Zen 3 build thread

As a finishing touch there’s a well regarded 12VHPR cable maker selling aftermarket PCIe 5 cables native to my EVGA (and other) power supplies so that adaptor dangling off of the GPU will get tidied up a bit. The 4080S will head into my main workstation where any concerns of waste heat are mitigated due to the ludicrous amount of 140mm airflow in and out of the 7000D. I suppose that after confirming that I like the PCIe 5 cable for the 4090 that itll make sense to get a second one for that machine as it’ll eventually get a 5xxx of a high-end flavor, and fewer junctions equal less waste to connector resistance. It’s not feeling smart to go replacing either 1200W or 1300W EVGA PSUs. They’re good gear.

Carhole said:

IRL the color is a strong chartreuse shifted into the green spectrum a tad.

Click to expand...

That sounds fantastic. I'm slightly envious, and I normally hate LED bling.

malor said:

That sounds fantastic. I'm slightly envious, and I normally hate LED bling.

Click to expand...

It’s really cheap to add some lighting to your PC these days. A couple of fans and a light strip or two and your whole case interior will glow like an aquarium.

Sigh. My generally rock solid Zen 3 build has started "no-posting" with increasing frequency. It did used to do it "once in a blue moon" but now it seems to be at least half the time I turn it on, probably more.

I'm guessing this is probably a memory issue. The memory configuration I have in there is 4x16GB 3600C16 DIMMs, which always been stable when in use but I'm wondering if something is happening somewhere to degrade that initial POST training (which would be very frustruating / disappointing if so). FWIW I have never overclocked this system beyond setting the "XMP" memory profile in the UEFI, altering the PBO power limits on the CPU to a limit my cooling can handle, and reducing core voltages via core optimiser. All other voltages have always been left on Auto, although I'm now wondering if that was a mistake, see for example
View: https://www.reddit.com/r/Amd/comments/z4jfym/intermediate_post_problems_on_x570_try_lowering/
Mine seems to be sitting at 1.1V going on what the UEFI readout is telling me, so doesn't seem excessive.

I could swap to a dual DIMM config given that DDR4 is so cheap just now, but 32GB DIMMs at any speed all seem to be C18, which seems like a bit of a step back from the C16 dual rank stuff I have in there now.

I can see I am a few UEFI and AGESA versions behind the latest so I'll update that, but any other ideas for things I should try welcomed, especially if anyone has a view of what voltages I "should" be seeing.

That's the thing, my system is (and remains, fingers crossed) basically completely stable. It's just the 'POST'ing that's the problem.

I really don't want to have to start twiddling ignorantly with voltages in UEFI, and while there's any number of forum Reddit posts on this the signal to noise ratio is abysmal.

I might just downclock to 3200 and see how that goes. But I'm annoyed to be in this situation at all.

steelghost said:

That's the thing, my system is (and remains, fingers crossed) basically completely stable. It's just the 'POST'ing that's the problem.

I really don't want to have to start twiddling ignorantly with voltages in UEFI, and while there's any number of forum Reddit posts on this the signal to noise ratio is abysmal.

I might just downclock to 3200 and see how that goes. But I'm annoyed to be in this situation at all.

Click to expand...

Pull and re-seat the RAM, dust out the entire rig, reset CMOS, confirm booting at 2666Mhz, then try XMP1, then XMP2 if XMP1 is sharty at 3600. I had a b550 that didn’t like pushing 3600/c16 at XMP1 and bizarrely that’s the same exact profile as XMp2, voltage and all, though perhaps it gooses the vDiMM at post, I dunno. Reboot and let it rip.

Once it posts and you’re aware that drives are healthy and no major hardware changes are happening enable Fast Boot if not already on, or disable fast boot for a full count on boot (don’t turn off SMART monitoring in your OS of choice). Note: I’ve not delved into the Reddit Miasma but if you’re seeing dying VRMs so early it’s maybe a mobo RMA session waiting to happen, but poke it with a stick first.

Reseat everything. Take your OS drive out too if you care about its data integrity and just boot an install thumb or any other clone if you’d like to get into an OS to verify timings, temps, etc with HWInfo64–strongly recommended.

