|
Paul MaudDib posted:i've been looking at the Asrock Rack X470D4U, it was on sale on Amazon for $220 a couple days ago, but they still haven't updated it for 5000 series CPUs yet...
|
#
?
Feb 21, 2021 12:44
|
|
|
|
| # ? Apr 23, 2024 21:29 |
|
|
Not sure if anyone would know the optimal way to do this. I'm trying to multibox three copies of a game at once on my 5800x. The game is pretty CPU heavy, so I told it to do cores 5-6-7-8 for 1st copy of the game, then 9-10-11-12 for second copy and 13-14-15-16 for the third and final copy. Seems to be somewhat better than telling all games they can run on all cores and have at it, but only slightly. I heard I should try to keep each instance of the game to it's own cores/threads, i.e. avoid giving Game 1 a thread that is on a core of Game 2 because they'll kind of battle for performance on that core from each other. Kivi posted:There's beta BIOS. My friend just picked up that board and claimed that it works with 5000 series. Yeah all that ASRock x470 stuff works fine with 5000 but you need an older CPU to flash the beta BIOS. My X570D4I-2T is working with a 5800x after I bought and returned a 3600 to flash it with.
|
|
#
?
Feb 22, 2021 08:19
|
|
|
priznat posted:This makes sense! I should ask the apps guys who deal with customers what they use for their servers, but for consumer some kind of FR-4 like that will probably be de rigueur. I was reading the AMD Rome/Milan design app notes and they are extremely vague when it comes to acceptable channel loss, it might be in a different document I didn't get a hold of. I think Microsemi has some retimers to along with their Gen 5 launch though -- they are not wasting any time. What is intra-pair skew / channel length matching like now from their design guides? Do you still have to saw a board in half to attach your Infinium / Tek DSA to be able to run the Rx / compliance tests?
|
|
#
?
Feb 22, 2021 19:10
|
|
|
movax posted:I think Microsemi has some retimers to along with their Gen 5 launch though -- they are not wasting any time. The diff pair match is low, 0.75ps whereas the lane to lane matching is 1100 ps. This is on the Rome/Milan, Ryzen is probably less strict. It is funny how they specify the the max lengths for gen1-3 and then for 4 it is a section that is basically “it depends, lol”. A lot is fobbed off to the CEM spec. For compliance tests on gen4 it was still using compliance load board (with an edge connector to SMP) for systems and compliance base boards (with a slot to SMPs) that would put the devices into compliance mode with a toggle to switch between presets. Gen5 is still a bit of a “emmm, TBD! Boards may be available in september!”, lol.
|
|
#
?
Feb 22, 2021 21:03
|
|
|
What's a good starting point for undervolting a 5800x? Also, someone mentioned there's a way in Ryzen Master to set a wattage ceiling? How can I do that?movax posted:I think Microsemi has some retimers to along with their Gen 5 launch though I thought this was some kind of pun. Like, first I was Microsoft, but now I've got a Microsemi Zero VGS fucked around with this message at 23:20 on Feb 22, 2021 |
|
#
?
Feb 22, 2021 21:24
|
|
|
Zero VGS posted:I thought this was some kind of pun. Like, first I was Microsoft, but now I've got a Microsemi Microsemi is now owned by.. Microchip  (not even joking!)
|
|
#
?
Feb 22, 2021 23:20
|
|
|
I've been having USB issues, especially with my mouse and keyboard, ever since I built my new PC (3900X, ASRock X570 Taichi) a year ago. USB has been kind of a shitshow on every PC I've built so I just assumed that's the way things are, but a combination of AMD's announcement this week and things getting especially bad last night (I was slip-sliding all around levels in Amid Evil because my keyboard kept losing connection) caused me to try some troubleshooting. Changing which port my USB hub was plugged into didn't anything, turning my PCI-E slots down to Gen3 didn't do anything, updating my BIOS didn't do anything. But changing "Global C-State Control" from "Auto" to "Disabled" in the BIOS does seem to have done the trick, at least so far. My question is, what am I giving up by having that setting off? I was given to understand that C-States had to do with suspend/hibernate modes, but sleep still seems to work.
|
|
#
?
