|
uhhhhahhhhohahhh posted:GN singled out tRFC as making a difference in one of their examples: TheFluff posted:As far as I understand it: RAM is basically a bunch of tiny capacitors, and they leak charge over time. In order for your memory to be persistent, they need to be recharged every so often. tRFC the number of clock cycles the memory controller should use to recharge a cell. While it's recharging it can't be read or written, so reducing tRFC just increases performance in basically every respect. The counterpart setting is tREFI, which is the number of clock cycles between recharges (so you want to increase this). Stock settings were somewhere in the neighborhood of 650 tRFC and 15000 tREFI for me, but check what your board does. Thanks! I remember seeing back when GN posted the video - makes sense that it would have a similar affect on Ryzen. I'll have to see if I can find some good recommended timings for Intel in the vein of the Ryzen timing calculator. It sounds like pushing it too low or tREFI too high can lead to data/OS corruption, which isn't ideal! Eletriarnation posted:It's probably worth mentioning that if you want to use an X470/B450 board without USB EFI flashing and don't have a source for a preflashed board, it's pretty frictionless to just get an older processor like a 2200G from Amazon and then return it for a 100% refund once you're done with the update. I dropped mine off at a Kohl's and the refund came back before I left the store. This is good advice! The integrated graphics makes the 2200G a great choice for troubleshooting, but if you want to go as cheap as possible get a $55 Athlon 200GE. Stickman fucked around with this message at 18:50 on Aug 20, 2019 |
#
?
Aug 20, 2019 18:46
|
|
|
|
| # ? May 7, 2024 11:07 |
|
|
That was released last year so is fine but as a reminder don't go too old (at least check with the BIOS update instructions), some of the Ryzen 3000 BIOS updates drop support for older construction cores. If you go used part shopping make sure it's compatible with both the BIOS you are flashing from and flashing to.
|
|
#
?
Aug 20, 2019 19:13
|
|
|
monsterzero posted:Well, the gamble paid off. I bought a 3600, a flashed Asus x470 Prime Pro, some DDR4-3200 and an Inland 1TB NVMe module at a Micro Center 500 miles from home. All the hair-on-fire reddit posters had me a little worried but other than a 15-second time to first post it was one of the smoothest upgrades I've done. Enable docp, you'll probably want the latest bios if it's not already running it.
|
|
#
?
Aug 20, 2019 19:13
|
|
|
pixaal posted:That was released last year so is fine but as a reminder don't go too old (at least check with the BIOS update instructions), some of the Ryzen 3000 BIOS updates drop support for older construction cores. If you go used part shopping make sure it's compatible with both the BIOS you are flashing from and flashing to. Yeah, I mentioned the 2200G because it seems like a good middle ground which is likely to be supported by any 400-series board at this point. Something like a 2600 would be ideal assuming you have a GPU anyway, but I think those still cost a bit more. It might be overly cautious but the preloaded EFI version on my X470 Taichi was the first version advertising support for Athlon 2xx series so I didn't know if there might be other models which haven't even been updated to support those yet, and like you say anything older than 2000-series especially with construction cores might be too far in the other direction. It's a little bit chicken-and-egg, honestly, since until you can get the board to boot up you don't necessarily know what version it's running to be able to select a compatible CPU. Eletriarnation fucked around with this message at 19:20 on Aug 20, 2019 |
|
#
?
Aug 20, 2019 19:16
|
|
|
GRINDCORE MEGGIDO posted:Enable docp, you'll probably want the latest bios if it's not already running it. Thanks. I couldn't remember what XMP was in AMDese.
|
|
#
?
Aug 20, 2019 19:30
|
|
Boys who love boys
|
Think that's called AMP. But Asus use the xmp profile and call it docp.
|
|
#
?
Aug 20, 2019 19:40
|
|
|
Gigabyte just calls it XMP. It used to be called AMP back when XMP was new and Intel branded. Some manufacturers call it docp.
|
|
#
?
Aug 20, 2019 19:45
|
|
|
Yeah my B450 gigabyte aorus pro wifi just called it XMP. I guess I should actually check what timings that set for a CL 16 3200mhz set of cheapo ram.
|
|
#
?
