Need assistance regarding secondary timings for a 32GB Patriot Viper III equipped with FX-8350.
Need assistance regarding secondary timings for a 32GB Patriot Viper III equipped with FX-8350.
I'm struggling to get 32GB (a matching kit, 4x8GB) stabilized on a new-to-me build (my PSU fried my old motherboard, when I originally tried to make the switch on it). I knew it would be a devil to switch from 16GB to 32GB on AMD's memory controller but I cannot function with 16GB anymore. I am not a gamer, but a heavy, heavy, heavy multitasker for school. I was willing to sacrifice an OC on both my CPU and RAM and speed in general to make the change, but I cannot seem to fully stabilize the RAM, no matter what I try. I am thinking a small secondary timings tweak is what I'm missing, but I am out of my pay grade dealing with it. Of note: I am desperately trying not to downgrade to a 6.66 multiplier on it.
System:
MB: Gigabyte GA-990FXA-UD3 R5
F2 BIOS (last released)
CPU: FX-8350 with a reduced OC:
BCLK 217 x20 multiplier (4.34 Ghz); I had my old setup at 4.53 with Crucial Ballistix OC'd to 2014 Mhz
COOLING: Cooler Master Seidon 120M water cooler set up on a push-pull configuration and mounted outside my case.
TEMPS @ 70° F ambient running Prime95 blend test
VRM's peaked at 66° C momentarily but stayed 60° C
CPU temps were 46.5°C Max and 38.2°C average
RAM: Patriot Viper III 1600 DDR3
Stock Timings on JEDEC 5 @800Mhz = 9-9-9-24-39 (2T?)/XMP Profile 1 @800Mhz =9-9-9-24-38 3T
@ 1736 Mhz (8.00 multiplier), I can run Prime95 for 12 hours without error at 10-10-10-27-40 2T.
AUTO on all other timings.
NORMS DISABLED for OC, e.g., APM, XMP, HPC MODE, C1, C6, etc. Fans on full for the water cooler.
Voltages:
VCORE 1.45625V (running 1.452V consistently in Prime95, sometimes spiking to 1.464V)
LLC Medium
NB CORE 1.3V (Stock = 1.2V)
NORTHBRIDGE 1.125V (Stock = 1.1V)
DRAM Voltage 1.675V to get it to pull 1.656 consistently. I am afraid to go any higher. The XMP profile on this RAM does pull 1.652V. So, it's rated around that range, but I don't think it would do 1.7V and I am not really comfortable where I have it. (Stock = 1.5V for XMP Disabled and 1.65V for XMP).
AUTO on all other voltages
The Prime95 disclosure does not mean the RAM is stable, and I'm sure I could not hold up for 24hrs. The trick is, my desktop environment is still RAM glitchy and at any number of timings I've tried that I may can run on Prime95 awhile. I want to do away with, for instance, the slow icon reload after the desktop refreshes on startup (and it's a RAM issue, not an icon cache issue, etc.). When the desktop flashes, I want it right back up, as it always is in any stable setup I've ever had. That's just one thing. The RAM is crappy, I'll warrant that, even taking the wonky controller into consideration.
I am thinking a tweak to a secondary or tertiary timing setting might help, but I don't know where to begin and haven't had much luck with what I've tried, e.g., tRFC. So, what do I need to do? Surely on such a mild OC I ought to be able to stabilize the RAM without bumping down. And, to be honest, when I have tried at 1451 Mhz/6.66 it was still glitchy. Granted, I haven't agonized over the timings there, though.
So, since it is not wolves that are killing me but little foxes, what is the tweak I need to make that I'm missing? I don't really have any voltage overhead, I guess, but I was hoping to at least get the RAM stable and then bump my CPU multiplier to 20.5 if I could. The big thing is I'm okay at 4.34 Ghz if I can just get the wonky RAM settled. I have tried researching it, but to no avail in finding a solution.
Can any of you heavy hitters/pros please help? I'll owe you. And thanks.
P.S. I'm not sure if you want other timings I've tried--I've gone much lower--but the above is my baseline that I have run the longest. I can add some of the others if needed. Thanks, again.
Thank you for trying to help. I have been worried about whether the RAM are inherently impossible to work with at 32GB (if not 16GB, too). The trick is, apparently DDR3 can only be bought used now and finding 4x8gb kits is hard--inevitably I went with what I could find. The batch numbers are the same on all four, though, not just the model numbers. I may have been wrong to do so, but I assumed those were a genuine kit (that is how I've always understood you can tell).
My PSU I bought brand new. It is a Thermaltake Smart BM3 850W. I am in grad school full-time (read "jobless") and on a budget, so it is not top of the line, but it was recommended at the price point. School is one reason I did not roll over to Windows 11 in October, i.e., because I did not have funds to spend on compatable hardware for a new build without going cheap-cheap (and I actually won't build again for the full three years MS update Windows 10 with security updates). School is also why having a solid rig is important.
On Memtest86, I ran the RAM already clocked but not on XMP, as I do my OC'ing with XMP disabled. And I ran it on my old build too, not this one (the build I killed switching to 32GB of RAM). I have switched XMP to enabled to see if I could eradicate the desktop environment glitches a time or two on this build, but it does not help and it seems like the RAM are more stable without it. This afternoon I bumped everything down to stock, to see how 1600 mhz and stock timings acted on the RAM, so I will run Memtest again on the stock timings with the stock CPU BCLK et al., and report back. Memtest86 did not post errors on the other attempt but it did post a warning about the hammer row test. I cannot find it in my notes now to quote that warning, but when I looked it up it was explained that Memtest may have found an error in one row that did not carry over to a second row. Thus, it was not a genuine error though it was flagged as a possible hardware issue. From what I read, it was explained that that did not actually mean I would have trouble. So, I was hopeful. I will try also try to find the exact quote and post that when I report back. All my Prime95 runs are with XMP off, including the twelve hours without errors.
