|
Kibner posted:what in the hell? lmao
|
#
?
Jun 1, 2023 19:20
|
|
|
|
| # ? May 29, 2024 15:24 |
|
|
https://www.amazon.com/gp/product/B09VCJ2SHD 5800X3D down to $290, sold+shipped by Amazon. $280 at Microcenter w/ $20 mobo credit.
|
|
#
?
Jun 2, 2023 04:22
|
|
|
props to that server for never having to be rebooted for almost 3 years
|
|
#
?
Jun 2, 2023 08:51
|
|
|
multiple servers, since the bug was reproduced
|
|
#
?
Jun 3, 2023 01:35
|
|
|
Is that a clock domain crossing / metastability thing where the MTBF hits around that mark? Kudos to the test lab / plan that caught and/or observed that + root caused.
|
|
#
?
Jun 3, 2023 03:37
|
|
|
movax posted:Is that a clock domain crossing / metastability thing where the MTBF hits around that mark? Kudos to the test lab / plan that caught and/or observed that + root caused. 1044 days times 10e11 is close to 0x7FFF_FFFF_FFFF_FFFF
|
|
#
?
Jun 3, 2023 04:41
|
|
|
karoshi posted:1044 days times 10e11 is close to 0x7FFF_FFFF_FFFF_FFFF Oh goddammit this poo poo again. 
|
|
#
?
Jun 3, 2023 05:38
|
|
|
Kibner posted:what in the hell? lmao CC6 is the deep sleep (power gated) state, so this is likely something along the lines of the core failing to power back up if it's in deep sleep after the time when some continuously incrementing counter in continuously powered uncore support circuitry rolls over. Some Redditor figured out that the time is roughly 0x380000000000000 ticks of the chip's 2800 MHz TSC, which they interpret as significant thanks to all those zeroes. I'm less sure about this numerology, because 0x38 doesn't tingle my RTL designer spidey senses, but sometimes poo poo is weird and it's someone else's design I have no insight into.
|
|
#
?
Jun 3, 2023 09:49
|
|
|
BobHoward posted:Some Redditor figured out that the time is roughly 0x380000000000000 ticks of the chip's 2800 MHz TSC, which they interpret as significant thanks to all those zeroes. I'm less sure about this numerology, because 0x38 doesn't tingle my RTL designer spidey senses, but sometimes poo poo is weird and it's someone else's design I have no insight into. A redditor deciding the 1044 days is wrong and it's actually 1042 days and titling their post that way because of bullshit numerology they pulled out of their rear end because it has a pleasing number of zeros is hilarious
|
|
#
?
Jun 3, 2023 20:44
|
|
|
crazypenguin posted:A redditor deciding the 1044 days is wrong and it's actually 1042 days and titling their post that way because of bullshit numerology they pulled out of their rear end because it has a pleasing number of zeros is hilarious Yeah, they are just guessing, after all
|
|
#
?
Jun 3, 2023 21:11
|
|
|
crazypenguin posted:A redditor deciding the 1044 days is wrong and it's actually 1042 days and titling their post that way because of bullshit numerology they pulled out of their rear end because it has a pleasing number of zeros is hilarious Well they're not entirely wrong about counter rollovers occurring close to a value with a ton of zeroes. For example, a 16-bit counter's maximum value is 0xFFFF; increment it by 1 and it rolls over to 0x0000 rather than 0x10000 (since the latter requires 17 bits to represent). The reason I don't fully buy the theory is that 0x37FFF.... is not a very plausible top value for a counter - a top value should be all-ones, but 0x37 hex = 0b110111 binary.
|
|
#
?
Jun 4, 2023 01:27
|
|
|
I guess if the top byte wasn’t part of the timer some how but that seems like it’d be weird.
|
|
#
?
Jun 4, 2023 03:50
|
|
|
Cygni posted:https://www.amazon.com/gp/product/B09VCJ2SHD
|
|
#
?
Jun 4, 2023 09:16
|
|
|
Well I got a 5800X3D installed. A few hiccups since apparently the 750 psu I have is...just dead? Is a 650 enough for 5800X3D, 6800, and a SSD? I think I need to run to microcenter anyway since the wraith prism cpu cooler doesn't seem like enough for this thing.
|
|
#
?
Jun 4, 2023 22:13
|
|
|
ChiTownEddie posted:Well I got a 5800X3D installed. A few hiccups since apparently the 750 psu I have is...just dead? Is a 650 enough for 5800X3D, 6800, and a SSD? Easily yes, if it's a good-quality 650. A 6800XT only pulls ~300 watts max, and the rest of the system under 200.
|
|
#
?