Aug 20, 2019 19:53
|
|
|
Had another go at 1900 RAM/FCLK, tried with all timings set to auto this time as last time I was pre-tweaking them. Can't get anything above 1800 to even POST. Don't know if I can be bothered attempting it again while moving ProcODT and the RTT_Park settings up one notch at a time to see if I can even get it to the BIOS. Can't get PowerDown mode Disabled stable at all, thought I had it with some tweaks. Played Prey for 2 hours and it was fine, MemTest passed but BFV caused a shutdown within 5 minutes. It's bizarre considering PowerDown is disabled by default, and it's generally suggested that leaving it disabled improves stability. Also 32GB is a fucken pain when MemTesting. I still don't get why I'm hitting 77ns latency though. There are people on the spreadsheet (at 3600 with worse timings than me with better read speed and sub 70ns latency. I'm also seeing AIDA reporting my FSB as <100 and my CPU multiplier at <41, whereas the other screenshots of 3600s i've seen are at 100 and 42. Can't figure it out, don't know if I've hosed a setting somewhere or if i've just got super unlucky with CPU and RAM binning. uhhhhahhhhohahhh fucked around with this message at 23:57 on Aug 21, 2019 |
|
#
?
Aug 21, 2019 23:54
|
|
|
E - are they also running the same board? GRINDCORE MEGGIDO fucked around with this message at 00:42 on Aug 22, 2019 |
|
#
?
Aug 22, 2019 00:40
|
|
|
Steve from TechSpot/Hardware Unboxed posted a video earlier about some testing he did with a bunch of different motherboards using the same 3800X across all of them (14 total, with 12 X570, a B350, and a B450 Tomahawk "Max"), and only 6 of the boards managed to hit the specified "max boost clock" of 4.5 GHz on any core. Some others would technically "round up," but that seems like it's being a bit too lenient on these companies. I think we're just still in the immature BIOS/AGESA stage and it's probably going to be a while longer before things like RAM compatibility and overall stability are at the level they should be across all manufacturers. I'm only running a 3600 and MSI X570-A PRO, but it seems to get up to the specified max boost on at least a couple of cores (like the Hardware Unboxed results for the 3800X). I tried to do some very basic overclocking of my DDR4-3200 to 3600, but the first attempt required a CMOS reset and I didn't want to deal with the hassle for a fairly minor improvement for my current uses, so I just left it at 3200 with timings from the "fast" preset from the DRAM Calculator since that seems fine at the moment (~71-73 ns in AIDA's tests).
|
|
#
?
Aug 22, 2019 07:46
|
|
|
I'm impressed with the Gigabyte X570 I. I tried a bunch of memory settings, some rendering it unbootable, and it just defaulted them, no bios reset required. (Handy considering where the bios reset jumpers are....)
|
|
#
?
Aug 22, 2019 08:19
|
|
|
GRINDCORE MEGGIDO posted:
Hard to tell subtimings if they don't post a screenshot. But I wasn't stable at 3600 with the DRAM Calc fast preset. Even the 3600 C16 should be ~70ns but if I rolled them timings I'd probably be at 80. One guy on the spreadsheet has the same motherboard as me (Aorus Elite) and hit 3600 C14 stable and got 67ns. https://docs.google.com/spreadsheets/d/1dsu9K1Nt_7apHBdiy0MWVPcYjf6nOlr9CtkkfN78tSo/edit?usp=drivesdk
|
|
#
?
Aug 22, 2019 09:24
|
|
|
Grog posted:I'm only running a 3600 and MSI X570-A PRO, but it seems to get up to the specified max boost on at least a couple of cores (like the Hardware Unboxed results for the 3800X) My 3600X on an Asus X570-P boosts to 4267 all core, which I've heard is pretty good. Unfortunately it also boosts to 4267 single core. No combination of PBO, Auto OC, or the Asus OC settings changes that.  I'm not super worried about it but I wouldn't mind if a future update fixes it.Edit: In other weirdness, I've been running a pair of Corsair 3000CL15 (Hynix A-die) at 3200 with all the timings set via XMP. This gets me 73ns in AIDA64. I saw the positive results some people have been getting with manual timings so I grabbed Thaiphoon Burner and the DRAM calculator. The timings from DRAM calculator on manual with the XML import from TB (including a drop to CL14) gets me...76ns. It did up the bandwidth from high 45GB/s to low 46GB/sec. I went back to auto timing on everything except leaving CAS at 14 and it went back to 73ns but kept the small bandwidth gain so I guess I'll just leave it at that. DRAM Calculator's V1 and V2 presets have even looser timings than auto so I haven't bothered trying them. I also tried boosting the FCLK. It's stable at 1900, but latency goes up to 80ns so I'm guessing this board ties UCLK to MCLK and I can't find a setting for it in the BIOS. (Not even under anything that seems like it might be a possible alternate name like Memory Controller Clock, Uncore Clock, IO Die Clock, SOC Clock or whatever) Theris fucked around with this message at 09:56 on Aug 22, 2019 |
|
#
?
