Which X570 board for 128GB?

[msroadkill612]

Came here to reply about this issue.
My machine was super unstable with 4 x 32GB Corsair Vengance PRO RGB 3200 running at XMP. At first I thought I had a bad processor or motherboard.. After bumping with this thread I clocked down the memory based on XMP to 2932 and was able to attain stability.
Today I've got another BSOD so even at 2932 I do not think it's possible to get a fully stable system with 4 x 32GB. I may lower it even further to 2666.

I was really hoping to have a good setup but so far it seems I should have gone with the TR instead.

I have a Ryzen 9 3950 @ 3.95 fixed clock, OC per CCX to 3950/3850/3950/3950 at 1.15v running on Gigabyte X570 Master

Everything appears to be perfect except for the RAM.

I use it for Photoshop/Indesign/Illustrator. I also have a 2 x MP600 using SoftRAID (very cool software tool) since AMD RAID is basically useless with the custom driver. I also have a RAMdisk and a Cachedisk to speedup writing of projects on RAM first.

The big question for x570 guys like u, is can u live w/ an 8 lane GPU?

There seem to be many gamers who say u can.

IF SO & u have 2x 8 lane slots on ur mobo AND one x8 slot can be bifurcated in bios to x4x4, THEN, a 16 lane quad m.2 port adapter from asus/asrock/... can be run at 8 lanes & yield 2x nvme ports, for a total of 3x true nvme on board.

These 3 nvme ports can be used for a fast raid 0 array for the page file to virtualise supplemental pseudo memory- maybe 4x nvme if the more latent chipset's m.2 doesnt overly degrade the triple array's perf.

since sustained write speed (& endurance) is the key metric for this, pcie 3 Samsung 570 pro's superior MLC NAND makes it the best choice ATM (tho better pcie 4 options seem just around the corner) - at 2.7GB/s sustained write - (x3 = 7.1GB/s write (~9.8 GB/s if using the chipset's fourth nvme port works out) & more for reads).

500GB 570 pro are $150 each atm.

 

[TheSlySyl]

The big question for x570 guys like u, is can u live w/ an 8 lane GPU?

I have x570 Taichi, 16 lanes dedicated to NVME (2 gen3 ssds at 4 lanes each and 2 gen4 ssds also at 4 lanes each) and 8 lanes dedicated to my GPU (1080ti, gen3) and yes. It's been perfectly fine.

My WD SN850 was benchmarking at 7500+ MBPS speeds, so there's no degradation in speed for the Gen4 NVME, and my 1080Ti benches in the top 5% of 1080Tis (it's got one hell of an overclock.) so there's no speed loss there either.

 

[msroadkill612]

I have x570 Taichi, 16 lanes dedicated to NVME (2 gen3 ssds at 4 lanes each and 2 gen4 ssds also at 4 lanes each) and 8 lanes dedicated to my GPU (1080ti, gen3) and yes. It's been perfectly fine.

My WD SN850 was benchmarking at 7500+ MBPS speeds, so there's no degradation in speed for the Gen4 NVME, and my 1080Ti benches in the top 5% of 1080Tis (it's got one hell of an overclock.) so there's no speed loss there either.

Ta. Yes, I know 2.7GB/s sustained write seems slow, but write is what counts for page files, & pcie 3 970 pro remains the best for now, tho not for long IMO, as I said.

By "guys like u", i meant editors using the usual demanding editing apps (adobe etc.) and really wishing for more "memory" than the fast/cheap 64GB sweet spot on AM4 x570.

The only effect of 8 fewer gpu lanes, should be the cpu to gpu pcie link bandwidth, & I had little fear it would matter - i am just not an editor & dont KNOW from experience. If using a pcie 4 gpu, we can be almost certain my question is rhetorical.

The big perceived catch, is that even with the hoped for new nvme drive's sustained write speeds, the arrays speed & latency is woefully slow vs ram.

This perception may ~mislead tho.

The swap memory should be considered a useful SUPPLEMENT to a tiered pool of cache.

Managed intelligently by the OS, it could cost effectively simulate a substantial increase (20-40%?) in available ram for peak loads, with acceptable degradation in perf.

It really depends on exactly what data memory is paging in and out. If it is chunks of static application code vs constantly changing video footage e.g.

W/ the former, it is only reading; undeleted pre written pages into ram - not constantly rewriting pages - a big difference. As u say - 7GB/s per drive is very doable for this & reading has no effect on endurance

At $150 per 500GB 970 pro & durability issues?, the economics may not seem too "flash" ???, but the above is just one approach to consider.

There are good drives like the new sk hynix gold costing half that, & a large fast array could have other uses like as a fast work space e.g.

 

[TheSlySyl]

Ta. Yes, I know 2.7GB/s sustained write seems slow, but write is what counts for page files, & pcie 3 970 pro remains the best for now, tho not for long IMO, as I said.

By "guys like u", i meant editors using the usual demanding editing apps (adobe etc.) and really wishing for more "memory" than the fast/cheap 64GB sweet spot on AM4 x570.

I'm using 64GB of RAM and I also spend a LOT of time on my computer transcoding/editing, Even as fast as my WD SN 850 is, doing anything straight to RAM is much, much faster. I use about 1/3 of my 64GB ram as a cache, mostly a diskcache for read speeds. I use Primocache for that.