|
Intel's moving to "chiplets" (tiles) in laptops sooner than they are for desktop, so there's presumably some benefit there if you do it right.
|
#
?
Apr 29, 2023 16:00
|
|
|
|
| # ? Jun 3, 2024 10:09 |
|
|
Dr. Video Games 0031 posted:Intel's moving to "chiplets" (tiles) in laptops sooner than they are for desktop, so there's presumably some benefit there if you do it right. They put 2 E-cores on the IO tile so they can turn off the lights on the CPU and GPU tiles during 99.999% of the time the system is waiting for user input while displaying the same image. (Display out is on the IO tile)
|
|
#
?
Apr 29, 2023 16:17
|
|
|
Something I've been wondering is why AMD puts so much physical distance between their chiplets. Intel putting all the dies in one homogenous cluster of tiles seems to make a lot more sense than how AMD's chiplet CPUs are packaged.
|
|
#
?
Apr 29, 2023 16:37
|
|
|
Dr. Video Games 0031 posted:Something I've been wondering is why AMD puts so much physical distance between their chiplets. Intel putting all the dies in one homogenous cluster of tiles seems to make a lot more sense than how AMD's chiplet CPUs are packaged.
|
|
#
?
Apr 29, 2023 16:40
|
|
|
Dr. Video Games 0031 posted:Something I've been wondering is why AMD puts so much physical distance between their chiplets. Intel putting all the dies in one homogenous cluster of tiles seems to make a lot more sense than how AMD's chiplet CPUs are packaged. Helps with cooling
|
|
#
?
Apr 29, 2023 16:48
|
|
|
Dr. Video Games 0031 posted:Something I've been wondering is why AMD puts so much physical distance between their chiplets. Intel putting all the dies in one homogenous cluster of tiles seems to make a lot more sense than how AMD's chiplet CPUs are packaged. I think spreading things out is better, all else equal, for thermal purposes. Ryzen chiplets are connected through a non-silicon interposer, which is cheaper. IIRC back when Zen 2 was announced one of the points of discussion was that it was impressive that AMD had made the chiplet thing work with standard organic substrate. Intel's tile stuff is silicon-on-silicon. For that every square mm is money. So it's gotta be tight so your bottom layer isn't huge. But since Intel is using silicon to connect the tiles, they're gonna throw some extra memory on it. Klyith fucked around with this message at 17:11 on Apr 29, 2023 |
|
#
?
Apr 29, 2023 17:05
|
|
|
karoshi posted:They put 2 E-cores on the IO tile so they can turn off the lights on the CPU and GPU tiles during 99.999% of the time the system is waiting for user input while displaying the same image. (Display out is on the IO tile)
|
|
#
?
Apr 29, 2023 17:11
|
|
|
e: gonna post this in the gpu thread, despite the cpu bottleneck
|
|
#
?
Apr 29, 2023 18:42
|
|
|
Assembled the new PC today. Decision time. ASUS only has a (day-old) beta BIOS available for the B650 mobo. This being a beta, they are denying all liability from using it. Which means the alternative option is using whatever bios the mobo came with in the box. Since it's "stock" they can't deny liability for it potentially bricking my CPU and itself (though they may try to push the blame onto someone else). Doesn't seem to be a good option in there. I'll give it a couple of days and if there's no progress I'll try out option 2 because I'm not taking liability for their poo poo work. Can't sit around forever because there's a limited "expedited return" period on the rest of the hardware and I'd like to test it out before then. I bought 4800MHz DDR5 because I don't care about 1.605% of extra performance, so if I've been reading this thread right, the risks might be somewhat reduced compared to higher frequency RAM?
|
|
#
?
Apr 29, 2023 19:39
|
|
|
If you aren’t overclocking you’re fine. DDR5-4800 is a JEDEC speed and the motherboard shouldn’t drive high VSOC.
|
|
#
?
Apr 29, 2023 20:05
|
|
|
BlankSystemDaemon posted:I have no idea, but I would be utterly unsurprising to learn that Intel has some kind of patent. Thermals. Intel and AMD have a patent share agreement that stem from x86-64 licensing.
|
|
#
?