On the OC: I was running 4.51 Ghz on 16GB (2x8GB) of Crucial Ballistix 1600 Mhz Low Profile sticks rated 8-8-8-24 at 1.35V. Those were a dream. I could do 4.62 Ghz stable on Prime but did not like the temps being borderline or the amount of voltage I had to be at to achieve that clock, so I pared it back to the 4.51 Ghz. The FSB OC on those RAM was 2005 Mhz, which bumped the timings to 9-9-9-something. My BCLK was 215 on it with a 21 multiplier. And that is where I stared with this RAM: I plugged it in knowing I would have to dial the OC down, as I was hoping not to have to start from scratch (as, ironically, I am now having to do).
The trick is, I didn't have the power on my original PSU (an EVGA 550W) and just did not catch that. Twas a blonde error if ever there was one. That old build is where I tested the RAM and with it jacked up with the BCLK. The thing is, I should have remembered I was at that PSU's threshhold (one reason I pared the original OC back because of voltage), but I forgot. It happens. Again, I've read the stories enough to know this would be a devil of an upgrade, but I did not think it would be
impossible
--that's why I had resigned out of the gate to having to bump down any OC'ing, i.e., as the memory controller could not handle it on 32GB. I'm okay with that--I don't game. I just don't want a snail's pace; I like a snappy desktop.
I will look at the guides before I come back, find the sundries mentioned above, and rerun Memtest at stock values. If this RAM won't work, I'm not sure what I'll do. When I found these there simply weren't a lot of options and the "kits" I was seeing were mismatched bundles, usually two (2x8GB) kits bundled. I was trying to avoid that. "Gloom, despair, and agony on me... whoa-oh-oh..."
Thanks, again, until you're better repaid.
P.S. (Just in case it helps you reconsider the Crucial.)
My issue with 16GB is mainly about browsing. Currently, I use Firefox across six windows with around 2000 tabs open. Not every tab is active, but I keep them all open, and I often let my PC sleep so I can resume work later. Sometimes there can be several hundred tabs running at once. My usage includes school research projects with separate windows for each project, a general browsing area, a shopping window, etc., plus streaming content mainly from YouTube-style lectures. Mostly, it’s school research for my thesis and class papers throughout the semester.
In short, it can really slow things down when I need to close Firefox because memory is running low. This happens whenever other programs are active—like Foxit PDF reader with books open, Word, PowerPoint, a video player, etc. So, having 32GB of RAM is a big advantage. Even if the system isn’t working perfectly, the performance difference during heavy use clearly shows that upgrading was necessary. In fact, I should have done this much earlier; it’s extremely frustrating to pause my work and restart frequently. When you do that, you have to re-enter all your pages, which can be very time-consuming and disruptive.
If all batch numbers are identical, they should generally be compatible initially. However, this doesn't guarantee perfect compatibility, especially with used items. It's possible to encounter issues over time.
I'm not familiar with the Smart series PSU, but it seems to have some drawbacks compared to other models like the TR2. The Smart BM3 hasn't been thoroughly tested by me, though eTechnix reviewed it and found it acceptable but lacking in detailed evaluation.
It's hard to say for sure without more details about the platform. For now, I think it might be a decent option but still worth considering carefully. It could be a budget-friendly choice, but there are concerns.
One stick of the RAM is bad. I have put in a return on them. When I ran Memtest86 the first time, I stopped it just shy of finishing the fourth test, thinking they'd all be the same. Well, the bad RAM was in the fourth slot and I just didn't give it enough time for the row hammer test to post errors. Had I let the test finish, I'd have known sooner. All four posted the "may be vulnerable to row hammer flip bits" spiel, though.
I also misspoke about RAM: Supposedly, you can buy 4x kits new from China still (HyperX was one variety), but I wouldn't trust that they wouldn't just throw four separate RAM in a package and send it, and it is too much of a hassle on returns and waiting to bother. I did find another 2x kit this time of my Tactial LP, which could not be found originally. I know mixing kits may not work, but I have ordered the kit to give it a try with my original pair of LPs. I'll just make sure one kit is in one channel and the other kit is in the other.
But, at the end of the day, the RAM was not just crappy, it was indeed defective. And I'm happy with the PSU. Thankfully, the Smart Fan bit is not automatic, but has to be manually turned on. I'll never use it. Thanks for the help. A lack of patience was my main problem here. If the Tactical LPs don't work, I'll pick up a mixed set of G.Skill, I guess. There are tons of those available in 2x8GBx2 kits. That's just about all there is, two kits paired, just like before.
There are still a few kits available new in the desired configuration if you're willing to pay for them and considering the price of DDR4 and 5 right now, despite them being DDR3, almost makes it not look so bad. LOL.
https://pcpartpicker.com/products/memory...Z=32768004
That's also why I said you need to run for 4 full passes anytime you are testing RAM. I've seen RAM take until the fourth pass of a given test before it started throwing errors so you definitely can't be shortcutting when testing memory.
Thank you for the link. If the Ballistix LP don't work with my original setup, I'll check other options. That was a place I hadn't considered before. Thanks again.