|
Gibbo at ocuk said he's got the parts but bound by nda not to say much except that there's not much difference between the ddr5 memory he has in stock and the faster 8pack ddr4 memory they sell. So basically ddr5 is only a marginal improvement, and the fastest ddr4 can match it for the time being.
|
# ? Oct 14, 2021 11:28 |
|
|
# ? May 22, 2024 13:10 |
|
Zedsdeadbaby posted:Gibbo at ocuk said he's got the parts but bound by nda not to say much except that there's not much difference between the ddr5 memory he has in stock and the faster 8pack ddr4 memory they sell. So basically ddr5 is only a marginal improvement, and the fastest ddr4 can match it for the time being. I guess there's multiple ways you could take that - if he means contextually that ddr5 performs like fast dual-rank quad-channel DDR4 in bandwidth that's great. If he means that it only performs like relatively modestly clocked RAM kit (2 stick kits have better clocks/timings than 4 stick kits, 8 are worse than 4) in latency that's bad. So you can read that rorschach test any way you want. ignoring that whole question for a sec, gibbo's a massive loving shitter who will say anything to make a buck. Like say what you want about him running a shop while also being a member of the tech review community/etc, or his poo poo being overpriced as hell, but someone like der8auer is at least an honest broker and I've never seen him say anything borderline questionable to make a buck. gibbo is a loudmouth huckster whose primary interest is making gibbo lots of moolah. Real "never stop the hustle" type. I don't doubt that ddr5 probably isn't going to be worth it at first, but I'm not forming that opinion on the basis of loving gibbo either. Paul MaudDib fucked around with this message at 17:26 on Oct 14, 2021 |
# ? Oct 14, 2021 16:55 |
|
3600 shits its pants and the 2nd memory channel dies on the CPU itself https://www.youtube.com/watch?v=-xJMFWtasIQ I've personally had 2 3600s that were not stable at stock settings. I had to do a voltage offset to get them to just boot and stay running.
|
# ? Oct 16, 2021 14:32 |
|
redeyes posted:3600 shits its pants and the 2nd memory channel dies on the CPU itself I don't know if he mentions that he cleaned it later because I skipped through to the pins & to see if he gave him the working CPU. Khorne fucked around with this message at 21:07 on Oct 16, 2021 |
# ? Oct 16, 2021 21:01 |
|
Bought a bundle that consisted of 5600x, aorus elite ax v2 and 2x8GB of 3600mhz memory. The system won't boot if I do XMP profile 3600mhz, but it will boot if I do default or XMP profile 3000mhz. I updated bios to F14e, the very latest firmware (which also enabled the rather excellent smart access memory mode). What could I be missing here? When I tried my old 3200mhz memory and its XMP, that worked just fine.
|
# ? Oct 16, 2021 22:04 |
|
Give the memory a little more voltage after setting it to the XMP settings. Try it out in 0.01 steps ie. 1.36, 1.37 etc. Don't worry about the CPU.
|
# ? Oct 17, 2021 00:47 |
|
Zedsdeadbaby posted:Gibbo at ocuk said he's got the parts but bound by nda not to say much except that there's not much difference between the ddr5 memory he has in stock and the faster 8pack ddr4 memory they sell. So basically ddr5 is only a marginal improvement, and the fastest ddr4 can match it for the time being. i think pretty much ddr generation this has been the case tbh. by the time a new generation hits, the old one's been juiced so much there's not much difference until the new one starts getting tight timing/high freq modules
|
# ? Oct 17, 2021 00:57 |
|
OTOH there may be room to push even the launch stuff if you want to do manual tuning. It's impossible to say for sure until we have better information.
