|
Craptacular! posted:If it doesn’t work, you can buy M2 to SATA boards on eBay, AliExpress, etc and use it that way. You don’t get the minimalist satisfaction of so much storage being a little add-in board screwed into your motherboard, but you can still use the drive in most cases. They’re a couple bucks at most and beat changing components over this. The M.2 to SATA boards that I am familiar with won't work with M.2 NVMe drives like the 970 Evo, since they are just passive adapters intended for M.2 SATA boards and not protocol converters. The equivalent would be a PCIe x4->M.2 card and I'm not sure if that would meet with any more success if we're talking about an EFI issue. That said, it seems like it would be pretty weird for the drive to not work. I didn't even check the compatibility list for my board because I figured that's the point of advertising support for a standard protocol.
|
![]() |
|
![]()
|
# ? May 29, 2024 17:58 |
|
That sounds very similar to a problem my brother had with his asrock mobo, and it was fixed with a bios update. I recommend installing the updated bios using the built-in flashing utility in the bios, since the new bios wouldn't install when we tried using a windows utility to do it.
|
![]() |
|
Looks like the new Threadripper chipsets will be branded as TRX40, TRX80, and WRX80. Which is good, because that "we will just use X399 before Intel gets there lol" was dumb as hell. Seems kinda obvious with the naming that the TRX40 is going to stay quad channel with the small IO die (and probably a 32 core cap?), while TRX80/WRX80 are full octo channel with the big IO die. Also bet that the difference between TRX80 and WRX80 is more PCIe lanes. https://videocardz.com/newz/asus-preparing-two-amd-trx40-motherboards
|
![]() |
|
Given the projected cost of the 3950X, I sure hope you'll get at least 24C for a 1000 bux on the Threadrippers. --edit: Well, either that, or a 16C spread over 4 compute dies and all their full cache. Also, AMD should finally divulge what's coming. I'm antsy as gently caress. Combat Pretzel fucked around with this message at 20:03 on Aug 29, 2019 |
![]() |
|
Yeah I’ve got huge blue balls over here waiting for the info on the 3950X and the new Threadrippers. One of those chips is going into my next machine, and I’d really like some info so I can start planning a build, preferably before Black Friday.
|
![]() |
|
One of those is going into my next machine, and the sooner DDR4-3600 unregistered ECC can get here the better.
|
![]() |
|
Have there been high speed ECC UDIMM modules announced?
|
![]() |
|
Samsung has sticks up to at least 3200
|
![]() |
|
Actually never mind, I can’t find that. But I could have sworn there was at least one kit at 2933 or something. Awful app is messed up so I can’t edit, sorry for the double post.
|
![]() |
|
Micron has a datasheet out for 3200. https://www.micron.com/-/media/client/global/documents/products/data-sheet/modules/unbuffered_dimm/ddr4/asf18c2gx72az.pdf![]()
|
![]() |
|
CL22 at 3200? Meh.
|
![]() |
|
You can limit SMT implementation in the BIOS right? I could see the WRX80 chipset, if Zen 3 is SMT4, being the only one allowed to scale to SMT3-4 at all as further product segmentation. So a 4990X is 64C/128T in TRX80 but 64C/256T in WRX80, besides professional level validation for WRX80 boards/bioses and more PCIE lanes. A little WMG, but maybe TRX40 is supposed to exist in the same space as X570, as some kind of budget/entry workstation HEDT platform? Like in theory how low could you push a quad channel board in price?
|
![]() |
|
Ecc at non jedec specs is stupid. The people who care about ecc aren't going to run it out of spec to tempt fate. And ecc is 9/8 more expensive naturally.
|
![]() |
|
If they can make modules run at up to 3600MHz reliably enough, why not just put out ECC modules with these speeds? For workstations and whatever other scientific/HPC applications that need bandwidth.
|
![]() |
|
Combat Pretzel posted:If they can make modules run at up to 3600MHz reliably enough, why not just put out ECC modules with these speeds? For workstations and whatever other scientific/HPC applications that need bandwidth. They likely can't at strict jedec requirements around voltage, reliability and signal integrity Malcolm XML fucked around with this message at 05:43 on Aug 30, 2019 |
![]() |
|
Malcolm XML posted:Ecc at non jedec specs is stupid. The people who care about ecc aren't going to run it out of spec to tempt fate. And ecc is 9/8 more expensive naturally. JEDEC guaranteed speeds are only slow if you don't regard the XMP "spec" that borderlines on false advertising.