Aug 22, 2019 09:45
|
|
|
Grog posted:Steve from TechSpot/Hardware Unboxed posted a video earlier about some testing he did with a bunch of different motherboards using the same 3800X across all of them (14 total, with 12 X570, a B350, and a B450 Tomahawk "Max"), and only 6 of the boards managed to hit the specified "max boost clock" of 4.5 GHz on any core. Some others would technically "round up," but that seems like it's being a bit too lenient on these companies.
|
|
#
?
Aug 22, 2019 10:20
|
|
|
Yeah, I'm hoping that I'll see improvements with future AGESA/BIOS versions. My 3600 went from hitting 4200MHz single threaded and 4GHz all core to 4117MHz/3.9GHz when I updated my X470 Prime from 1.0.0.2 to 1.0.0.3AB.
|
|
#
?
Aug 22, 2019 16:10
|
|
|
I’ll confirm that memory latency has increased with the latest BIOS from AsRock. I was getting ~72ns but now I’m closer to 79 using the same profile. They did something wonky to make things more compatible I think? Either that or the memory training loosened timings a bit
|
|
#
?
Aug 22, 2019 17:08
|
|
|
Zen 2 thermal paste application advice? Specifically a 3900X. Buddy suggested an X with an extra dot over each of the chiplets and the I/O die. (un)fortunately we ran out of thermal paste mid-build and are delayed by several hours 
|
|
#
?
Aug 23, 2019 00:22
|
|
|
I feel like anytime you have a heatspreader touching a smooth block, the best way is going to be a single dead center drop big enough to coat the whole thing (at least touching all four edges, a bit of ooze-out is fine) when it spreads out under pressure. With full coverage and no voids, the topology under the heatspreader doesn't really matter much. If your heatsink has direct-contact heatpipes with significant valleys between, you might get good results from the single-drop method but I'd probably double check and see how it spreads since you're buying more paste anyway. If channels between the pipes are keeping the outer pipes from getting good coverage you could put a tiny bead on those directly. e: All that said, I feel like I've seen an article testing this and like most things thermal paste it doesn't make much difference as long as you aren't doing something idiotic. e2: yeah, might have been a different one but this article indicates that 1) X is best and 2) it doesn't really matter. Eletriarnation fucked around with this message at 00:39 on Aug 23, 2019 |
|
#
?
Aug 23, 2019 00:33
|
|
|
Yeah GamersNexus did a vid on this and IIRC they had the same conclusion, X gave the best spread but from a temps standpoint all of the patterns were similar.
|
|
#
?
Aug 23, 2019 00:46
|
|
|
Yeah, I've seen other tests of various patterns where they all just splooge out into a circular patty of paste under pressure. I guess I'm wondering if that impairs heat transfer, since with Zen 2's design the heat is coming from off-center, so having the edges be uncovered might be a liability. It's going under a heatspreader (Noctua NH-D15) for what it's worth, not a cooler with direct-contact heatpipes.
|
|
#
?
Aug 23, 2019 00:47
|
|
|
The chiplets aren't centered but they're still pretty close, since they and the I/O die fit into the center 50% or so of the package's area. Look up a picture of a delidded one to see for yourself. If you get a circle that hits all four edges of the HSF you'll have both chiplets and the I/O die covered with distance to spare. Still, the X is going to be even better at getting out to the corners so that seems like the way to go.
|
|
#
?
Aug 23, 2019 00:54
|
|
|
Azuren posted:Zen 2 thermal paste application advice? Specifically a 3900X. Buddy suggested an X with an extra dot over each of the chiplets and the I/O die. (un)fortunately we ran out of thermal paste mid-build and are delayed by several hours If you are ultra paranoid, draw a thin line from top to bottom. Otherwise, a normal center dot will do fine.
|
|
#
?