Feb 23, 2021 03:24
|
|
|
some of the deeper C-states are capable of clocking down internal or external busses on the CPU, maybe that has something to do with it. it's unclear why a processor would be entering a deeper C-state while it's doing VR gaming though, that is not really a "park things and go to sleep" scenario. it's ultimately probably just a driver/firmware bug and they'll probably pin it down now that it's on their radar
|
|
#
?
Feb 23, 2021 03:35
|
|
|
Hello, I had a bit of a mishap removing my amd wraith prism cooler off my 3700x, the whole cpu came with it and some of the cpu pins were bent. I managed to get them straight and socket the cpu back but one of pins was slightly bent from the very tip making it a bit shorter than the other pins, do the pins need to go all the way back to make contact to work? It does boot fine and finds all my memory also so I don't see an immediate problem with it at least, I'd rather not fiddle with the pins any more in fear of breaking one(not even sure how to straighten just the end of a pin without bending the whole pin) Thanks.
|
|
#
?
Feb 23, 2021 09:28
|
|
Delerion posted:Hello, I had a bit of a mishap removing my amd wraith prism cooler off my 3700x, the whole cpu came with it and some of the cpu pins were bent. Fyi, I've had this happen a couple of times with the wraith prism and the trick is to run the PC for a while. Even run a benchmark or something to heat up the paste, then shutdown and remove the block. This gives the paste time to loosen up a bit and not bind
|
|
|
#
?
Feb 23, 2021 09:37
|
|
|
CyberPingu posted:Fyi, I've had this happen a couple of times with the wraith prism and the trick is to run the PC for a while. Even run a benchmark or something to heat up the paste, then shutdown and remove the block. This gives the paste time to loosen up a bit and not bind Yeah I read about this method after I did it(apparently its a common thing with the stock paste atleast)in any case I'm on my way to buy another cooler and after that I'll do some stress test on the cpu and just hope it'll work for a year or two still.
|
|
#
?
Feb 23, 2021 10:01
|
|
|
Intel solved this problem in 2004. It's a shame AMD don't provide an effective retention mechanism on the PGA chips.
GRINDCORE MEGGIDO fucked around with this message at 10:38 on Feb 23, 2021 |
|
#
?
Feb 23, 2021 10:35
|
|
GRINDCORE MEGGIDO posted:Intel solved this problem in 2004. It's a shame AMD don't provide an effective retention mechanism on the PGA chips. Yeah the lever for AMD boards is loving awful.
|
|
|
#
?
Feb 23, 2021 10:37
|
|
|
What I've been doing is to unscrew the retention bracket from the motherboard backplate, while holding the cooler in place, so that the tension is released gradually from the cooler from that direction. It sucks, and the ideal solution is to still have a bracket between the CPU and the cooler so that the thermal paste never causes the cooler to rip the CPU off the socket (I posted a while back about a manufacturer that's trying to make a third-party accessory like this), but I still feel that AMD's socket lay-out is still better than Intel's.
|
|
#
?
Feb 23, 2021 10:48
|
|
gradenko_2000 posted:What I've been doing is to unscrew the retention bracket from the motherboard backplate, while holding the cooler in place, so that the tension is released gradually from the cooler from that direction. It sucks, and the ideal solution is to still have a bracket between the CPU and the cooler so that the thermal paste never causes the cooler to rip the CPU off the socket (I posted a while back about a manufacturer that's trying to make a third-party accessory like this), but I still feel that AMD's socket lay-out is still better than Intel's. So funnily enough. I've never had this issue with my Fuma 2 cooler. Which doesn't use those god awful bracket arms that the wraith prism and other coolers use. It uses a 2 screw bracket system which releases the pressure gradually
|
|
|
#
?
Feb 23, 2021 10:51
|
|
|
This has happened to several goons in this thread, another tip was to twist the cooler off, some of them say that in their directions.
|
|
#
?
Feb 23, 2021 16:39
|
|
|
Quaint Quail Quilt posted:This has happened to several goons in this thread, another tip was to twist the cooler off, some of them say that in their directions. Although I feel like this would make me even more worried about bending pins, but maybe my intuition is way off there.
|
|
#
?