Jun 4, 2023 22:31
|
|
|
Just to jump on the bandwagon, I got a 5800X3D as well. figured I would max this build out, put my 5800X into my plex box (very inefficient AV1 decoding ahoy, also hosts game servers) and gave the 2600X to my cousin! That's totally a justified upgrade lol.
|
|
#
?
Jun 4, 2023 23:13
|
|
|
Klyith posted:Easily yes, if it's a good-quality 650. A 6800XT only pulls ~300 watts max, and the rest of the system under 200. Cool, I could have probably figured that out myself but I was still super annoyed with the 750W haha
|
|
#
?
Jun 4, 2023 23:33
|
|
|
https://twitter.com/VideoCardz/status/1665701503552684038 it's not super earth-shattering news, given that we haven't even seen a whiff of Ryzen 7000 desktop APUs, but: * it's nice if later generations of APUs keep getting updates to their graphics core, rather than four generations of Vega * this is the first "confirmation" that the next generation is still going to be on the AM5 socket. Of course, two gens on a socket is still only about as good as what Intel does/did, but still
|
|
#
?
Jun 5, 2023 14:01
|
|
|
if they ever released the rumored zen 3+ apus on AM5 it would technically be three generations but they're in seemingly no rush to put those out anytime soon
|
|
#
?
Jun 6, 2023 04:25
|
|
|
I'm pretty certain that they confirmed they'll be at least two generations on AM5 when they first announced it. And that slide actually confirms they plan on releasing products for AM5 at least through 2026.
|
|
#
?
Jun 6, 2023 06:42
|
|
|
https://twitter.com/AMDGPU_/status/1665754093170614272 link to the full article: Zen 4c: AMD’s Response to Hyperscale ARM & Intel Atom quote:Bergamo, AMD’s upcoming 128-core server part sets new heights in x86 CPU performance. Architected to be cloud native, Bergamo represents an important inflection point in datacenter CPU design as Moore’s Law grinds to a crawl. At the heart of Bergamo is Zen 4c, a brand-new CPU core variant of their successful 5nm Zen 4 microarchitecture that enables the push toward more cores per socket. While official details of Zen 4c have been rather scant so far, AMD’s Chief Technical Officer had this to say at their Ryzen 7000 Keynote: some of this is going over my head, but the headline seems to capture the gist: it's a variation on the design of Zen 4 that's "smaller", giving AMD something that will let them compete with ARM server-grade chips, while also competing with Intel's Atom-grade CPUs
|
|
#
?
Jun 7, 2023 06:57
|
|
|
It's very impressive that AMD could cut the core size practically in half with so few compromises. The challenges the article highlighted were illuminating too, e.g. the package limiting the number of CCDs, 3D cache TSVs taking enough area that it was worth removing them, and having to use lower performance SRAM cells for the sake of density. The last point in particular is interesting since it shows one approach they'll take to components that aren't scaling with node shrinks anymore. They had to cut the L3 in half, leading to the weird situation where there's only twice as much L3 per core as L2. Maybe they should have kept the TSVs and 3D stacked the L3 straight on top of the core  Here's a useful pic for those who don't want to read the whole article: 
|
|
#
?
Jun 7, 2023 11:52
|
|
|
ConanTheLibrarian posted:They had to cut the L3 in half, leading to the weird situation where there's only twice as much L3 per core as L2. Maybe they should have kept the TSVs and 3D stacked the L3 straight on top of the core I get the impression that the cloud stuff these sorts of million-core CPUs are going to aren't as big on CPU cache. The Neoverse N1 & 2 ARM chips only have 512KB of L2 per core, half as much as 4c, and similar amounts of L3. OTOH ARM is also doing a version of the Neoverse with way more cache, but that's supposed to be for AI poo poo.
|
|
#
?
Jun 7, 2023 13:58
|
|
|
ConanTheLibrarian posted:Maybe they should have kept the TSVs and 3D stacked the L3 straight on top of the core The die area available for stacking would be very small and I doubt you'd get more than 4MB of cache per core, which is what's already available on the normal 2D chips. I guess that begs the question of what would be cheaper to produce. Heat density would also probably become a serious problem. Since this is how AMD is doing it, I assume this is the cheaper or less problematic way. What's more likely is AMD doubling down on heterogeneous designs. Imagine if the 7950X3D had one of these as the 2D chip. One 8-core 3D CCD for cache-heavy tasks and one 16-core 2D CCD for raw compute. Dr. Video Games 0031 fucked around with this message at 14:17 on Jun 7, 2023 |
|
#
?
