|
terrorist ambulance posted:Can you update a motherboards bios with processor etc connected? I reviewed the manual for my motherboard and it says it's just as simple as putting it on a UDB and pushing a button but don't want to gently caress it up Generally, no. They want no cpu/ram/cards installed.
|
#
?
Dec 12, 2020 03:15
|
|
|
|
| # ? Jun 3, 2024 18:11 |
|
|
Updated w/ M-Flash and it worked really well easily, after all. Latest MSI drivers appear to have fixed the memory problem, I set XMP to 3600 mhz w/ no problems, no errors in eventviewer
|
|
#
?
Dec 12, 2020 04:04
|
|
|
RocketSurgeon posted:Hey, anyone on an msi board with bios flashback? I have a b550 tomahawk and the flashback led keeps flashing for over 20 minutes now. The cpu is on the board the ram aswell but both the hdd's are disconnected. Ive read that its ok to update with the cpu installed. On a B550 tomahawk while it is flashing the motherboard will display fun colors, it will do more than just blink the rear LED. If only the rear LED is blinking that seems like it's hanging. I did it right out of the box nothing connected besides the 2 power connections to the board.
|
|
#
?
Dec 12, 2020 04:32
|
|
|
bus hustler posted:On a B550 tomahawk while it is flashing the motherboard will display fun colors, it will do more than just blink the rear LED. If only the rear LED is blinking that seems like it's hanging. I did it right out of the box nothing connected besides the 2 power connections to the board. Alright. I'll try a different thumbdrive first just in case then yank the cpu and try again. E: Apparently the board didnt like my choice of thumbdrive afterall. RocketSurgeon fucked around with this message at 09:55 on Dec 12, 2020 |
|
#
?
Dec 12, 2020 08:36
|
|
|
Apparently Cyberpunk (or at least parts of it) were compiled with Intel's C++ toolchain, which dispatches to the baseline slow path on AMD CPUs even if they're capable of running faster paths https://old.reddit.com/r/Amd/comments/kbp0np/cyberpunk_2077_seems_to_ignore_smt_and_mostly/ https://old.reddit.com/r/Amd/comments/kbp0np/cyberpunk_2077_seems_to_ignore_smt_and_mostly/gfjf1vo/?context=3 There's a hex patch that supposedly overrides the vendor check and improves performance 
|
|
#
?
Dec 12, 2020 19:49
|
|
|
repiv posted:Apparently Cyberpunk (or at least parts of it) were compiled with Intel's C++ toolchain, which dispatches to the baseline slow path on AMD CPUs even if they're capable of running faster paths I manually did the hex edit (you just need to change one value from 75 to 74) and It definitely looks like I've got more performance on my 5800x now. It could be just placebo because I didn't measure the FPS I was getting before but it seems to be better/smoother overall.
|
|
#
?
Dec 12, 2020 20:20
|
|
|
Cygni posted:I think everyone is using the same Foxconn sockets on their AM4 boards (at least the boards with metal arms), and Asus sells a version with wifi/bt? Ya for some reason ASUS has their main brand ASUS and then their Asus Rog line. But for the x570 the jump from the highest regular Asus mobo to the first Rog setup was not worth it. Their pricing is much better in their intel mobo lineup. CaptainSarcastic posted:Personally I don't want Bluetooth/wifi on my mobo, and would rather run them as discrete parts. It just strikes me as one more possible point of failure, and would rather pick and choose my wireless components. Ya I don't mind the no wifi sinces I am running a fiber optics 1.5 gbps. But been loving this Bose Bequiet headset, it is so freaking comfortable, looks amazing compared to most of the gaming oriented junk I had previously. Works so smooth for a bluetooth, none of the connectivity issues of reconnecting or not finding devices. You connect to several devices and it prioritize whatever you actually are. Also I went from a Intel rig to AMD and the asus prime pro doesn't have a bios flashback slot either. So if you end up in that position I recommend getting a ryzen 3600x, I think it might be the only ryzen chip with integrated graphics? You can get into bios and upgrade through a usb. It was a whole mess. After when I put back my 5800x and gpu, I completely forgot I still had the dp plugged into the motherboard instead of the gpu, so stood there like a dumbass waiting for something to appear. Was nervous I had hosed something up. I realize now to not cheap out on the mobo, since the extra features are good to have.
|
|
#
?
