Is DDR5 at 6000 mhz still optimal for Ryzen 9950X memory controller?

What the topic says… :smile:

I’m getting a 9950X + Asus ProArt X670E build, currently was able to get Crucial Pro 48x2 5600mhz kit but still trying to source a 48x2 kit at 6000mhz if you folks would confirm that 6000mhz is till the optimal speed for max CPU memory controller (FSB?) performance?

Thank you!

I just made a similar build and all the information I found indicated that 6000mt/s is the correct out-of-the-box speed optimization.

If you are interested in overclocking and tweaking, there are advantages to be found at 6600mt/s and probably some nosebleed stuff you could do with the 8400 sticks. On the other hand, you may not win the lottery and your processor or motherboard may never work at those speeds. I am happy with the 6000.

1 Like

You could but there is a roll of the dice at higher speeds…
Unless your use case demands the high speeds, better of saving the money and investing elsewhere…

1 Like

Thanks! Good to know.

I’m currently running a stopgap 32x2 kit at 6000mhz but I really wanted 48x2 (preferably black non RGB) although availability is a bitch where I am located (India) so I will likely have to settle for the Crucial Pro 48x2 kit at 5600mhz, hopefully not too much of a perf drop, not that I care too much, I’d rather have the extra 32 gigs.

Thanks again!

I thought 6400 was the easy to do sweet spot for Ryzen 7000? And the memory controller didnt change in 9000 so should be the same speed?

Depends on how much RAM and how many sticks…

Not all CPUs manage 6400. But yeah, highest you can do 1:1 is sweet spot. It’s just that the vast majority can do 6000, most 6200, some 6400 and rare or extremely cooled samples 6600.

If you can reach 8000+ in 2:1, then that becomes faster than 6400 1:1. but this requires the right memory kit and motherboard too.

So 6000 is the best almost guaranteed set it and forget it generally.

3 Likes

+1. Tight 6000 timings edge out 6400 1:1 and 8000 on some benches, 6400 edges out 8000 on others, and…

Looks like 48x2 kits that can do 6000mhz aren’t available, so I’ll be going with 5600mhz. It’s a Crucial Pro set.

1 Like

Where do you live? Do you mean kits that do exactly 6000 or kits that do >=6000?

You can get a 6400 one and run it at 6000…

1 Like

Sometimes its better to buy a faster kit to run it slower, especially with better timings.

That’s broadly my preference, particularly if the cost’s much the same, which is typically what I’ve been seeing lately (6000, 6400, and 6600 2x48 kits within ~3% of each other). Not sure how much awareness there is one can apply the XMP/EXPO profile and then just turn down the clock and maybe the voltage, though.

1 Like

I’m in India, yep I looked for higher 48x2 spec kits (I wanted non-RGB black preferably but RGB is ok) but my main choice of 48x2 capacity (RGB or non) is currently not available, except this 5600 Crucial Pro set.

Observed an approx 5 to 7 % perf drop in a few quick benchmarks and workloads I tried (cpuz was ok, geekbench and linux kernel compile showed these numbers) which I guess is expected :slight_smile:

BTW - if I get a really high spec kit, like, say, 8000 mhz then will the perf in geekbench/kernel compile go up accordingly? Or should I just stick to the 6000 optimal speed/ratio for the memory controller?

Ah in that case just get what’s available. At least the crucial kit is significantly cheaper than the 6000+ kits with hynix dies (where it is available). For most applications there’s no difference and as you said modest drops when it does matter.

AFAIK there’s no 8000MT dual rank kits available. Only single rank (16GB or 24GB modules) and 4 modules won’t hit those speeds either.

1 Like

If you consider like +1-2% with unrestricted PBO to be accordingly, sure. See TechPowerUp and Phoronix scaling data.

Highest 2R UDIMMs I know are Corsair and V-Color’s 2x48 7200s. Micron’s sampling 12800 MRDIMMs but desktop relevance’s over the event horizon.

1 Like

Honestly tighter timings is more important than frequency with Zen4 & Zen5. The CPU cores waste too much time waiting on the memory.