Jun 7, 2023 14:13
|
|
|
Klyith posted:I get the impression that the cloud stuff these sorts of million-core CPUs are going to aren't as big on CPU cache. The Neoverse N1 & 2 ARM chips only have 512KB of L2 per core, half as much as 4c, and similar amounts of L3. Dr. Video Games 0031 posted:The die area available for stacking would be very small and I doubt you'd get more than 4MB of cache per core, which is what's already available on the normal 2D chips. I guess that begs the question of what would be cheaper to produce. Heat density would also probably become a serious problem. Since this is how AMD is doing it, I assume this is the cheaper or less problematic way. Dr. Video Games 0031 posted:What's more likely is AMD doubling down on heterogeneous designs. Imagine if the 7950X3D had one of these as the 2D chip. One 8-core 3D CCD for cache-heavy tasks and one 16-core 2D CCD for raw compute. The funny thing is that average desktop users are probably still best served by an 8 core CPU. Who knows, maybe AMD return to monolithic CPUs for the low and mid end.
|
|
#
?
Jun 7, 2023 18:54
|
|
|
Klyith posted:I get the impression that the cloud stuff these sorts of million-core CPUs are going to aren't as big on CPU cache. The Neoverse N1 & 2 ARM chips only have 512KB of L2 per core, half as much as 4c, and similar amounts of L3. AWS has spent more than a decade at this point getting people used to 1 vCPU = 1 hyper-thread on a Sandy Bridge Xeon. With each new CPU generation, they use lower clock parts so that performance per vCPU stays roughly the same even though IPC has been improving. AMD gave you more performance than that, somewhat accidentally, because they don't have parts that slow. This Zen 4c is a way for AMD to deliver Sandy Bridge level performance, in-line with all of the other AWS vCPUs, much cheaper.
|
|
#
?
Jun 7, 2023 19:12
|
|
|
What's the best way to undervolt my 7900X? CPU load is low in Diablo 4 but the temps do rise above 70C every few minutes causing the fans to ramp up which then cools it down to like 63C before it heats up again.
|
|
#
?
Jun 8, 2023 22:01
|
|
|
Josh Lyman posted:What's the best way to undervolt my 7900X? CPU load is low in Diablo 4 but the temps do rise above 70C every few minutes causing the fans to ramp up which then cools it down to like 63C before it heats up again.
|
|
#
?
Jun 8, 2023 22:09
|
|
|
Josh Lyman posted:What's the best way to undervolt my 7900X? CPU load is low in Diablo 4 but the temps do rise above 70C every few minutes causing the fans to ramp up which then cools it down to like 63C before it heats up again. It'd be easier to change your fan curve.
|
|
#
?
Jun 8, 2023 22:10
|
|
|
Josh Lyman posted:What's the best way to undervolt my 7900X? CPU load is low in Diablo 4 but the temps do rise above 70C every few minutes causing the fans to ramp up which then cools it down to like 63C before it heats up again. Easiest was is a negative offset for core voltage. -15mv is easy and almost always stable, -20 or -25 will probably work but needs testing, -30 is where you start hoping you have a good chip. Better way is PBO2 curve tuning but that's a can of worms I don't have experience with. Alternately, just fix your fan curves so you don't have an annoying ramp-up. The zen 4s are happy operating at higher temperatures, so you could set that 70C as a smaller increase from the 60C zone.
|
|
#
?
Jun 8, 2023 22:11
|
|
|
BlankSystemDaemon posted:Wouldn't it be easier to fix your fan curves? Kibner posted:It'd be easier to change your fan curve.
|
|
#
?
Jun 8, 2023 22:11
|
|
|
If you really want to go through with this, you will want to use something like Core Cycler to verify your system isn't loving up unnoticeably. A "safe" undervolt on the 12-core 7900x will likely take 144 hours to validate. quote:If, for example, you're after a 12h "prime-stable" setup which is common for regular overclocks, you'd need to run this script for 12x12 = 144 hours on a 5900X with 12 physical cores, because each core is tested individually, and so each core also needs to complete this 12 hour test individually. Respectively, on a 5600X with its 6 physical cores this would be "only" 6x12 = 72 hours.
|
|
#
?
Jun 8, 2023 22:12
|
|
|
Josh Lyman posted:I did try messing fan with the fan curves in my Asus bios but the fundamental issue is that, to avoid the fan spin up, I have to have them running at a higher RPM all the time which I'd like to avoid. Have your fans max out at an rpm that you are audibly happy with. Odds are, 100% speed isn't going to be that much more effective than 50% speed.
|
|
#
?
