I work at a bioinformatics facility and we purchasing a compute node and are debating the relative merits of a couple of the EPYC cpus to put in a Dell poweredge R7525.
We have a total of 2Tb of RAM in 32x64Gb 3200Mhz RDIMMS irrespective of the CPUs.
We could afford 7763s but we can’t get them in time for our deadline from our supplier (probably) so we are focused on comparing the 75F3 to the 7713.
I’ve looked over the benchmarking results here and which comes out on top seems to depend very much on the specific benchmark but I’m not sure which of these benchmarks might map best to our workloads.
We do quite a lot of sequence alignment of various types (nothing 3rd/gen long read at the moment) as an example we run pipelines like this one for analysing RNA-seq data from nf-core https://nf-co.re/rnaseq/3.0
This is pretty varied using tools written in java, C, C++, perl, & R
Does anyone have any experience with these sorts of workloads on these CPUs, or more generally on the trade-off between clock speed and number of cores in sequencing alignment and related workloads?
7713 all the way. Especially with that much memory. Either will be fine, and probably shred through whatever workload you give it. You’ll either have a good enough SNR that you can do it with any modern processor, or your SNR will be so bad nothing can save you (except maybe trained GPU-accelerated algorithms). If you’re a facility and suspect you’ll be running multiple things in parallel, then the 7713 is a better deal because it’ll let you work on maybe 8-16 individual projects. If you suspect you’ll only be working on 1-2 projects at a time, I doubt either of them will be significantly faster.
I have two 7282s at my disposal. Let me know if you want me to try a particular computation with them.
It’s the less heavily multi-threaded loads like if we use it interactively to explore visualizations of single-cell sequencing data - which can be quite high memory, that’s giving me pause. Maybe I’m just overly impressed by the boost clock speeds on 75F3 but i’m wondering how much of a difference this would really make make on how long things will take to run when working interactively.
So far our demand is not high enough that we are likely to need to run several highly multi-threaded loads simultaneously.
Thanks for the input, we will probably only be working on 1-2 projects at a time which is part of why I’m hesitating to go 7713 but demand may go up over time so would be good to have extra cores.