DDR5 6400 CL32 vs 6800 CL32

Melodist · October 30, 2023

Hello there,

thinking about getting a pair of 64 GB Corsair Vengeance RAM.

Deciding between 6400 CL32 vs 6800 CL32.

Is it worth getting the 6800 at this lower latency?

System will be Intel 14th Gen.

filpo · October 30, 2023

3 minutes ago, Melodist said:

Hello there,

thinking about getting a pair of 64 GB Corsair Vengeance RAM.

Deciding between 6400 CL32 vs 6800 CL32.

Is it worth getting the 6800 at this lower latency?

System will be Intel 14th Gen.

Doesn't really matter too much. If you like the higher speed and can spend the money then sure go for that but 6400 is already quite fast and I don't really see a benefit to going faster

YoungBlade · October 30, 2023

If they're the same price, you might as well. But if the DDR5-6800 costs more, I wouldn't bother. The real world performance difference will be within a couple of percent if that.

You can see what Intel 13th gen memory scaling looks like in this video from Hardware Unboxed - it's about a 4% difference between low end and high end memory out of the box. Tuning the timings on a cheap kit gives more of an uplift, if the uplift is even worth it. We're still talking about just a few percent when gaming at 1080p with an RTX 4090.

overbuilt_gaming · October 30, 2023

DDR5 is a marketing scam. Each stick is dual channel. No, each channel is not running at the rated speed, it's half the speed. That's why initially when it was released it performed the same or less as DDR4. I'm sure there is other minuscule advantages of DDR5 but unless you're living in AiDA64 or some other memory bandwidth benchmark, you aren't going to notice. I'm sure they'll be a response of someone going, it's 5%!!!112. Sure, I guess if you can tell a difference between 100fps and 105fps, or 200fps vs 210fps.

If it were me, I would get a good DDR4 motherboard, and a memory kit with tight timings and low latency. Once you get into the upper echelon of PC Parts, you are essentially comparing apples to apples from diminishing returns alone.

ConnectorMan · October 31, 2023

I just bought 96GB of the Corsair 6400MHz 32CL for 7950x. I was able to get it to 83,xxx MB/s Memory Threaded bandwidth (memory bandwidth with multiple threads), the only useful memory test within Passmark Performance Test benchmark, IMO. This is for work number crunching 3D solvers. It tracks with my real world solve times. If I go with JDEC settings of 4800MHz, it shows 60,xxx BW and in fact slows my solve time from 41m to 56m, reasonably close to calculated of 13m. I also compared to Threadripper 3955wx 8CH 128GB box that has way slower memory sticks but eight instead of two, but also a 81,xxx rating and it was only a bit slower proving to me this is reasonably accurate benchmark. It is also where the same memory on Intel 13900k or 14900k can do a bit more BW of 96,xxx MB/s. I have not tested them to confirm real world improvement, but I believe they have a different memory controller that could very well do it. I also had some instability with XMP settings, but reducing the clock to 6000MHz resolved the issue, even with lower voltage and kicking CAS to 30 cycles to maintain the same 10ns delay. Like I said way too many variables in play with memory tuning.

However, for gaming, I agree with above, won't matter to much as most games run in cache, not copying GBs of data around. Hence the X3D CPUs with larger cache help many games. You can data mine Passmark's benchmark data on RAM to see other people's large variation in performance due to lots of variation in MB, CPU, and RAM. For the 48GB sticks and AMD CPU with 2Ch of ram, that is pretty close to state of the art that I can find. My old 5600x for gaming gets 43,xxx MB/s BW but I don't know if that is due to 6C vs 16C or just the memory settings. I'm trying to relay my findings the best I can here, but there are too many variables and opinions. I tried to keep this factual, but opinions, bad data or unknowns always sneak in. Still lots more to learn. I hope it is helpful, if not, oh well.