|
# ? Oct 17, 2021 00:59 |
|
Icept posted:Give the memory a little more voltage after setting it to the XMP settings. Try it out in 0.01 steps ie. 1.36, 1.37 etc. Don't worry about the CPU. That's DRAM voltage? Very rusty with this aspect. It would make sense - the default is saying 1.20 but the rated voltage is 1.35 edit: I managed to solve the issue by changing the sticks from A1 & B1 to A2 & B2, as well as manually setting the DRAM voltage to 1.300v. It seems to work just great. Cheers for the tip on the voltage thing, I never would have caught that. For some reason it defaulted to 1.200 even in XMP mode. Zedsdeadbaby fucked around with this message at 09:28 on Oct 17, 2021 |
# ? Oct 17, 2021 07:54 |
|
Zedsdeadbaby posted:That's DRAM voltage? Very rusty with this aspect. It would make sense - the default is saying 1.20 but the rated voltage is 1.35 Glad you got it working! And yeah, if you ever get bored one day, you can spend a long time dialling in specific RAM timings/voltages with Ryzen. It's one of the few areas left for the end user to optimize. I haven't really bothered myself, but the main DRAM voltage is the one where you can really run into stability issues if it's too low.
|
# ? Oct 17, 2021 11:51 |
|
Zedsdeadbaby posted:edit: I managed to solve the issue by changing the sticks from A1 & B1 to A2 & B2, as well as manually setting the DRAM voltage to 1.300v. It seems to work just great. Set it to 1.35v if you want to spare yourself the headache later of random bluescreens and forgetting why that may be happening.
|
# ? Oct 18, 2021 04:47 |
|
That's definitely on my to-do list. I will do that next time I start up the PC for sure. I've played a bunch of Destiny and Monster Hunter and no problems so far at least.
|
# ? Oct 18, 2021 08:23 |
|
Say I have four DDR4 sticks with JEDEC 3200 1.2v spec, what are the chances four of them will run in a 5950X, or rather upcoming VCache Zen3, at actually 3200MHz, with tightened CAS at 1.25v? (--edit: they're single rank, if that matters.)
|
# ? Oct 18, 2021 19:43 |
|
Combat Pretzel posted:Say I have four DDR4 sticks with JEDEC 3200 1.2v spec, what are the chances four of them will run in a 5950X, or rather upcoming VCache Zen3, at actually 3200MHz, with tightened CAS at 1.25v? (--edit: they're single rank, if that matters.) Probably pretty good, depending on how tight you try to go with the timings and what type of die it is.
|
# ? Oct 18, 2021 19:47 |
|
They're Micron E-Die, which back then was said to be good. I can't make head or tails of that die stuff. They're CL22 stock, but they're ECC DIMMs, so they probably erred on the right side. I dropped them to CL18 and raised the voltage +0.5v proactively (--edit: doesn't appear to hurt it, the built-in temp sensors say they run pretty "cool" with 36-40°C).
Combat Pretzel fucked around with this message at 19:57 on Oct 18, 2021 |
# ? Oct 18, 2021 19:55 |
|
https://twitter.com/VideoCardz/status/1450873395558690824 Some early ES leaks, seems pretty far along.
|
# ? Oct 20, 2021 20:03 |
|
https://twitter.com/aschilling/status/1451197966379741185?s=21
|
# ? Oct 21, 2021 17:58 |
|
You can thank me for Rembrandt APUs moving toward production, since I decided to buy a 5600G yesterday. I'm finally doing a serious refactoring of my desktop machine, which was 100% made with spare parts until I upgraded it from a 1600 to a 3600 last year. The motherboard is still an A320 model that I picked up open-box at Microcenter right around the time the B450s were released, so I figured it could use some work in the name of reliability. I'm ditching an RX 550 in favor of the APU's graphics, swapping in a spare X570 ITX mobo, and will be using the PCI slot for an NVMe host card. That will let me retire the 2.5" form-factor SATA SSD (whose age I can no longer remember) that all my personal data is currently sitting on, in favor of a new Samsung M.2 drive.
|
# ? Oct 21, 2021 19:55 |
|
It's curious because the latest Win11 insider beta build .282 supposedly fixed that too? Or is this a different issue? quote:We fixed an L3 caching issue that might affect performance in some applications on devices that have AMD Ryzen processors after upgrading to Windows 11 (original release). owls or something fucked around with this message at 21:53 on Oct 21, 2021 |
# ? Oct 21, 2021 21:50 |
|
owls or something posted:It's curious because the latest Win11 insider beta build .282 supposedly fixed that too? Or is this a different issue? Different issue.