Apr 30, 2023 01:30
|
|
|
SwissArmyDruid posted:Thermals. Intel and AMD have a patent share agreement that stem from x86-64 licensing. It's a super fun agreement from way back in the day: AMD got a license from Intel to manufacture x86 CPUs to help them meet demand. This license was apparently quite open ended because 20 or so years later Intel tried to force Itanium on everyone and AMD lol-noped and built extensions onto x86 giving us x86-64. As it stands now Intel could revoke x86 from AMD but at the same time AMD can revoke -x64 from Intel. They're both holding loaded shotguns to each others' heads.
|
|
#
?
Apr 30, 2023 02:42
|
|
|
Fabulousity posted:It's a super fun agreement from way back in the day: AMD got a license from Intel to manufacture x86 CPUs to help them meet demand. This license was apparently quite open ended because 20 or so years later Intel tried to force Itanium on everyone and AMD lol-noped and built extensions onto x86 giving us x86-64. As it stands now Intel could revoke x86 from AMD but at the same time AMD can revoke -x64 from Intel. They're both holding loaded shotguns to each others' heads. Shouldn't those patents be close to expiring? Or is it just incremental patents on each little addition that add up
|
|
#
?
Apr 30, 2023 03:43
|
|
|
Shipon posted:Shouldn't those patents be close to expiring? Or is it just incremental patents on each little addition that add up Important portions could also be copyrightable.
|
|
#
?
Apr 30, 2023 03:46
|
|
|
Gamer's Nexus video on the 7800X3D https://www.youtube.com/watch?v=kiTngvvD5dI
|
|
#
?
Apr 30, 2023 05:53
|
|
|
grack posted:Gamer's Nexus video on the 7800X3D is there a tl:dw version
|
|
#
?
Apr 30, 2023 06:02
|
|
|
Palladium posted:is there a tl:dw version AMD bad, Gigabyte stupid, Asus bad and stupid
|
|
#
?
Apr 30, 2023 06:04
|
|
|
Palladium posted:is there a tl:dw version 1) AMD had poor communication with the mobo OEMs and needs to do better validation 2) ASUS was a standout for major issues since they were really cranking the volts up while not setting other voltages properly and just generally didn't handle the whole situation well, also ther voltage protection scheme is screwed up and can cause other failures 3) Other mobo vendors weren't nearly as bad (vSOC was generally raised higher than 1.2v but less than 1.3v) but also weren't good. Gigabyte had a bug where BIOS settings were getting ignored for instance 4) New BIOS'es mostly address the issue but they're still not totally perfect, absolutely install them since they help 5) GN had to try and force the issue to cause the CPU's to get killed immediately. Its not something that is a major worry for most people 6) Excess volts can still drastically shorten the lifespan of the chip even if things seem 'fine' 7) Checking your voltage settings with hwinfo isn't a bad idea just in case 8) More vids coming since other bugs were found but were unrelated to the main topic of the video. 9) AMD is willing to immediately replace and pay for shipping on effected CPU's even if only EXPO was turned on, mobo vendors might be a different story Its a bit long of a vid for some people I guess but its a good vid. Key parts are split up and labelled properly so you can easily click to where you need. PC LOAD LETTER fucked around with this message at 06:44 on Apr 30, 2023 |
|
#
?
Apr 30, 2023 06:39
|
|
|
Updated to the 1202 bios with my 7800x3d on an ASUS X670E-E Gaming Wifi, just tested with EXPO on and off and saw VSOC at 1.394 and VDDIO/MC at 1.447 V, versus stock settings of 1.057V and 1.137 V respectively. This was supposed to be the "fixed" bios as of a few days ago but apparently not lol
|
|
#
?
Apr 30, 2023 06:45
|
|
|
Shipon posted:Updated to the 1202 bios It looks like thats from the 21st. They have 2 newer BIOS'es on their site now with the latest being 1401 with a issue date of yesterday. Give the latest a shot and see if it helps. If it doesn't stick wtih defaults only for now or manually set your volts.
|
|
#
?
Apr 30, 2023 06:58
|
|
|
PC LOAD LETTER posted:mobo vendors might be a different story this is gonna be great
|
|
#
?
Apr 30, 2023 07:27
|
|
|
I especially liked the shout out to that one Biostar board with volts even higher than Asus
|
|
#
?
Apr 30, 2023 07:33
|
|
|
gradenko_2000 posted:I especially liked the shout out to that one Biostar board with volts even higher than Asus i think they went don't ask, don't tell on 1.4V+ VSOC in those offices
|
|
#
?
