As a test of my new Supermicro development box I loaded up lukminer, the Monero miner written specially to take advantage of the Xeon Phi by an Intel employee in his spare time.
The result on my 7290 indicates a serious problem with this hardware. I got a mere 493 Hashes/sec whereas on my Dell T7610 with the lowliest first generation 3120A inserted I achieved 544 H/s. I have alerted Supermicro via their reseller (WiredZone) and will be working with them to diagnose the problem.
Testing by ASRock suggests this CPU should achieve more like 3,075 H/s.
If anyone else has the official so-called "Ninja Development Platform" and would like to confirm or deny these results, I used the executable file luk-phi from the lukMiner-0.10.1-cpu-ocl-phi.tgz package available at www.lukminer.net/releases. The command line used was:
./luk-phi -a xn --host xmr-usa.dwarfpool.com --port 8080 --user 4ALcw9nTAStZSshoWVUJakZ6tLwTDhixhQUQNJkCn4t3fG3MMK19WZM44HnQRvjqmz4LkkA8t565v7iBwQXx2r34HNroSAZ.3e2ff2735ec20b4a749bc2ad636135da081a569dd6a81ca8c760f688a9070081 --pass x
For testing purposes you can use this account, I'm not seriously mining to it. On first invocation you will see a message to enable huge_pages support. This should be set to at least 5000.
The program was run under Ubuntu 16.04.4 and the system configuration includes 48GB of RAM but this is not required as it should run in internal memory on the CPU. The OS is booted from a 64GB SSD SATA-DOM.
Expiry date for this test is 6 April, after which the Monero algorithm will change.
After emailing Lukas Alexander (not his real name) about the mining results he responded that my bios settings probably needed tweaking. This was definitely a case of newbie anxiety, he was totally right.
The system was shipped with AllToAll and Flat settings. Changing these to Quadrant and Cache made all the difference.
The luk-phi miner now gets a respectable 2,600H/s which, while still significantly short of the ASRock results, indicates that the machine is not faulty and that probably a bios upgrade will make up the difference eventually.
So we're now running at almost 5x the speed of a 3120A. Impressive indeed.
Almost a year later and 87% down :)
To make the nr_hugepages stick between boots, you would do:
# sysctl -w vm.nr_hugepages=8192
vm.nr_hugepages = 8192 [this is what the system should reply]
I was never interested in the mining itself, just the technology behind the phi processor. Around the same time you wrote this, I reached 28K on the Fujitsu trial https://docs.fujitsu.fr/knlrap/doku.php running between 16 and 24 nodes; it shouldn't have been possible for me to that, but I figure out a way of bypassing the PBS manager. The 28K was with hugespages disabled; I can only imagine what would be if I had set them to 8192, but I did not have root access.
Regardless, I hope your Ninja station is not in a corner getting dusty. I just came across this https://www.ebay.com/sch/i.html?_from=R40&_trksid=m570.l1313&_nkw=%22Intel+Xeon+Phi+7285%22&_sacat=0... which is a tottaly different beast from the one you have as you can see here https://ark.intel.com/products/series/132784/Intel-Xeon-Phi-72x5-Processor-Family The x205 was launched in December 2017 while the Xeon Phi 7210, 7210F, 7230, 7230F, 7250, 7250F, 7290, and 7290F were discontinued in July and can no longer be ordered.
The 7285 should plug right into your Ninja station and turn it into an amazing general purpose machine, including running virtualization, something the x200 series could not do. Keep in mind that those are "Engineering Samples". Should you decide to go this way please keep me posted; I would love to try out this combination.