Feb 23, 2021 18:36
|
|
|
If the CPU isn't fully seated, there's nothing supporting the bits of the pin that are exposed, so they're much more fragile. It's pretty robust when it's fully seated, enough to twist the heatsink and break any seal.
|
|
#
?
Feb 23, 2021 19:00
|
|
|
I have no feelings one way or the other on whether or not AM5 sticks with ZIF sockets, but if AM4 was ZIF as a way of throwing a bone to mobo manufacturers instead of shifting the increased cost of LGA sockets onto them to get them on board for Zen 1, maybe they have enough good will to move to LGA for AM5?
|
|
#
?
Feb 24, 2021 00:46
|
|
|
I don't know if there are any AMD specific mobo manufacturers, but all the mainstream ones are already making Intel boards, so it's not like would they would have to move mountains to put a better socket on there.
|
|
#
?
Feb 24, 2021 00:56
|
|
|
Intel is going to a new socket retention bracket (and heatsink hole pattern) later this year for the first time in 12 years. If AMD really wanted to do the board makers a favor and keep BOM down, they would find a way to use the same parts... but i imagine intel has got a patent on the mechanism anyway so!
|
|
#
?
Feb 24, 2021 01:31
|
|
|
I thought Intel had a patent on the LGA until Threadripper sockets came out with them 🤷♂️ Idea: make the lever on the ZIF a big plate that has a hole shaped like the IHS but slightly bigger an make it clip or screw into place on the side of the socket opposite the hinge Munkeymon fucked around with this message at 06:44 on Feb 24, 2021 |
|
#
?
Feb 24, 2021 06:41
|
|
|
I've been building PCs since 1997, and I also worked in a data center for three years -- during which time I socketed and desocketed more CPUs than I can remember (in the dozens for sure, but probably not a hundred). And I did three CPU swaps in the past 10 days, so my old rear end is still doing this without killing CPUs. A fair amount of them have come up with the HSF, because yeah, that's a thing that happens. And I have,on occasion, slightly bent a CPU pin. Maybe around 10 in total. So yes, also totally a thing that happens. What I don't get is what's going wrong for people who are unsocketing a CPU and bend many pins, all at once, some horribly so. Or even more confusingly, people who are socketing a CPU and bend some ridicuous fraction of the pins (these seem to mostly come from reddit, so there's already grains of salt involved). I do not understand how to accomplish either of those things without (1) actual malice, or (2) a lack of care and attention so staggering that it is indistinguishable from malice. I really wish there was multi-angle video to go along with these stories. I prefer LGA over ZIF sockets too, but I've never had a CPU mangled by either socket type. I have bent more pins through mishandling than via socket trauma. I find these reports genuinely befuddling. LGA or ZIF, you're not supposed to be pushing. The chip will either drop into place once aligned, or require so little pressure that you can't really feel it. If you can feel pressure in your fingertips, something is not right and you need to stop and examine the situation. Also, LGA or PGA, you should never handle any CPU except by the substrate PCB. And wipe your hands with an alcohol pad/windexed paper towel before handling a CPU, to make sure your fingertips aren't getting thermal compound on the pins/contacts. It's electrically conductive and can cause some all manner of fuckery if it bridges pins. And never touch a socket with your fingers at all, of either type. If you must know what it feels like; you can touch it just before you throw it away. And if the CPU does come up with the HSF, that's nothing to panic about and not hard to solve. Sit the HSF upside down, then place the blade of a flathead screwdriver (a small one like a terminal driver, NOT that 5/8" blade with the 11" shaft from the machine shop) under the lip of the package substrate, and gently lever it free. Dessicated thermal compound may act as cement against a bit of linear tension, but it isn't meant to be a cement and it doesn't stand up to torque or shear forces (which is also why the "twist gently before pulling" thing helps). And when you go to remove a HSF or unsocket a CPU, you should be pulling directly away from the motherboard. Which should always be the same as "up", because you should always be doing this on a work surface with the motherboard flat on said surface. If you've got room to do this in the case, fine, but the mobo should still be horizontal. And any time you separate a CPU and HSF, clean both thoroughly. And don't hold CPUs while you're cleaning them. Sit them in their little clamshell, or socket them, and clean them there. And use alcohol pads or windex-dampened pieces of paper towels. And swap out the pad/towel frequently; do NOT push globs of thermal compound around because you don't want it getting where it shouldn't be. And wet the paper towel, not the CPU. That's everything I can think of. I hope this helps less CPUs die.
|
|
#
?