Dec 12, 2020 20:28
|
|
|
Ulio posted:Ya for some reason ASUS has their main brand ASUS and then their Asus Rog line. But for the x570 the jump from the highest regular Asus mobo to the first Rog setup was not worth it. Their pricing is much better in their intel mobo lineup. 3600X does not have graphics onboard. The APUs like the 3200G do.
|
|
#
?
Dec 12, 2020 21:14
|
|
|
repiv posted:Apparently Cyberpunk (or at least parts of it) were compiled with Intel's C++ toolchain, which dispatches to the baseline slow path on AMD CPUs even if they're capable of running faster paths Did the patch and got this. Seems it worked? I just loaded a save before and after and waited about 10 seconds for it to stabilize. https://imgur.com/a/l1dnpqH edit. 5600X and 3060TI, d'oh Jeff Fatwood fucked around with this message at 22:06 on Dec 12, 2020 |
|
#
?
Dec 12, 2020 21:54
|
|
|
Hah, it's looking more like this is actually AMDs fault The problem is that Cyberpunk only launches a worker thread for each physical core on AMD processors, rather than each logical core to take advantage of SMT, and it turns out there's a handy little library on AMDs GPUOpen for detecting how many cores the system has. For whatever reason it checks if it's running on an AMD processor, and if so it only counts logical cores if it's a Bulldozer chip, otherwise it returns the number of physical cores. CDPR are almost certainly using this: https://github.com/GPUOpen-LibrariesAndSDKs/cpu-core-counts/blob/master/windows/ThreadCount-Win7.cpp#L69 repiv fucked around with this message at 22:24 on Dec 12, 2020 |
|
#
?
Dec 12, 2020 22:16
|
|
|
Possible, but one of the most helpful rules of coding is that the comments are lying to you.
|
|
#
?
Dec 12, 2020 22:57
|
|
|
I'm not sure why the comments matter? The code clearly intends to ignore SMT on AMD unless the CPU family is 15h (Bulldozer) I wonder how many other application are leaving Ryzen performance on the table because the developers copy pasted that snippet
|
|
#
?
Dec 12, 2020 23:05
|
|
|
Ulio posted:Ya I don't mind the no wifi sinces I am running a fiber optics 1.5 gbps. But been loving this Bose Bequiet headset, it is so freaking comfortable, looks amazing compared to most of the gaming oriented junk I had previously. Works so smooth for a bluetooth, none of the connectivity issues of reconnecting or not finding devices. You connect to several devices and it prioritize whatever you actually are. I've noticed Windows 10 is way better at handling Bluetooth than Windows 7 was. I'm running Bluetooth off a ASUS PCE-AC55BT PCIe card and it's been great - it's basically a laptop Intel wifi/Bluetooth combo that sits in a PCIe card. It's been reliable and handles my headphones really well - even hotswapping between my JBLs and backup Skullcandys.
|
|
#
?
Dec 13, 2020 00:35
|
|
|
  Yep, that's a whole lot more CPU being used! What a weird oversight.
|
|
#
?
Dec 13, 2020 03:51
|
|
|
repiv posted:Hah, it's looking more like this is actually AMDs fault If it's an AMD processor, and if the processor is a Bulldozer, then the library will return a number of cores equal to the number of logical cores If it's an AMD processor, and if the processor is NOT a Bulldozer (i.e., is a Ryzen), then the library will return a number of cores equal to the number of physical cores (i.e. not taking advantage of SMT) I don't get it. Why would the function work that way?
|
|
#
?
Dec 13, 2020 06:18
|
|
|
gradenko_2000 posted:If it's an AMD processor, and if the processor is a Bulldozer, then the library will return a number of cores equal to the number of logical cores Perhaps the logic predates Ryzen?
|
|
#
?
Dec 13, 2020 06:23
|
|
|
Subjunctive posted:Perhaps the logic predates Ryzen? but Ryzen were the first AMD CPUs to have SMT - if you were running a triple-core Phenom or whatever, it was a "3-physical-core, 3-logical-core" processor early in Bulldozer's lifecycle, Windows would treat an FX-4300 as a four-core processor, which means the function wouldn't matter either way later, Windows would see it as a two-core-four-thread processor, in which case the function still wouldn't make sense (at least to me), because you actually want to do it the other way around where a Bulldozer would only return as many cores as there are "physical cores" so you're only launching one thread per "module"
|
|
#
?
Dec 13, 2020 06:43
|
|
|
gradenko_2000 posted:but Ryzen were the first AMD CPUs to have SMT - if you were running a triple-core Phenom or whatever, it was a "3-physical-core, 3-logical-core" processor Since Bulldozer threads share a front-end (and running two threads at once does bottleneck each thread) maybe it was desirable to only run 1 thread per module to keep performance up.
|
|
#
?