|
# ? Oct 21, 2021 22:12 |
|
So this AMD Ryzen Powerplan in Windows, do you need it or don't you? Several releases back, the scheduler in the OS couldn't deal with Ryzen quirks, thus AMD made that plan. Later on, it was said that Microsoft fixed their poo poo and that you don't need it anymore. Now I'm suddenly being told, for those recent issues in Windows 11, I should use the plan again. I have it installed now, but it's not listed as recommended within Windows (I have a Zen+), whereas it appears to do so with Zen2 CPUs. And Zen3 doesn't need any external power plan. What is going on?!
|
# ? Oct 22, 2021 13:22 |
|
Combat Pretzel posted:So this AMD Ryzen Powerplan in Windows, do you need it or don't you? Are you on Windows 10? If so, you can pretty much ignore everything because it's tiny 1-2% changes in performance either way. The Ryzen power plan changes the timings for how fast the CPU changes clock speed, from 15ms to 1ms (on Zen 2). Faster reactions for clock changes is nice, but minor. W10 more or less just works with Ryzen, fuhgeddaboudit. Did you install a day 1 release OS to be an unpaid beta tester? If so, you want the chipset drivers / ryzen power plan because it fixes threads getting sent to weak cores instead of strong cores. OTOH the main W11 problem with Ryzen, where the L3 cache is hosed up, is not solved by a driver / power plan at all. Are you a competitive benchmarker? If so, you want to use this other custom power plan that, when combined with 10 other tweaks, will allow you OC slightly higher and get 2% higher scores on a selection of benchmarks.
|
# ? Oct 22, 2021 14:25 |
|
Both Ryzen bugs in Windows 11 are fixed as of yesterday, the L3 cache and preferred core. There was a window patch and a chipset driver update. W11 performs just as well as W10 for Ryzen now.
|
# ? Oct 22, 2021 16:15 |
|
Yeh, the chipset driver update is what's bothering me. When I tried to install it, it told me that all drivers were current, except the power plan, which I didn't install in the past, on the notion claimed in the past that it wasn't necessary anymore (only for early Win10 releases). I'd have figured the CPPC2 stuff was an operating system thing, not needing external help. But now the recommendation is to use the Ryzen power plan for it to work properly.
|
# ? Oct 22, 2021 18:18 |
|
https://twitter.com/ExecuFix/status/1451221315919171595 so TRX40 is a one-and-done socket I guess...
|
# ? Oct 23, 2021 03:23 |
|
CoolCab posted:ayyyyyy scored a 3600 for my rig and a x570 and cooler to resell for £120 total. six cores here i come and it's installed! i'm quite proud of myself, with that i have now replaced every single part on this machine at least once. it's a 3600 and i remember gamers nexus's steve saying that you could easily achieve 3600x performance with a mild overclock, does anyone know what he was talking about? i have thermal overhead i think, my cooler is a little above spec.
|
# ? Oct 23, 2021 20:40 |
|
CoolCab posted:and it's installed! i'm quite proud of myself, with that i have now replaced every single part on this machine at least once. The last part isn't a replacement but installing all your other parts on a new machine when you think about it.
|
# ? Oct 23, 2021 20:50 |
|
Paul MaudDib posted:so TRX40 is a one-and-done socket I guess... Joins FM1, AM1, and Quad FX/4x4 with that distinction for AMD dekstop.
|
# ? Oct 23, 2021 21:24 |
|
I actually liked AM1 though sure, it wasn’t very good in desktop terms but as a little home server it was OK, and being able to get a cpu and motherboard for $50 changes your level of expectations
|
# ? Oct 23, 2021 21:38 |
|
A 3600x is literally only a few % better than a 3600, so yeah any sort of OC you put on the chip will let you match it.
|
# ? Oct 23, 2021 21:54 |
|
Paul MaudDib posted:I actually liked AM1 though AM1 motherboards had so little on them because the CPUs were a more complete SoC than usual if I recall correctly, they were cool.