Feb 24, 2021 07:32
|
|
|
Some heatsinks require you to really manhandle the bracket to be able to get them attached. I'm guessing that they are undoing such a heatsink, the undo the bracket enough that it suddenly jumps up from the board, carrying the CPU, and then lands on the pins, bending a bunch. I definitely had a heatsink a few years ago that terrified me when I was putting it on. I had to screw one side down a bit, and then put a decent chunk of my body weight into getting the other side down to get it screwed down. It scared me enough that I studied the instructions over and over again to make sure I had everything in the right orientation.
|
|
#
?
Feb 24, 2021 07:56
|
|
|
I've had to repair two CPUs with bent pins in my admittedly short time in this hobby, but that's because I buy second-hand from the Aliexpress market where they come to you in a small plastic tray/box swathed in multiple layers of bubble wrap
|
|
#
?
Feb 24, 2021 08:17
|
|
|
Just to update i did get the pins straightened out but the motherboard was done for(it broke the plastic part where the cpu is seated on the motherboard) Anyways the cpu seems to be working fine now, i couldn't get the pins completely straight but it did socket in and im seeing no issues(all 4 memory slots working at dual-channel)
|
|
#
?
Feb 24, 2021 19:44
|
|
|
Delerion posted:(it broke the plastic part where the cpu is seated on the motherboard) Are you Shrek?
|
|
#
?
Feb 24, 2021 20:34
|
|
|
Wow, I didn't realize what a difference RAM settings could make. The top image is my latest 3D mark score with 32GB of 3600 RAM overclocked to 4000 with some timings borrowed from a friend who took the time to tune his RAM. The bottom image is my first run when I first installed my 5800X with old RAM that was 3200 with only the stock DOCP profile. It's just too bad that most games are GPU bottlenecks, so I need to get a 3080 or something.
|
|
#
?
Feb 25, 2021 21:27
|
|
|
Overclocking RAM doesn’t translate neatly to games like overclocking the CPU/GPU so I wouldn’t read too much into synthetic benchmarks.
|
|
#
?
Feb 26, 2021 01:09
|
|
|
Manually tuning ram timings can have considerable improvements in game, but only in cpu bound scenarios.
|
|
#
?
Feb 26, 2021 01:27
|
|
|
shrike82 posted:Overclocking RAM doesn’t translate neatly to games like overclocking the CPU/GPU so I wouldn’t read too much into synthetic benchmarks. It’ll help with frame time consistency if you’re having issues with that which won’t show up as easily on a FPS counter.
|
|
#
?
Feb 26, 2021 01:28
|
|
|
we can argue about qualitative improvements but that CPU number jumping by 16% is pretty misleading my personal anecdote is not noticing my XMP settings reset after a bios update and running games/ML for half a year at 2400. i recently noticed it and turned it back on, and haven't really seen or felt a substantial change in performance (even running game benchmarks) 
|
|
#
?
Feb 26, 2021 01:33
|
|
|
hobbesmaster posted:It’ll help with frame time consistency if you’re having issues with that which won’t show up as easily on a FPS counter. tongue-in-cheekly: surely there's a way to reduce frame consistency to a single number. Calculate the variance across the ms values of a benchmark run?
|
|
#
?
Feb 26, 2021 01:33
|
|
|
And how many people will set that up.
|
|
#
?
Feb 26, 2021 01:34
|
|
|
Benchmarks thorough enough as to the point of tediousness, wherever will we get those?
|
|
#
?