|
![]() |
|
Cygni posted:If you bought a FX-8120, FX-8150, FX-8320, FX-8350, FX-8370, FX-9370, or FX-9590 from AMD.com or at a retailer in California, you may be able to get some money from a false advertising class action settlement regarding Bulldozer having 8 "cores". I wonder if Newegg counts as "at a retailer in California." My FX-8350 got fried a year and a half ago but I'll print out the receipt for it for $35
|
![]() |
|
Combat Pretzel posted:If they can make modules run at up to 3600MHz reliably enough, why not just put out ECC modules with these speeds? For workstations and whatever other scientific/HPC applications that need bandwidth. I think Threadripper and Ryzen may be the only product that actually supports ECC XMP/DOCP. I doubt the Intel C2xx series supports it, I super doubt their server chipset supports it. I remember reading at one point that Epyc doesn't even support UDIMMs, not sure if that's true or not. High-binned ECC UDIMMs are a super niche product and nobody ever really considered that was a product worth producing before about a year ago. The Wendell callout around then probably helped that. Before that there was straight-up nothing... like 2666 and that was it.
|
![]() |
|
Uh, where does anyone find faster than 2666? It's still all I can find, so nothing really changed.
|
![]() |
|
I've been playing around with the Ryzen Calculator. So far no luck even getting past the BIOS. If I am unable to get those RAM timings to take, is it better just to run in coupled mode at 3733Mhz rather than mismatching RAM and IF speeds due to decoupling? My understanding is you can get better performance if you run your RAM faster than FCLK but you'll only reap the benefits if the timings are tuned. Is that about right?
|
![]() |
|
Combat Pretzel posted:Uh, where does anyone find faster than 2666? It's still all I can find, so nothing really changed. Amazon? It may vary depending on where you live, but they seem to have a large stock of everything.
|
![]() |
|
Combat Pretzel posted:Uh, where does anyone find faster than 2666? It's still all I can find, so nothing really changed. https://pcpartpicker.com/product/MTsmP6/crucial-32-gb-1-x-32-gb-registered-ddr4-3200-memory-ct32g4rfd432a https://pcpartpicker.com/product/hym323/crucial-16-gb-1-x-16-gb-registered-ddr4-3200-memory-ct16g4rfd832a
|
![]() |
|
fuckpot posted:I've been playing around with the Ryzen Calculator. So far no luck even getting past the BIOS. If I am unable to get those RAM timings to take, is it better just to run in coupled mode at 3733Mhz rather than mismatching RAM and IF speeds due to decoupling? My understanding is you can get better performance if you run your RAM faster than FCLK but you'll only reap the benefits if the timings are tuned. Is that about right? https://www.youtube.com/watch?v=xSAFs-t0DCg&t=780s I found the whole video pretty informative, but the TL;DW of it all (which I timestamped to) is: If (FCLK - UCLK) > 166MHz then its better to run your FCLK at its best speed.
|
![]() |
|
Rexxed posted:These are all I can find via pcpartpicker but they seem to be in stock:
|
![]() |
|
Combat Pretzel posted:Those are RDIMMs not UDIMMs. That's the crucial difference. Fast UDIMMs seem non-existent. Oh sorry, misread your original post. Yeah, I can't find any either. Maybe when the base ram speed of the next chipset supports it they'll appear.
|
![]() |
|
I’m jizzin’ for Ryzen!
|
![]() |
|
fuckpot posted:I've been playing around with the Ryzen Calculator. So far no luck even getting past the BIOS. If I am unable to get those RAM timings to take, is it better just to run in coupled mode at 3733Mhz rather than mismatching RAM and IF speeds due to decoupling? My understanding is you can get better performance if you run your RAM faster than FCLK but you'll only reap the benefits if the timings are tuned. Is that about right? There is no guarantee that every Zen2 chip can run at any IF clock over 1800. If you can, you got lucky. I got unlucky on both fronts, CPU and RAM. It won't even pass the DRAM training stage with anything above 1800 FCLK, even with stock RAM settings and any combination of voltages or ProcODTs, and I get instant memory errors on my e-die using even the new Safe timings on the calculator at 3600, I still have to relax them a little bit.
|
![]() |
|
I've seen a couple of people complain that their infinity fabric won't even go over 1600, and apparently that is all that AMD will guarantee hence the ddr non-OC being 3200. It probably has to be a terrible chip, but it is possible it seems to get them, even on 3900X chips.
