What’s the fastest processor for single threaded/single process of running sha256sum?
1 Like
Probably the Intel i9-10900k or the lesser core 10th gen mainstream desktops might do it too with the right RAM oc etc
1 Like
Is there benchmarking data available that you can share that would lead you to that conclusion so that I can take a look at that as well?
Thank you.
cant you opengl that and use gpus? i remembering reading of some people doing that with gpus to break every possible windows pass in under a minute years and years ago
1 Like
Single core price/perf probably the Ryzen 3 3300X. Might be able to push it a little as well. But I’m basing that strictly on CB R20 scores.
1 Like
Intel has held an advantage in clock speed and IPC (instructions per clock) for the better part of a decade. AMD’s Ryzen architecture, introduced a few years ago now, is much stronger in heavily-threaded workloads and is close to parity in lightly-threaded workloads; however, AMD still cannot match Intel’s top-end single-threaded performance.
Benchmarks for pure, single-threaded workloads are rare, since it’s not a very common use-case. Games and productivity software tends to benefit from additional cores, up to a certain point (even Photoshop these days!). But if you compare a “pure” multi-threaded benchmark, like code compilation, to lightly-threaded benchmark, like Photoshop, it’s clear which way the performance is leaning.
Intel just released a new series of CPUs—the so-called “10-series” (aka Comet Lake). All the chips are very similar in all respects except for core count. The i9-10900k has 10 cores/20 threads, the i9-10700k has 8 cores/16 threads, and the 10600k has 6c/12t. Single-core performance should be very similar, especially with a manual overclock. Memory tuning has a much larger impact than expected, and in some cases pushes the 10600k past the 10900k—although, the 10900k might reclaim its crown if it received the same type of memory tuning.
So, if money is no object, buy Intel’s top-of-the-line and overclock it. If money is even the slightest concern, get the 10600k instead; any regressions should be margin-of-error. There are also Ryzen chips with comparable performance levels which may be even less expensive, though I’m not sure by how much.
1 Like
Yeah, the 10600k seems pretty damn strong in that regard. In Cinebench single core the i5 is around 530 points while the Ryzen 3300X is around 495. If cost is important the AMD part is still the way to go, I think. Cheaper platform, cheaper chip, less power draw and it even comes with a cooler.
Would be an interesting test to run those two against each other.
[EDIT] Actually, the i5 is a bit behind the 10900k it seems. And way closer to the 3300X in single core.
2 Likes
Not that I’m aware of, no.
I am not aware of any GPU accelerated sha256sum computations.
But if you come across something, if you don’t mind passing it along, and I’ll take a look at it.
Anything that can help process a single 6.5 TB file would be greatly appreciated.
Yeah…I’m not really concerned about the price/perf metric.
I’m just looking for raw sha256sum performance.
But that’s where I am trying to see if there is data behind it that would support that statement.
It isn’t that I disagree with you.
I’m just looking to make a data-driven decision.
Do you think that Cinebench single thread is really a good, representative benchmark of sha256sum performance?
Thanks.
1 Like
not entirely sure what your doing with it, checking file integrity? breaking passwords?
but it’s been worked on by a lot of people for various task. at least some of them can use opengl to use the gpu, and with password cracking and mining it exponentially better then using the cpu.
https://www.geeks3d.com/hacklab/20181219/hash-functions-md5-sha-1-sha-256/
1 Like
video? sounds wicked
either lv1 or tek syndicate waaaaay back int eh day reported on it, was using all amd 2xxx? 3xxx? gpus and like dozens of them. cant find a link sadly.
1 question:
- what is the speed of your data source?
You may find that the speed of your data source is the limiting factor on modern CPUs. Unless you can fit 6.5 TB into RAM it is going to have to be streamed from somewhere (and even if you can, it’s going to have to be loaded from somewhere into RAM still?) and I doubt you’ll have more than a few GB/sec of bandwidth to load it from?
You might not need the “fastest” CPU for sha256 if you’re bandwidth limited anyway.
3 Likes
checking file integrity as the file is being moved over the network.
It’s being read from my RAID0 array consisting of four HGST 6 TB 7200 rpms HDDs, which are capable of a total of somewhere around 750 MB/s read speed for the entire array. (~187.5 MB/s per drive based on iotop).
Currently, best case scenario (using an Intel Core i7-3930K, a single thread can only read at around 311 MB/s with 64 GB of DDR3-1600 memory I think, I forget the timings).
But if I am reading the file over the network, I’m reading it over 4x EDR 100 Gbps Infiniband (which has a theorectical peak bandwidth of 12.5 GB/s, which I should be able to obtain if I move to U.2 NVMe SSDs).
But at the current rate, I think that I am definitely CPU bound.
When I run openssl speed sha256, I get around 350 MB/s for 8 kiB block sizes.
Thanks.
1 Like
Interesting.
What OS are you running?
Kubuntu 20.04 with Linux kernel version 5.6.14
This is what I get with Intel Core i7-3930K, on Asus P9X79-E WS, with 64 GB of DDR3-1600, running CentOS 7.7.1908 running Linux 3.10.0-1062.el7.x86_64 kernel, running openssl 1.0.2k-fips.
Time to update your kernel?
1 Like
Not sure if it is the kernel, openssl, the RAM, the motherboard, and/or the CPU.
This is part of what makes running this test so difficult in order to come up with some kind of a meaningful answer/conclusion.
The more variables there are, the complexity and the number of permutations of the test that will need to be run out.
sigh…
One of the biggest reasons why I haven’t updated a lot of stuff on the system is because I also don’t want to break other stuff that’s already working (e.g. NFSoRDMA).
Support for that is rather “fragile”.
There are no kernel updates for rhel based distributions only patches. While could update to 7.8 this still would not pull a newer kernel. He would have to update to CentOS 8 which runs the 4.18 kernel.
1 Like