Jun 8, 2023 22:14
|
|
|
Josh Lyman posted:I did try messing fan with the fan curves in my Asus bios but the fundamental issue is that, to avoid the fan spin up, I have to have them running at a higher RPM all the time which I'd like to avoid. Something with high airflow/static pressure, but low noise is what you would be looking for. While there are a few brands out there that can fit the bill, I'd go with Noctua without a second thought, other than they are "pricey". In BIOS news. The latest from Gigabyte has AGESA 1.0.0.7 A, and EXPO worked without a hitch. That latest version is to "Addresses Download Assistant Vulnerabilities Reported by Eclypsium Research", aka "we need to cover our asses", but hey, it's giving me working Expo, which I haven't had since BIOS version F2 or 3 I think it was (current is F6B).
|
|
#
?
Jun 8, 2023 22:36
|
|
|
Koskun posted:Are these the fans that came with the cooler and/or case? If so, the vast majority of the time they are the cheapest things they can ship. Only a few brands will ship actual good fans with their cases, and a few more with their coolers. Something like Ventoo coolers, the fans are there because they work, not much else. Noctua though, they work and are quiet.
|
|
#
?
Jun 8, 2023 23:10
|
|
|
Josh Lyman posted:Yeah I have the stock fans with my Thermalright Peerless Assassin as well as the 2 stock intake on my Fractal North. I thought those fans are supposed to be pretty okay though? They do what they need to do, which is move air. How well, how quiet, that isn't much of a concern. As an example. The Noctua NF-S12B (their black version) lists the dBA at 18.1, where Thermaltake's Toughfan 12 Turbo says theirs are 28.1. For reference, normal conversation is about 50 to 60 dBA. Now these noise level measurements can be taken with a bit of a grain of salt. Are they measuring them while on a table top with no restrictions, in a case, on a heatsink/radiator. There's a bunch of different circumstances that will effect noise of a fan. But Noctua make some of the quietest fans on the market, it's kind of what they are really known for (and their brown on brown color till recently). Your case is all mesh on the front, back, and side. Not that surprising you can hear the fans spinning up. I switched to an open bench style, and while I do sometimes hear my fans spin (I only have fans on the cpu cooler and GPU now), it's only under serious load, and with my headset on I don't notice it at all from less than 5' away. I happen to have the same CPU cooler too, but I added a third fan just because. One additional thing. I would get a fan and add it as an exhaust at the top of your case. Right now with only two fans as intake, all that heat is building up inside the case and only getting out from positive pressure and convection. Adding a single fan to the top will help get that settled heat out, and would probably help maintain/lower the temps in the case a bit. Virtually everything in a computer makes heat, and modern systems are making a lot of it. 300-400 watts is about 1/10 of a small space heater. Koskun fucked around with this message at 23:49 on Jun 8, 2023 |
|
#
?
Jun 8, 2023 23:39
|
|
|
The fans on a Peerless Assassin are adequate, noctua or whatever fancy poo poo is a bit better but wouldn't make a huge difference. If you care about noise you need to tune your fan curves, whether you have meh stock fans or fancy ones. A lot of noise is super specific to the setup, because there will be some PWN x% makes y RPM which generates extra resonance and a more noticeable sound than a tiny x-5% and y-50 RPM change. (Or maybe even +5% and +50!) Get in there and move the fan speed around to find some good spots where air seems to be whooshing rather than whining. Also the generic fan setups on a DIY mobo are built for stock CPU coolers, so they ramp the CPU fan way up when it gets hot. If you have a good tower cooler you can probably run the fans much lower and have zero change in CPU temperature. 5nm CPUs get hot, they generate heat in such a tiny amount of space that it can't conduct out fast enough.
|
|
#
?
Jun 9, 2023 01:17
|
|
|
I wonder, just a little bit, if we might not be better off going back to planar transistors. Increase the area of a chip to make cooling it a little easier, but no, gotta pack as many chiplets into a wafer as possible
|
|
#
?
Jun 9, 2023 06:20
|
|
|
That would work but everything would be physically bigger, therefore cost more in raw materials, transport, etc.
|
|
#
?
Jun 9, 2023 11:32
|
|
|
|
| # ? May 29, 2024 15:24 |
|
|
Josh Lyman posted:I did try messing fan with the fan curves in my Asus bios but the fundamental issue is that, to avoid the fan spin up, I have to have them running at a higher RPM all the time which I'd like to avoid.
|
|
#
?
Jun 9, 2023 14:35
|
|
