I work as a bioinformatics contractor, and I am frequently running into difficulty with 128GB RAM just not cutting it. I waste a lot of time dealing with out of memory issues. Realistically, I need >= 384 GB RAM.
I run a 12700k, which is fantastic for its single core performance and because many scripts I run are not optimized for multithreading, this is a priority. I would like to at least match this performance with a platform that can support more RAM, however it seems this is not possible and is also expensive (thinking high-end Xeon W’s or threadripper pros).
Am I missing a blind spot here, or will I have to take a single core hit in order to expand RAM?
I am betting you will take a slight single-core performance hit to get the higher workstation level of RAM.
It depends precisely on what code you are running.
I have an Intel Xeon W7-2495X 24C 48T, W790 ACE, 512GB ECC, RTX-3090 system.
I can run it easily at 4.8GHz single-core or all-core without messing around with Overclocking.
But its standard single-core performance doesn’t compare to the 13900K or even the 12700K.
Multi-core the W7-2495X beats the 13900K with ease. 24 P-Cores.
Since the W2400 and W3400 series are unlocked, you would have to play with overclocking to try to get similar single core performance as the 12700k, and that all essentially comes down to the silicon lottery.
If you really require the RAM memory, you may not have a choice though.
Passmark: 12700K ST 4712 MT 34763
Passmark: 2495X ST 3656 MT 61668
Regarding “expensive”, my W7-2495X system totalled $13,000 CAD.
I am not a TR Pro guru so I don’t know their overclocking potential if any.
FWIW. Looking at the Threadripper Pro line, the 5955WX 16C goes up to 4.0GHz.
It’s Passmark rating are: ST 3292 MT 49546
If you aren’t in a hurry and you can wait 6-12 months, you could see what the upcoming Threadripper 7000 series performance is like or the upcoming new Intel W-Series.
Do you have any idea if your scripts benefit from additional CPU cache? The AMD EPYC CPUs with 3D vcache could give you proportionally more performance even with lower clock speeds while allowing you to exceed 128 GB. Otherwise, the Xeon W CPUs are probably the best alternative to what you have.
The only way to give more RAM to the OS on consumer platforms is caching. For example on Linux zram is popular.
You could get an Intel Optane 280-400GB and designate the entire drive as zram. With Optane you get better random performance then with NAND drives and you don’t have to worry about write cycles.
Interesting results @twin_savage
The W7-2495X is 4.8GHz while the W5-3435X is 4.7GHz, yet your single-thread is higher.
Are you overclocking by chance?
This is my W7-2495X at 4.8GHz CPU-Z bench.
My single-thread value is almost at the 12900K, 94% comparatively, probably so close you couldn’t tell the difference in real-world.
…yeah just a bit, that was 5.1Ghz.
The real hurdle with W790 seems to be the scheduler/power profile doesn’t like ramping up the clock speeds fast enough to hit the really high single core speeds without manual tuning (perhaps this will improve as the platform matures).
Are you sure you can use zram like that? zram is used for creating virtual block devices in RAM, meaning the system thinks it has a block device but it is actually a section of the main memory, but as far as I know you can’t use it the other way around, to extend you main memory?!
