#
?
Aug 20, 2023 20:32
|
|
|
Those are some real lovely release notes.
|
|
#
?
Aug 20, 2023 20:39
|
|
|
Subjunctive posted:Those are some real lovely release notes. Agreed. I have three machines of my own this applies to, as well as two more owned by family members I end up as default tech support for, and I don't feel like I have any idea what it is actually doing.
|
|
#
?
Aug 20, 2023 21:53
|
|
|
Well the intern had to take a break from coding FSR3 to write those, so cut them some slack!
|
|
#
?
Aug 20, 2023 21:56
|
|
|
Not really a parts question, but im putting together a new pc for my gf. I got two codes for starfield from the parts i bought her. Is there a way to use a code for me and her since she wont be using both? E: its through amd rewards
|
|
#
?
Aug 22, 2023 03:19
|
|
|
ZombieCrew posted:Not really a parts question, but im putting together a new pc for my gf. I got two codes for starfield from the parts i bought her. Is there a way to use a code for me and her since she wont be using both? It'll depend on if it generates codes or redeems it to a steam/epic/whatever account. If it's redeemed via Steam then a Steam account has to be linked to the AMD Rewards site, and it'll add it to the account. I think you can probably add a Steam account, redeem one code, unlink that account and link another, and redeem the other code.
|
|
#
?
Aug 22, 2023 03:24
|
|
|
Branch Nvidian posted:It'll depend on if it generates codes or redeems it to a steam/epic/whatever account. If it's redeemed via Steam then a Steam account has to be linked to the AMD Rewards site, and it'll add it to the account. I think you can probably add a Steam account, redeem one code, unlink that account and link another, and redeem the other code. But id have to do it on the machine im building so it sees the parts?
|
|
#
?
Aug 22, 2023 03:27
|
|
|
Yes, that is correct. You enter the redemption code for one of the parts and you get a verification tool that looks for the specific component, once it sees that component it allows you to add the game to your Steam library or gives you the game code or w/e.
|
|
#
?
Aug 22, 2023 03:58
|
|
|
Is it confirmed to be a Steam copy?
|
|
#
?
Aug 22, 2023 04:14
|
|
|
Rinkles posted:Is it confirmed to be a Steam copy? According to this Reddit post it is, indeed, a Steam copy. https://www.reddit.com/r/Starfield/comments/14z5eml/confirmed_that_amd_bundle_gives_a_steam_key/
|
|
#
?
Aug 22, 2023 04:18
|
|
|
Klyith posted:The other question to ask is: how many things are actually bandwidth limited by 2 channels of memory on a normal desktop CPU? ... Yep. I honestly think this was part of the calculus for keeping quad-cores for so long too. that's what the business world wants as a "decent but not extravagant workstation" (not HEDT) in the 2012-2017 era. the drive for higher cores wasn't there, and if you were a prosumer the HEDT platform was a lot cheaper and more accessible.
|
|
#
?
Aug 22, 2023 04:56
|
|
|
no breaking news per se, but check out that on-board fan, good lord https://twitter.com/VideoCardz/status/1698675506478981164 
|
|
#
?
Sep 4, 2023 14:23
|
|
|
kliras posted:no breaking news per se, but check out that on-board fan, good lord which one? each board has two, one over the chipset (or where the chipset used to be) and one on the VRMs.
|
|
#
?
Sep 4, 2023 14:38
|
|
|
https://twitter.com/Kepler_L2/status/1698531961294205308 I don't think anyone is actually boycotting AMD like this (lol), but the rumor mills are very bullish on Zen 5 lately
|
|
#
?
Sep 4, 2023 15:02
|
|
|
gradenko_2000 posted:https://twitter.com/Kepler_L2/status/1698531961294205308
|
|
#
?
Sep 4, 2023 15:51
|
|
|
The worst part about waiting for Zen 5 is you have to wait another half year for the X3D parts
|
|
#
?
Sep 4, 2023 17:43
|
|
|
Icept posted:The worst part about waiting for Zen 5 is you have to wait another half year for the X3D parts The internet thinks Zen 5 (AKA Ryzen 8000) has a street date of about a year from now (H2 2024), so AMD will probably unveil it at Computex -- next May? I don't even have the energy to get hyped about that, no matter how badly twitter leakers want me to. Edit: embarrassing apostrophe error mdxi fucked around with this message at 19:43 on Sep 4, 2023 |
|
#
?
Sep 4, 2023 19:04
|
|
|
I'm eagerly waiting for Zen5 to upgrade my 3700X and avoid the first gen issues DDR5 and new chipsets have.
|
|
#
?
Sep 4, 2023 19:10
|
|
|
gradenko_2000 posted:https://twitter.com/Kepler_L2/status/1698531961294205308 There’s probably one person out there doing this and their reasoning would have to be “AMD is so bad at this. Now, Intel, there’s a company that actually knows how to gently caress over other hardware vendors”
|
|
#
?
Sep 4, 2023 19:22
|
|
|
Sininu posted:I'm eagerly waiting for Zen5 to upgrade my 3700X and avoid the first gen issues DDR5 and new chipsets have. Rumor is that its the exact same IO die as Zen4 without any silicon revision, so might not be much different for memory stuff. Still a ways off from knowing for sure, and I think your sentiment is smart regardless.
|
|
#
?
Sep 4, 2023 19:26
|
|
|
The current AGESA versions seem to be making good use of the IOD, at least to the capabilities of the zen4 CPUs. We’ll have to see if this is more like Zen1 to 2 or 2 to 3.
|
|
#
?
Sep 4, 2023 19:28
|
|
|
Seems the most recent AGESA has tidied up a lot of the memory bullshit?
|
|
#
?
Sep 4, 2023 19:35
|
|
|
|
| # ? May 30, 2024 12:12 |
|
|
Sininu posted:I'm eagerly waiting for Zen5 to upgrade my 3700X and avoid the first gen issues DDR5 and new chipsets have. Same, but would be upgrading from an Intel 9900k assuming I can wait that long.
|
|
#
?
Sep 4, 2023 19:36
|
|