Dec 13, 2020 07:30
|
|
|
Paul MaudDib posted:Since Bulldozer threads share a front-end (and running two threads at once does bottleneck each thread) maybe it was desirable to only run 1 thread per module to keep performance up. yeah, I get that - if core 1 and core 2 are on the first module, and cores 3 and 4 are on the second module, and you have two threads you want to run, then it's better to run them on core 1 and 3 but like I said, the function would seem to work the opposite of what I might expect: "If it's an AMD processor, and if the processor is a Bulldozer, then the library will return a number of cores equal to the number of logical cores" - this statement is going to return a "4" on an FX-4300 regardless of whether Windows considers it a 4c/4t CPU, or a 2c/4t CPU in order to run only 1 thread per module, you'd have to consider it a 2c/4t CPU (which is what Windows does nowadays), and then only return the number of physical cores, which would be two
|
|
#
?
Dec 13, 2020 07:48
|
|
|
Subjunctive posted:Perhaps the logic predates Ryzen? That's what I guessed, but the same code appears in this AMD presentation from 5 months ago https://www.youtube.com/watch?v=d0TMCbDDg5g&t=798s They say SMT is usually a performance win and should be used unless profiling indicates otherwise... while showing code that will default to ignoring SMT on modern AMD processors idgi
|
|
#
?
Dec 13, 2020 14:26
|
|
|
repiv posted:That's what I guessed, but the same code appears in this AMD presentation from 5 months ago The presentation says that most applications benefit from SMT, but many games have a main thread that is responsible for updating the results from all of the other threads. Turning on SMT can negatively impact performance because that single thread needs all the power it can get. They also caveat that your game may be different, and it's up to the devs to actually profile how the game runs. The bulldozer exception is because bulldozer doesn't do real SMT, it has 2 physical cores that report to windows as a single core because they have some shared integer processing units. Ironically I remember seeing games perform better by setting core affinity in windows to avoid those IPU collisions in the bulldozer lineup. It does seem like these are recommendations based on Zen 1 and older game engines, Cyberpunk's engine is threaded to hell to make it function on consoles, and it's possible that Zen 2+ doesn't lose as much to SMT as Zen 1 may have.
|
|
#
?
Dec 13, 2020 17:23
|
|
|
Blorange posted:The presentation says that most applications benefit from SMT, but many games have a main thread that is responsible for updating the results from all of the other threads. Turning on SMT can negatively impact performance because that single thread needs all the power it can get. They also caveat that your game may be different, and it's up to the devs to actually profile how the game runs. The bulldozer exception is because bulldozer doesn't do real SMT, it has 2 physical cores that report to windows as a single core because they have some shared integer processing units. Ironically I remember seeing games perform better by setting core affinity in windows to avoid those IPU collisions in the bulldozer lineup. Phrasing it this way made me realize that SMT is a teeny tiny step towards heterogeneous processing on x86. I’ve been wondering if binning could get better if we let cores run at different max clocks like Android has been doing for years, or if we’ll see x86 CPUs do full on big and little cores like Apple and Qualcomm do. Did that Intel thing with 1 core core and 4 atom cores ever ship? It seems like right now, if you have one thread that’s a bottleneck for everything, make sure nothing else gets scheduled on it’s SMT partner.
|
|
#
?
Dec 13, 2020 17:29
|
|
|
Twerk from Home posted:Phrasing it this way made me realize that SMT is a teeny tiny step towards heterogeneous processing on x86. I’ve been wondering if binning could get better if we let cores run at different max clocks like Android has been doing for years, or if we’ll see x86 CPUs do full on big and little cores like Apple and Qualcomm do. Did that Intel thing with 1 core core and 4 atom cores ever ship? Intel's 12th generation "Alder Lake" is rumoured to be due 2H 2021 and be a mainstream consumer CPU using big (Golden Cove) little (Gracemont) on 10nm, supporting DDR5 and PCI-E 5.0 apparently.
|
|
#
?
Dec 13, 2020 17:56
|
|
|
Anyone else having problems with MSI Afterburner and the 5600X? The on screen display doesn't seem to display monitor CPU temps, even when I've selected it in the OSD/Monitoring settings. I didn't have this problem with my 3600x.
|
|
#
?