This seems fixable. Also, disable or only at most auto PBO your CPU once you restore normal board power, then work backwards from a working system there. I suspect that your undervoltt first and foremost is showing its limits. Don’t be shy about reseating your CPU when I recommend to re-seat everything.

Edit: double check the seating of your ATX 12V pin makeups. Top left of most boards.

Got a 5800X3D at a bargain price, and this is how I discover whether my TIM application was good:

Yes, I forgot to take the protective film off

With it on, the CPU drew 88W and hit 90c while running through the Akila City gate in Starfield. After fixing that, it drew 91W and maxed out at 72c. Minimum FPS was 56 on mostly high settings and DLSS Balanced @ 1440p, a big improvement over the mid-30's the last time I tried (albeit there has been at least one patch since).

A Handbrake test took 5:50 with the film on, and 5:19 with it off (compared to 7:31 with the 5600X). Still hits 90c, but it takes 40 seconds to get there with the film off instead of instantly. I need to fool with the fan curve a bit as the ones that came with the Thermalright Peerless Assassin get too loud for my tastes above 1200 RPM.

What a nice CPU. I would’ve gotten one for the shop build if it would’ve offered anything meaningful for 4K everything, but went with the four additional cores of the 5900X for about thirty bucks less IIRC, though very closely priced the 5800X3D was still fetching a decent premium earlier this year. I do admit to not being sure that I made the correct decision seeing as how I will likely not render much or ever load up all twelve cores out there unless game engines make a massive leap in SMP over the next year or four. Congrats on yours, I bet that you’ll have seamless loading of very large game levels as well.

grommit! said:

A Handbrake test took 5:50 with the film on, and 5:19 with it off (compared to 7:31 with the 5600X). Still hits 90c, but it takes 40 seconds to get there with the film off instead of instantly. I need to fool with the fan curve a bit as the ones that came with the Thermalright Peerless Assassin get too loud for my tastes above 1200 RPM.

Click to expand...

Went looking to see if I could reduce the 90c temps by lowering the power limit, even though there's no eco mode. Turns out you can.

Package Power Limit Control is found in AMD CBS > NBIO Common Options > SMU Common Options > Package Power Limit Control.

Click to expand...

Set mine to 115W, and the Handbrake test goes from 5:19 to 5:29 while losing about 500MHz (all-core). Most importantly, it stays around 85c with the fans @ 1400RPM. Still boosts to 4.5GHz on lightly threaded loads too.

grommit! said:

Went looking to see if I could reduce the 90c temps by lowering the power limit, even though there's no eco mode. Turns out you can.

Set mine to 115W, and the Handbrake test goes from 5:19 to 5:29 while losing about 500MHz (all-core). Most importantly, it stays around 85c with the fans @ 1400RPM. Still boosts to 4.5GHz on lightly threaded loads too.

Click to expand...

That’s efficient!

Tho, after just recently trying a premium Dark Rock Pro 5 and finding it deafening after AIO builds, if temps and case noise really matter that much save up a hundred bucks and put an Arctic Freezer 240, 360, or 280 in there (their even larger radiators just won’t make a difference) and be done with both heat and noise. They’re very much worth the small expense overhead of a good air cooler. Maybe keep an older air-cooled sink around as a backup if say, an RMA need for the AIO arises around year four as a just-in-case? They can theoretically lose some coolant via exchange through the tubing walls very slowly, but I’d really not be concerned about it. Set the fan profile down slow and you’ll still get desktop temps in the 30s, and probably 70s while transcoding.use the AMD offset mounts. It drops temps even more and you will likely see that all-core PBO just goes fast without tweaking anything.

Nah, I'll stick with this for now as it was meant to be a cheap upgrade. The fans only need to spin that fast when transcoding. They can keep it in the mid-30's @ 900 RPM, while gaming increases that to around 1100RPM, at which point the GPU fans are louder.

I might give the old Scythe Mugen 5 a try for shits and giggles, but suspect it won't be able to handle the transcoding heat load.

