Server:Lenovo SR590
CPU:Intel 6226R,full core 3.6Ghz
Memory:512GB 2133RECC
I am currently DDR4 RECC 2133, but the seller said that it can be replaced with DDR4 2933/3200, which requires extra money.
I don’t know if the bladebit depends on the memory frequency because I don’t have the condition to test.
Does someone know about this situation?
Thanks a lot.
A Higher Frequency in general means that the Memory is faster. But there are some caviats. For example, 2 bars of “slow” memory may be faster than one bar of fast memory.
Also I would evaluate how large of a farm you want to build. If you only want to plot 200TB for example, slower hardware might be well sufficient, it only takes a little longer…
Does the seller offer upgrade-pricing (so only the difference) or does ‘replace’ mean you have to pay full price and keep the 2133?
For full price I would never consider it, 512GB must be pretty expensive in Registered ECC.
Depends on your wallet/wishes I guess.
In general higher clockfreqency means faster but also check the CL (latency) specs.
2933/2133 would suggest an 37.5% speed increase, but with higher latency (for instance CL21 vs CL17) the real speed difference would be 2133/17 vs 2933/21 so only 11.5%.
Must be nice to have a bladebit capable machine 
2pb for now.
At present, each plot takes 8 minutes, 180 per day. If I upgrade to 3200 (2933 and 3200 are same price), assuming that each plot can be faster by 1 minute, then there are 205 per day, 25 more.
It is the upgrade fee, make up the difference, and 2133 will be returned.Theoretically, the frequency is increased by 37.5%. Is it actually that high? (Assuming the same latency)
It is amazing how your single xeon gold performs, even with slow memory.
My system 2x E5-2683v4 (16c) with DDR4 2400 spits out a plot in 9 minutes. Yours is total of 32 threads performs better than mine total of 64 threads.
Must be the clock speed and generational improvements. I think it will be able to take advantage of the faster memory.
no no no,
I’m 2 cpus.
6226 * 2, I forgot that…
However, your configuration gave me a big hint, 8min (with the chia plotter mover,without this ,1 min faster) is only 1 minute faster than you, maybe the L3 cache may play a huge role.
by the way,I’m using centos stream 9.
Not related to memory but bladebit works best on RHEL such as Centos.