Apr 30, 2023 08:49
|
|
|
Shipon posted:Updated to the 1202 bios with my 7800x3d on an ASUS X670E-E Gaming Wifi, just tested with EXPO on and off and saw VSOC at 1.394 and VDDIO/MC at 1.447 V, versus stock settings of 1.057V and 1.137 V respectively. This was supposed to be the "fixed" bios as of a few days ago but apparently not lol Disable EXPO immediately and update to a newer BIOS. edit: GN confirmed that 1202 exhibits the behavior that can lead to this failure, so yeah, don't use it. Dr. Video Games 0031 fucked around with this message at 13:53 on Apr 30, 2023 |
|
#
?
Apr 30, 2023 10:34
|
|
|
Gigabyte's recent bioses are bugged. GN reported that in the newer versions, any manually input VSOC voltage isn't overwritten when loading defaults or setting it to auto, and I can confirm this was happening to me too. I also confirmed the "actual VSOC voltage is 50mV higher than input VSOC voltage" issue was happening to me. After some tinkering, i input the soc voltage on the tweaker tab to 1.15 and I found a separate VSOC setting in the AMD overclocking menu and set that to 1.2V, and now both the VDDCR_SOC and VCORE SOC readings are at 1.2V on the dot in hwinfo. That seems safe, so I'll just leave it at this from now on.
|
|
#
?
Apr 30, 2023 12:14
|
|
|
Fuckin' tech jesus, holy crap. I expected a good video when he was pimping it on twitter, but that is something else. grack posted:AMD bad, Gigabyte stupid, Asus bad and stupid MSI somehow neither bad nor stupid, what?????
|
|
#
?
Apr 30, 2023 14:51
|
|
|
Palladium posted:is there a tl:dw version Its far worse than I thought jfc.
|
|
#
?
Apr 30, 2023 15:51
|
|
|
i like how people in there gets surprised by "how could this also happen to my $700 AM5 halo-tier mobo?", as if  weren't be happy picking enormous profit margins over delivering actual QC
|
|
#
?
Apr 30, 2023 15:53
|
|
|
Palladium posted:i like how people in there gets surprised by "how could this also happen to my $700 AM5 halo-tier mobo?", as if I think it was tech jesus that said that most of the consumer mobo manufacturers have one developer assigned to UEFI. So, despite whatever you're paying on the motherboard they're just packaging whatever phoenix technologies/whoever and AGESA that AMD gives them because thats all they have resources to do.
|
|
#
?
Apr 30, 2023 15:58
|
|
|
Correct me if I'm wrong, but wasn't there a whole lot of complaining when the 5800X3D launched and the voltages were locked down? Did they walk back on that for the 7800X3D or did the board partners just juice it the gently caress up behind the locks?
|
|
#
?
Apr 30, 2023 16:16
|
|
|
Icept posted:Correct me if I'm wrong, but wasn't there a whole lot of complaining when the 5800X3D launched and the voltages were locked down? Did they walk back on that for the 7800X3D or did the board partners just juice it the gently caress up behind the locks? VSoC was never restricted - mine is sitting at 1.16V vs default of I think 1.05V (1.00?) right now. VCore is and you couldn't even set negative offsets in UEFI, though you can from windows.
|
|
#
?
Apr 30, 2023 16:24
|
|
|
Icept posted:Correct me if I'm wrong, but wasn't there a whole lot of complaining when the 5800X3D launched and the voltages were locked down? Did they walk back on that for the 7800X3D or did the board partners just juice it the gently caress up behind the locks? Yeah there was. There were ways to get around it to and extent if you were determined but it wasn't easy and was still limiting. For the 7800x3D AMD limited the multiplier and some of the volts (core volts but not SOC for instance) but it seems like the mobo guys were setting the volts much higher if certain XMP or EXPO settings were used. AMD's screw up was allowing them to do that crap in the first place which is where GN's comment about AMD needing better communication and validation comes in. The motherboard guys were pulling similar shenanigans years ago as someone else noted. They didn't get any chips to blow up that time though and Intel put a stop to it before it turned into a real mess. https://www.youtube.com/watch?v=qQ_AETO7Fn4
|
|
#
?
Apr 30, 2023 16:27
|
|
|
https://twitter.com/harukaze5719/status/1652659459024654336
|
|
#
?