@Carhole I consider an AIO as cost of business in my personal builds. To each their own, but I'm OK with replacing it on average every 5 years. $40/year for cooling I can absolutely tolerate, YES. I have recently built two machines with good air coolers and after playing with fan curves a lot, it got to a decent level but never to "AIO quiet".

hansmuff said:

@Carhole I consider an AIO as cost of business in my personal builds. To each their own, but I'm OK with replacing it on average every 5 years. $40/year for cooling I can absolutely tolerate, YES. I have recently built two machines with good air coolers and after playing with fan curves a lot, it got to a decent level but never to "AIO quiet".

Click to expand...

And on the efficient Zen 3 cores, even a 240mm from Arctic seems to work incredibly well and my shop temps are dry 70°F or so. Eighty bucks. It was cheaper than the Dark Rock Pro 5.

following Carhole I picked a new 5950X to be added to various other Zen3 boxes I have kicking around; in part because they were 'cheap', part uncertainty about Zen4 and the chipsets and part an irrational fear because I'm doing esoteric cluster things and worry about VM migration issues.

Paired it to some noctua tall cooler (all my other boxes are AIO) and to be fair to it it's quiet so far. My only complaint is that X570 boards are basically impossible to come by now this side of the pond and I wanted as much PCIE/m.2 as I could get. Like Amazon had precisely one board left in stock and the smaller retailers were all out/not listing them. Glad I did when I did...

ETA: oh one other complaint: the mobo (MSI MAG X570S TORPEDO MAX) wanted two sets of power for the CPU, the normal 8 pin ATX and then an additional 4 pin (you know how the 8pin is composed of two 4-pins, well, yeah one of them in addition to the 8). I had enough parts kicking about but meant I had to shuffle a bunch of stuff around in other boxes to get that elusive cable free.

wobblytickle said:

following Carhole I picked a new 5950X to be added to various other Zen3 boxes I have kicking around; in part because they were 'cheap', part uncertainty about Zen4 and the chipsets and part an irrational fear because I'm doing esoteric cluster things and worry about VM migration issues.

Paired it to some noctua tall cooler (all my other boxes are AIO) and to be fair to it it's quiet so far. My only complaint is that X570 boards are basically impossible to come by now this side of the pond and I wanted as much PCIE/m.2 as I could get. Like Amazon had precisely one board left in stock and the smaller retailers were all out/not listing them. Glad I did when I did...

ETA: oh one other complaint: the mobo (MSI MAG X570S TORPEDO MAX) wanted two sets of power for the CPU, the normal 8 pin ATX and then an additional 4 pin (you know how the 8pin is composed of two 4-pins, well, yeah one of them in addition to the 8). I had enough parts kicking about but meant I had to shuffle a bunch of stuff around in other boxes to get that elusive cable free.

Click to expand...

I’m still not even quite recovered enough from surgery to put another clean Win11 install on my new 5900x shop PC. Once my right arm decides to play along I think that’ll be the time as gaming left-handed prooved a non-starter.

Love my 5950X. In regular desktop usage I cannot tell the difference between it and a 5800X (or 5900X) but once the heavy multitasking comes into play the 5950X is hard to bog down. It seems like there’s always an available thread left for Teh Snappay.

Good job grabbing a Zen 3 component set before they were all gone.

Yup, 5950X here as well and it's hard to beat it down. I have, a while ago inspired by someone here to actually disable CCD2 and run my workloads just to realize that hey, by themselves, yes, 8 cores satisfy most of my workloads perfectly well. Thing is, oftentimes multiple things overlap, I have to restore huge databases while debugging code, compress/decompress things that come in, have 50 things running and then the 16c shines.

But I admit I could live with 8. I just don't wanna, for those glorious moments where everything fires on all cylinders

And of course there's always the lingering "well what if something comes along that I want/need that scales with all cores" and that really gets me. Has that happened, well, I fired up Star Citizen now and again and it ate 30GB RAM (I have 64) and every bit of CPU and GPU. Too bad it's not really an enjoyable game yet.

@hansmuff I hear that Star Citizen will go gold any year now. Fire up the cores!