Dec 13, 2020 20:12
|
|
|
mA posted:Anyone else having problems with MSI Afterburner and the 5600X? The on screen display doesn't seem to display monitor CPU temps, even when I've selected it in the OSD/Monitoring settings. I didn't have this problem with my 3600x. Afterburner hasn't been updated yet to link in with Zen3's thermal sensors--in fact, a lot of hardware monitoring software doesn't detect Zen 3's temps right now, or don't detect them correctly. Best I can suggest is to also run Ryzen Master, which picks up the temps fine, and circle back on Afterburner in a month or so and see if they've updated it.
|
|
#
?
Dec 13, 2020 20:52
|
|
|
I just installed a 5800x from ANTOnline last night, so far it's been pretty easy. Had a brief moment of panic when that first boot took a while to POST, but aside from that the flashback process was painless, install seems to have gone well, and appeared to take the XMP profile with no problems so far. I thrashed the hell out of the CPU for a bit with Prime95, I'm able to get it up to around 85-86C but it's still staying under the 90C max so whatever.
|
|
#
?
Dec 14, 2020 01:44
|
|
|
I'm using Ryzen Master, HWMonitor and the mobo's vendor program, and they're all reporting different CPU temperatures on my 5900x. Which one should I trust?
|
|
#
?
Dec 14, 2020 04:13
|
|
|
taco_fox posted:I'm using Ryzen Master, HWMonitor and the mobo's vendor program, and they're all reporting different CPU temperatures on my 5900x. Which one should I trust? Only one of those is actually written by AMD, so I'd go with that one as the authoritative reading.
|
|
#
?
Dec 14, 2020 04:20
|
|
|
AMD just benchmarked their 128 core Epyc: https://www.notebookcheck.net/128-core-AMD-EPYC-Milan-server-CPU-scores-an-eye-watering-87878-on-CineBench-R23.509223.0.html
|
|
#
?
Dec 14, 2020 04:28
|
|
|
What's good software to manage fan speeds with these days?
|
|
#
?
Dec 14, 2020 06:15
|
|
|
Argus Monitor.
|
|
#
?
Dec 14, 2020 06:28
|
|
|
Micro Center has now decided that the 3900X is on sale at $530, because they list the regular price as $600 These numbers all seem perfectly legit, and not inflated in any way.
|
|
#
?
Dec 14, 2020 07:12
|
|
|
Farmer Crack-rear end posted:What's good software to manage fan speeds with these days? Argus monitor is dope and what I'm using, but there's also Fan Control which was released earlier this year and is freeware.
|
|
#
?
Dec 14, 2020 09:32
|
|
|
mdxi posted:Micro Center has now decided that the 3900X is on sale at $530, because they list the regular price as $600 I got that for $380 months ago. :o
|
|
#
?
Dec 14, 2020 14:21
|
|
|
.
sincx fucked around with this message at 05:50 on Mar 23, 2021 |
|
#
?
Dec 14, 2020 14:26
|
|
|
tomshardware benchmarked the cyberpunk ryzen SMT fix https://www.tomshardware.com/news/cyberpunk-2077-amd-ryzen-performance-bug-fix-testing seems to be a major performance win on <=8 core chips, but a slight regression on >8 core chips repiv fucked around with this message at 14:36 on Dec 14, 2020 |
|
#
?
Dec 14, 2020 14:27
|
|
|
Does anyone know of a stock tracker for Ryzen 5 3xxx CPUs? I looked into some stock drops discords but they only track 5xxx series.
|
|
#
?
Dec 14, 2020 14:32
|
|
|
sincx posted:Buying my 3900X back in February was definitely the right move, in retrospect Yeah Covid has wrecked normal pricing.
|
|
#
?
Dec 14, 2020 14:45
|
|
|
Yeah, I've just completely written off building anything new. I'm struggling to even find used parts that aren't going to cost me $ARM + $LEG + $KIDNEY. Yields for 5000 series must be loving amazing, though. I can't imagine that if AMD had a surplus of defective chiplets, that they wouldn't be trying to rush 5300s out the door as 4c/8t parts. Losing cores never feels good, but going from a 1600 to 5300 is probably an upgrade. SwissArmyDruid fucked around with this message at 14:54 on Dec 14, 2020 |
|
#
?
Dec 14, 2020 14:51
|
|
|
|
| # ? Jun 3, 2024 18:11 |
|
|
I’m not even sure how I managed my new build at RRP to be honest.
|
|
#
?
Dec 14, 2020 16:16
|
|