Feb 26, 2021 01:56
|
|
|
Some Goon posted:Benchmarks thorough enough as to the point of tediousness, wherever will we get those? One of the biggest points to remember during all of this -- and any other memory testing published by other outlets -- is that motherboard matters almost more than the memory kit itself. Motherboards are responsible for most of the timings auto configured on memory kits, even when using XMP, as XMP can only store so much data per kit. The rest, including unsurfaced timings that the user never sees, are done during memory training by the motherboard. Motherboard manufacturers maintain a QVL (Qualified Vendor List) of kits tested and approved on each board, and we strongly encourage system builders to check these lists rather than just buying a random kit of memory. Motherboard makers will even tune timings for some kits, so there’s potentially a lot of performance lost by using mismatched boards and memory.
|
|
#
?
Feb 26, 2021 02:06
|
|
|
Yes? You can't just buy one of the kits they have listed and expect to perform exactly the same, but when the kit where they manually tuned a bunch of the clocks on pulls well ahead in every test, there's clearly a quantitative benefit to well-tuned ram.
|
|
#
?
Feb 26, 2021 02:26
|
|
|
shrike82 posted:One of the biggest points to remember during all of this -- and any other memory testing published by other outlets -- is that motherboard matters almost more than the memory kit itself. Motherboards are responsible for most of the timings auto configured on memory kits, even when using XMP, as XMP can only store so much data per kit. The rest, including unsurfaced timings that the user never sees, are done during memory training by the motherboard. Motherboard manufacturers maintain a QVL (Qualified Vendor List) of kits tested and approved on each board, and we strongly encourage system builders to check these lists rather than just buying a random kit of memory. Motherboard makers will even tune timings for some kits, so there’s potentially a lot of performance lost by using mismatched boards and memory. it's remarkable how well AMD has done at eliminating the hassle of CPU overclocking, like humans are basically unable to outperform default settings by more than trivial amounts without pumping dangerous amounts of voltage through the chip and vastly increasing power consumption/etc and that's really something. Per-CCX overclocking is sorta a thing but that will probably be integrated within a few generations too. Hopefully they can eventually branch this out to memory and IF tuning as well, have some sort of factory-approved tool that works like MSI OC Scanner does for GPUs and eliminate this last friction point, because I just look at the vagaries of memory tuning and cringe. It would be so nice to just push a button and come back in an hour and the board has found some reasonable approximation of "optimal" tunings and fabric speed within some manufacturer-approved voltage limit for your CPU, your memory kit, and your board/AGESA/etc. I know there are programs which can guess some good timings for you but it's still imposing and it's not officially approved in the same way that Ryzen factory boost clocks are. It is a bit of a bummer that you can no longer get a free 30% performance just by dragging a couple sliders, but that has gone away naturally due to the delicateness of newer nodes and the need for integrated boost management to keep cores within stable operating ranges (as this can no longer be done purely on a design/validation basis). And I think for most users the performance headroom is pretty much outweighed by the Ryzen OC experience of "put chip on board and stop messing with it, the chip is better at it than you are". (and of course that's looking at it with rose-colored glasses, GPUs like Maxwell, Hawaii, and Polaris were super easy to get huge gains out of, but CPUs have always been a bit of a pain if you wanted to push them to the limit, tuning ringbus and cache clocks has been around for a long time for example, and memory has always been a mess.)
|
|
#
?
Feb 26, 2021 02:31
|
|
|
There seems to be a fair amount of magic and/or voodoo involved in over-clocking these days, so yeah, it would be real nice to have some robust tuning software like that. Maybe there's just too much turnover in mobos/cpus to make such a thing worthwhile, although the AM4 has had a nice run.
|
|
#
?
Feb 26, 2021 04:00
|
|
|
|
| # ? Apr 23, 2024 21:29 |
|
|
shrike82 posted:we can argue about qualitative improvements but that CPU number jumping by 16% is pretty misleading Idk if you have Cyberpunk but while messing with stuff I tried JDEC RAM speeds and I got huge FPS drops running through crowds in the city, putting the memory back to 3800 I saw no drops running through crowds vs doing anything else. Also I've definitely seen at least like 5%+ FPS gains going from XMP to tuned timings testing at 1080 high.
|
|
#
?
Feb 26, 2021 06:35
|
|