|
# ? Oct 23, 2021 22:03 |
|
I went with a 3600X on this machine partly because I figured it gave me at least slightly better odds in the silicon lottery. And I know the conventional wisdom is and has been that a 3600 should be about the same, but at this point I can't help but wonder if that wasn't a little overstated. It's hard to search on a longer-term comparison of benchmarks (and loving Userbenchmark pollutes the search results a lot), but looking at the average benchmarks on CPU-Z the 3600X holds a significant lead, and the 3600X and 3600XT are much closer, more like the distance I imagine the 3600 should be if it really were so easy to make it perform equally to a 3600X. People who install, run, and post results using CPU-Z seem like they are also likely to be the people who tune their machines for performance, so with a year or two of data I would've expected the results to be closer than they are if the differences between the 3600 and 3600X were that trivial.
|
# ? Oct 23, 2021 22:22 |
|
Paul MaudDib posted:so TRX40 is a one-and-done socket I guess... Sucks that prosumers get screwed out of high lane count systems, but alas, what can you do. Of course, maybe he's just wrong.
|
# ? Oct 24, 2021 00:34 |
|
Twerk from Home posted:AM1 motherboards had so little on them because the CPUs were a more complete SoC than usual if I recall correctly, they were cool. it was basically AMD’s counter-product to Atom SOCs and a decent chunk of the market for that is stuff that’s pretty cost-optimized. I’m sure in most of its normal applications it’s a soldered BGA chip, and tbh the socket was a little pointless here too, just give me a 5350 soldered down, done. I know there were other models like E350 that were soldered… For its time and role, it was a neat product - Atom really only got good with Baytrail (I think it’s airmont?), Jaguar was better than the old in-order atom cores. I’m a weirdo but I would actually love to see an Atom built with just tremont or nextmont cores, as well as a server style thing like denverton. Like, skylake perf at atom power levels sounds amazing. If your task threads well then just give me 128 of them on a HEDT socket, like a multi CCD processor of just small cores.
|
# ? Oct 24, 2021 00:43 |
|
I always wanted to play with the embedded Opterons and Epycs.
|
# ? Oct 24, 2021 00:50 |
|
CommieGIR posted:I always wanted to play with the embedded Opterons and Epycs. I was disappointed Intel seems to have let the Xeon-D sort of wither away, they are great little setups. An AMD version would be great!
|
# ? Oct 24, 2021 00:52 |
|
priznat posted:I was disappointed Intel seems to have let the Xeon-D sort of wither away, they are great little setups. An AMD version would be great! Yeah they have the Epyc 3000 series I wanna find a board that has it. This looks fun: https://www.servethehome.com/supermicro-m11sdv-4ct-ln4f-review-low-cost-amd-epyc-3101/
|
# ? Oct 24, 2021 00:55 |
|
Paul MaudDib posted:I actually liked AM1 though Ive been wanting to pick up an AM1 board, I have a certain... uh... affinity for weird sockets. for example, I present my CPU collection for the completely forgotten FM2 platform: Athlon X4 750k, 760k, 860k, 850, and 845. One each of Piledriver, Piledriver+, Steamroller, Steamroller+, and Excavator, architectures people famously love! I finally got my testbench running again so ill prolly run a bunch of benchmarks on these things someday just for fun to really see how bad/good they were in retrospect.
|
# ? Oct 24, 2021 00:57 |
|
FYI to the thread: I installed these newest 3.10 chipset drivers on my W10 machine and started getting consistent system crashes. This is on a X370 board. Not blaming anyone, but AMD might have been rushing these out for the performance fix over other considerations. (On 10 the performance fix was irrelevant, but I noticed that I hadn't updated my chipset drivers for a while. The crash is the hard lock type -- display freeze, no input, looping audio -- pretty typical for a driver induced crash.) edit: and just to confirm, reverting to 2.17.25 has cleared everything up Klyith fucked around with this message at 12:48 on Oct 25, 2021 |
# ? Oct 24, 2021 13:48 |
|
|
# ? May 22, 2024 13:10 |
|
I've seen some loose AM5 schematic about how devices are wired up. 16 PCIe lanes come from the CPU, eight go directly to a PCIe 16x, the other eight go through a "mux", which distributes eight to the first slot and eight to the second. But this isn't really a mux, right? Just a switch, so it's 16x or 8x/8x, but not load balanced on demand (say the device in the second slot idling, raising the first slot with the GPU to 16x intermittently)?
|
# ? Oct 25, 2021 17:02 |