|
![]() |
|
uhhhhahhhhohahhh posted:There is no guarantee that every Zen2 chip can run at any IF clock over 1800. If you can, you got lucky. I got unlucky on both fronts, CPU and RAM. It won't even pass the DRAM training stage with anything above 1800 FCLK, even with stock RAM settings and any combination of voltages or ProcODTs, and I get instant memory errors on my e-die using even the new Safe timings on the calculator at 3600, I still have to relax them a little bit. Well, if you makes you feel any better, even my samsung b-die didn't like safe settings. Though it's G.skill which I heard is thoroughly binned already. Different topic, I heard that one big guy on LTT mention that HWinfo might not show cpu's idling down, but I could have sworn I saw them go down into the low 3k mhz before. Now they seems to just sit at 4.1k roughly. Even if I go back to default settings on the Mobo. Is there any issues to be worried about trying Ryzen master? I thought I read it messing something up when it was uninstalled. Golluk fucked around with this message at 23:34 on Aug 30, 2019 |
![]() |
|
If you can decouple the IF clock from the DRAM one, why isn't it just running at 1800MHz all the time?
|
![]() |
|
Combat Pretzel posted:If you can decouple the IF clock from the DRAM one, why isn't it just running at 1800MHz all the time? I'd imagine the same reason it isn't running at 1900MHz all the time, not every chip can do it. And if you run memory higher than 1600MHz, it's better to keep at 1:1 than go go 1800MHz. Decoupling adds about 10ns of latency.
|
![]() |
|
Finished upgrading a friend's old 4790K rig to 3600 + MSI B450 Mortar board. 2x16GB DDR4 3200C16 worked like a charm from the get go after setting it in the BIOS. This thing is a beast at CS:GO with 31 bots enabled, gets around 5 more FPS on average than my stock 8700K. Seriously impressed by AMD with my first hands-on experience with Zen 2 CPUs, I'm more convinced than ever that nobody should take the 9400F instead of the 3600 for any purpose.
|
![]() |
|
3600 with new safe timings and PowerDownMode off passed 1000% in MemTest helper ![]()
|
![]() |
|
uhhhhahhhhohahhh posted:3600 with new safe timings and PowerDownMode off passed 1000% in MemTest helper Did you enter all the values? Like the Termination Block ohm and CAD_BUS Block ohm settings? OK, yeah. Just downloaded the latest version. Now safe timings isn't recommending CL14, but CL16. Golluk fucked around with this message at 17:13 on Aug 31, 2019 |
![]() |
|
Golluk posted:Did you enter all the values? Like the Termination Block ohm and CAD_BUS Block ohm settings? Yep, everything on the front screen except BGS, which I can't find in my BIOS
|
![]() |
|
Dumb question time: When my MSI X470 Carbon Pro shows me a colorful MSI logo with the spinning circle of dots from Windows below, is that still a BIOS animation or is it actually loading Windows already? Did MSI switch out the Windows logo in thr loading screen when I installed the board drivers? This is taking 7-10 seconds, then the logo vanishes and the spinning dots come back for 1-2 seconds and I'm on my desktop. I'm using a PCIe M.2 drive.
|
![]() |
|
mcbexx posted:Dumb question time: When my MSI X470 Carbon Pro shows me a colorful MSI logo with the spinning circle of dots from Windows below, is that still a BIOS animation or is it actually loading Windows already? Did MSI switch out the Windows logo in thr loading screen when I installed the board drivers? I have an MSI X570 board and that's exactly what they do, they replace the Windows load logo with a custom one. Annoyed me at first because I kept wondering why I couldn't get into BIOS with the custom logo up.
|
![]() |
|
Pittsburgh Fentanyl Cloud posted:I have an MSI X570 board and that's exactly what they do, they replace the Windows load logo with a custom one. Annoyed me at first because I kept wondering why I couldn't get into BIOS with the custom logo up. Hmmm, not that it matters much, but aren't these some lovely loading times for an M.2 drive (and a fairly fresh Windows install) then? I assumed the new Bioses for Zen2 were not optimised yet/kind of slow and this was actually a custom BIOS splash screen.
|
![]() |
|
mcbexx posted:Hmmm, not that it matters much, but aren't these some lovely loading times for an M.2 drive (and a fairly fresh Windows install) then? I assumed the new Bioses for Zen2 were not optimised yet/kind of slow and this was actually a custom BIOS splash screen. I'm on an x570 a pro, and I also get the custom BIOS splash screen, and I also get the lovely load time on boot. I think it must be something odd about the BIOS, because it takes almost twice as long to boot as my old i7 4770k on the exact same drive. Once it's booted the performance is better, of course, but it's a bit weird that my boot time is noticeably slower with a much faster 3900x in there.
|
![]() |
|
![]()
|
# ? May 29, 2024 17:58 |
|
I get the same longer boot with a 3700x in an x470. Back to normal (slightly faster, by a few seconds) with a 2700x in. Both are slower then my 6600k. (Csm is off, everything set to fastest boot, unused stuff like LAN is disabled) Something is fucky in bioses I think. GRINDCORE MEGGIDO fucked around with this message at 13:10 on Sep 1, 2019 |
![]() |