Is there a slack channel since you mentioned that in the website of the zeropond.
Weâve started blogging our progress!
here it is:
ROFL ⌠
Seems about right
Hahaha, unfortunately (fortunately?) I donât wear vests and know nothing about ASIC design. 
A lot of people told us the pricing was confusing, so weâve update our pricing into packages to make it easier to understand:
http://cloud.zeropond.com/
My single core testing so far indicates 4 cores is only 2.8x more than 1 core on the network. Itâs far from statistical significance (only 9 data points). The expected error on this is at least 1.9 to 3.7. Iâll up my estimate to say 4x to 5x CPU is a fair statement, close to your 6x laptop statement.
But how are you getting 0.6 H/s for your GPU? Is it somehow the same benchmark software?
What is your electrical expense for the 0.6 H/s? On my 2013 machines, if 4 cores = 3x one core, it is 250 W for 0.6 H/s.
As @5a1t mentioned, you were right that we were measuring Equihash solution count not H/s, so we needed to adjust by 2x. We are getting about 1.2 H/s for a typical high-end GPU. We assume 300W draws if you include the motherboard.
He only said the Xeon data was transferred incorrectly. If youâre at 1.2 H/s then youâre back up to 1.2 / (0.05 x 2 For4Cores) = 12x the laptop.
The parameters changed in z8 precisely in order to reduce the solution latency.
The difficulty starting point is likely to be significantly higher on mainnet.
1 Like
I see your web page is calling 0.075 a CPU equivalent, maybe something you added today. Your laptop data indicates 0.1 should be an equivalent. As an independent party, Iâm saying 0.15 H/s would be a more fair comparison. I think you should start high with my number, which would cut your CPU numbers in half.
This is slightly misleading, because the longer solve times give more solutions. There are two stages to the algorithm: finding âpartial solutionsâ, and reconstructing + difficulty checking each partial solution. The average number of solutions is slightly less than 2 (the Equihash paper made an approximation); sometimes it is none. Finding the partial solutions, and testing each partial solution take roughly the same time for n=200, k=9. The benchmark reports the time to check all solutions; it therefore reports a longer time for runs that are more likely to pass the difficulty check (and the shortest time for runs that have no chance of passing the difficulty check, because they found zero partial solutions). This makes the benchmark results somewhat hard to interpret, although if you measure over enough runs (Iâd recommend at least 50), then results should be comparable between systems. Itâs definitely necessary to apply @str4dâs PR to get meaningful results for the multi threaded benchmark.
Summary: you actually canât measure âhashes per secondâ yet. Measure âfull solution runs per secondâ instead. (Just make sure that youâre not comparing numbers that have been multiplied by 2 with numbers that havenât.)
4 Likes
Please note that the 12-core box scaled only slightly worse than linearly up to 12 cores (and less well when it started using hyperthreads, as expected). Memory bandwidth is a factor, but youâre not going to saturate that with only a few cores. Therefore, I would say that comparing a GPU against the single-core benchmark is misleading. In fact if you compare against a 4-core or 8-core, then the claim of only a 4Ă GPU advantage in the Equihash paper stands up very well (so far, allowing that both the GPU and CPU solvers can be further optimised).
4 Likes
Benchmark, multithreaded benchmark, GPU benchmark, and testnet measurements are all different. The most reliable way to compare is to use testnet. I need one of your GPUâs mining for 2 days to get about 60 blocks. It took 24 hours for my 5 PCs to get 30 blocks [edit: it may have been as high as 40 because I have not looked at unconfirmed blocks]. 60 is really needed. I would like to buy some of your services, but I want to confirm your H/s numbers. I donât know what metric youâll use in selling 0.075 H/s, but I at least want to know all your measurements are self-consistent and I want to know how your H/s compares to the H/s I am getting for my CPUs.
Does anyone know if the price currently displayed is before or after the 50% discount?
The price currently displayed is before the 50% discount
1 Like
Hi All! Sorry, but i all the same do not understand about price. Writen âEach contract runs for three months of continuous mining.â Then below read â1 CPU - 0.015625 BTC / monthâ (for Alpha-Miner). So how much is this contract - BTC 0.015625 x 3 ?. So, or not?
Sorry for my bad english, also.
Sorry, Iâm not an English expert, but the wording on the site suggests the price currently displayed is AFTER 50% and 20% discounts. It says "âIncludes 50% off first month & early adopter discounts (20% off our regular price after launch).â.