Apr 30, 2023 16:31
|
|
|
Klyith posted:Fuckin' tech jesus, holy crap. This definitely happens on MSI boards as well, Asus and Gigabyte are just the two most focused on because Steve bought already exploded CPU/Motherboard combos from both of those manufacturers. Literally every AM5 board will pump vSOC to high heavens if you turn on a >=6000MHz EXPO profile. There had to have been a failure of communication, or a straight mistake, from AMD for all the partners to make the same mistake. I don't like that the GN video focuses really hard on the Asus failure, because it makes it seem like it is isolated to Asus. Steve's implication that while other motherboards will kill the CPU, only Asus boards will continue to pump power into them until they melt, just isn't true because there are CPUs that have melted down in this way in boards from every major manufacturer. The first melted CPU was the one that De8aeur did a video on a month ago, and that one had the solder drip out from under the heatspreader running in an X670E Aorus Master. The AGESA changelog leaked by Igor, posted a page ago, lists PROCHOT as being non-functional in prior AGESA revisions. In the GN video at 13:30 he says that PROCHOT should be triggering at this point to prevent the meltdown, so if that's really the cause then it's not an ASUS problem but an AGESA one. Hopefully the followup video from GN is a bit more directed, he mentioned their lab found a whole host of broken features and protections in AM5 that they will cover separately. E: in the time it took me to slowly phone-type this I got semi-beaten by a helpful post from DR.VG about AGESA and PROCHOT
|
|
#
?
Apr 30, 2023 16:34
|
|
|
Congrats to Asus for making overcurrent protection that doesn't work at all Special mention to Gigabyte for making a uEFI UI that lies about voltages Thank you AMD for letting motherboard makers cook your processors If you were wondering who is at fault for this, the answer is everyone, this is a collaborative failure
|
|
#
?
Apr 30, 2023 16:42
|
|
|
PC LOAD LETTER posted:The motherboard guys were pulling similar shenanigans years ago as someone else noted. They didn't get any chips to blow up that time though and Intel put a stop to it before it turned into a real mess. my old MSI z370m autopushed a somewhat insane 1.3V into the VCCSA with XMP on for a mere 3200C16 i don't want to know how much higher that voltage will get if the RAM was set faster, lol
|
|
#
?
Apr 30, 2023 16:48
|
|
|
BurritoJustice posted:Literally every AM5 board will pump vSOC to high heavens if you turn on a >=6000MHz EXPO profile. Uh, that's a very strong claim that I've not seen from any professional source. Especially seeing as how the L1T collab in the GN video showed boards not doing that (and the MSI board staying at 1.2V flat). Palladium posted:i like how people in there gets surprised by "how could this also happen to my $700 AM5 halo-tier mobo?", as if I kinda wonder if some aspects -- like asus having functionally zero OCP to prevent throwing 35 amps of current into the SOC rail -- might actually be because these are $700 halo-tier mobos. These things are advertised partially on the back of extreme overclocking nonsense. That's the image they're selling, even if 95% of the people buying the boards will never touch a thermos of LN2 in their life. Supporting real extreme overclocking means having the ability to push volts and amps that will destroy a CPU under normal conditions. And then the mobo makers say "derbaurer used this mobo to set a record OC, you want to be like derbaurer right?" (And yeah they should just add a special "safeties off" mode for the extreme OCers, but see the bit where they have very small numbers of people working on the bios & systems software.)
|
|
#
?
Apr 30, 2023 16:54
|
|
|
i remember raising an eyebrow at some of the xmp behaviors iirc it did boost cpu voltage on my asus z390e + 8700k, but I just shrugged and left it on and it survived like 5 years. that said like a year ago or so it stopped being able to boot at all if I turned on XMP it would just boot loop after I got out of post so I had to manually set the timings and the speed of the ram and leave xmp off and it was good until i finally retired it.
|
|
#
?
Apr 30, 2023 16:55
|
|
|
Klyith posted:Uh, that's a very strong claim that I've not seen from any professional source. Especially seeing as how the L1T collab in the GN video showed boards not doing that (and the MSI board staying at 1.2V flat). I have an X670E Tomahawk, which is a newly released board so it doesn't have much coverage and before applying the latest beta BIOS I was seeing 1.35v in HWiNFO. Now it's at 1.3v.
|
|
#
?
Apr 30, 2023 17:03
|
|
|
|
| # ? Jun 3, 2024 10:09 |
|
|
I've never been so happy to have an Intel that only runs at 100c. hahaha
|
|
#
?
Apr 30, 2023 17:05
|
|