Aug 23, 2019 01:12
|
|
|
I spread a thin, even layer across the whole IHS and then draw the sigil for speed in it.
|
|
#
?
Aug 23, 2019 02:37
|
|
|
How is Powercolor as a brand? The Red devil looks really nice for a slightly above midrange card (or whatever you want to call this market segment)
|
|
#
?
Aug 23, 2019 02:56
|
|
|
monsterzero posted:I spread a thin, even layer across the whole IHS and then draw the sigil for speed in it. I'm finally going to have a computer that can run Morrowind!  On the thin, even layer note, my buddy was supposed to bring over some Kryo Grizzly Xtreme or whatever the hell, the kind you warm up and then spread flat into an even layer across the whole processor. But, he couldn't find it or lent it to our coworker or something, so he comes over this afternoon and we get to building (this is my second build and he's done a shitload, he was excited to see the new Ryzens) and I splooge on an X with a couple dots, using the Noctua branded paste that came with the NH-D15. I over-do it a bit, and buddy says, "Nah that's a bit too much, let me clean it off with isopropyl and I'll redo it cleaner" and after he lays a tiny bit down, the tube runs out So, we're trying again tomorrow after work! Praise the Lord of POSTs.
|
|
#
?
Aug 23, 2019 02:58
|
|
|
The only downside of using "too much" thermal paste is cleaning it up afterwards, as far as performance goes only using too little is a problem (when it fails to cover the entire surface), it is basically impossible to use too much. You could literally coat your entire socket in the stuff and it wouldn't hurt the thermals, but it would be a huge pain to clean up. So always just use enough that you can see a small amount get squeezed out of the edges uniformly and it will give you optimal cooling without being excessively difficult to clean up should you ever need to redo it.
|
|
#
?
Aug 23, 2019 03:00
|
|
|
You know those machines that glaze donuts? I use one of those filled with thermal paste and run the whole motherboard through.
|
|
#
?
Aug 23, 2019 03:04
|
|
|
some people think there is no incorrect way to apply paste, but they are wrong
|
|
#
?
Aug 23, 2019 03:32
|
|
|
Azuren posted:I'm finally going to have a computer that can run Morrowind! Ha, I bought a tube of the fancy youtube guy thermal paste for this build and when I opened up the tube and pulled out the syringe I felt like Derek Zoolander. Good thing I still have every single tube of thermal paste I've ever purchased. Anybody need some Ceramique?
|
|
#
?
Aug 23, 2019 03:37
|
|
|
Paul MaudDib posted:some people think there is no incorrect way to apply paste, but they are wrong They should have used this method.
|
|
#
?
Aug 23, 2019 03:37
|
|
|
|
|
#
?
Aug 23, 2019 03:39
|
|
|
HardOCP used a piece of cheese in place of thermal paste in one article. This is something that’s hard to do wrong with.
|
|
#
?
Aug 23, 2019 04:48
|
|
|
Craptacular! posted:HardOCP used a piece of cheese in place of thermal paste in one article. This is something that’s hard to do wrong with. If it's gouda'nuff for them, it's gouda'nuff for me.
|
|
#
?
Aug 23, 2019 06:23
|
|
|
good god have some self respect
|
|
#
?
Aug 23, 2019 06:49
|
|
|
What? You aren't fondue of cheesy puns?
|
|
#
?
Aug 23, 2019 11:03
|
|
|
More like fond-ew.
|
|
#
?
Aug 23, 2019 14:13
|
|
|
Paul MaudDib posted:some people think there is no incorrect way to apply paste, but they are wrong 
|
|
#
?
Aug 23, 2019 14:54
|
|
|
Well I'm posting this from my new build (3600 and Pulse 5700 XT). This is probably my easiest windows installation ever. It's nice to play Doom 2016 at 150+ FPS without even optimising the CPU and RAM.
|
|
#
?
Aug 23, 2019 15:19
|
|
|
|
| # ? May 7, 2024 11:07 |
|
|
Hellman's Mayonnaise is the best thermal paste, this is known
|
|
#
?
Aug 23, 2019 15:45
|
|