Does Zeropond also have its own pool mining sofware to co-ordinate its miners? Is it possible / reasonable for zcashd or similar CPU miners to join that pool to boost the mining power (and get a share of the reward) to help balance the effect of another big, competing miner?
@tim_olson @austin-williams @daira To my great surprise 4 cores on my CPUs appear to be only 2.84x faster than 1 core (28 data points on testnet). Maybe something changed in z9 because I tested it z7 and z8. I believe it was z9 when the RAM/core was no longer be so stable which made me wonder how big a change the optimization was and what other things could have been affected. That optimization improved single core speed, but could have come at a cost that affected multicore performance. (i.e., it could have used the bandwidth more effectively, which by definition blocks other cores, making single core results faster, but not improving multicore.)
My testing uses identical groups of computers with the only difference being genproclimit=1 or 4. I’m going to repeat the test because they had different start times which had some effect due to unconfirmed blocks, which I adjusted for by approximation.
I believe my 2013 desktops are like 2015 laptops, so I consider them a standard CPU. The single core H/s is 0.044. So I consider a CPU at this point to be 2.84 x .044 = 0.125 H/s. People looking at their pricing should therefore be aware my best estimate is that where they claim “50 CPUs” you are really getting 30 CPUs. This assumes that no other mistakes are being made behind the scenes in how they measure 0.075 H/s. For example, they currently claim their GPUs are 1.2 H/s in contradiction to their previous comments, and they have so far refused more realistic free audits of their results by mining on testnet head-to-head against our CPUs.
Their 50 CPU plan [30 CPUs] price is $400/month in terms of Zcash without the ramp. It will cost me $166 in electricity to do in 30 CPUs the 3.75 H/s they claim and offer for $400 a month. Their comments indicate it will cost them $190 (I’m using $0.14 per kWh, 300 W for 0.6 H/s). Paying $190 in electricity from a gross of $400 seems a little too efficient, but it’s possible. I mention this as a check on their ability to deliver what they claim.
If you are pricing options as carefully as I am, I urge caution until their ability to acquire blocks in beta is known. If you have no other option, or do not wish to do the work, and think they will do as good as botnets and other miners so that your coins will be cheaper through them than on exchanges, then it seems like good pricing even if they are off a factor of 2.
The network hashrate is knowable. What if customers do not get the correct percentage of blocks after the sale of a H/s promise? They can’t promise the number of coins you will acquire, but they have already promised the number of coins per network hashrate (determined by difficulty) by stating a H/s in the sale, and it is measurable after the sale.
If you look here:
https://benchmark.minezcash.com/index.php?title=Main_Page
You will notice the majority of machines do not reach near 0.044 h/s on a single thread. Your pricing does not take into account hardware purchases (including CPU, RAM, motherboard), shipping, time spent managing the hardware, internet connection costs, risk of various sources of downtime, or risk of hardware failure.
I challenge you to find a VPS that beats our price on hashrate with real tests, since this accounts for all of the above details.
As for knowing the network difficulty after launch, of course it is a risk. It’s a risk to anyone involved with the project, as is the price of ZEC. It is a risk to us, and our customers. This is the nature of making an investment; If it was an absolute guarantee of ZEC output we would certainly be in a different position.
EDIT: Additionally, we have not refused any sort of test. We will do a public test and have accepted your request. We’re just working out the best timing, but it will be before sales open up. Please try to be patient.
1 Like
How are you benchmarking with multiple cores? I’m having one hell of a time trying to figure that out. Seems the zcbenchmark solveequihash [NUM_ITERS] [NUM_THREADS] doesn’t actually measure what one might think it measures.
Have you seen this question of mine? Perhaps you could help me figure it out.
@Austin-Williams It’s not merged yet, you need to pull down @str4d’s PR here:
And compile.
3 Likes
We’ve increased the h/s that we consider 1 fully fledged CPU to 0.1 h/s, and adjusted pricing to match (restructured some bulk savings) .
0.0208 is new CPU cost at Zeropond for 0.1 h/s at our highest pricing.
Hopefully you will agree that these are basically real cpu equivalents now.
1 Like
There is indeed a typo.
Genesis Block contracts begin on Genesis Block (currently October 28th) and run for 90 days. Regular contracts begin December 1st, 2017 and run until February 28th, 2017.
Should be
begin December 1st,2016 and run until February 28th,2017
You are correct, thank you. I have fixed this.
Edit- Thanks for the clarification 5alt
@Shawn The price per hash is the same, and actually cheaper on the low end.
We simply called a CPU = 0.1 h/s and then adjusted prices * 0.1/0.075, carrying some of the bulk discount down to the lower end.
When orders open you will be able to enter how many CPUs or h/s you’d like instead of buying a package, so you can still order whatever you were planning on before.
We’re almost done with a calculator that you can use to find what you’d pay for your desired hashrate.
1 Like
gpu x20 faster it sucks
@5a1t On the benchmark you will notice one of the “CPUs” is an atom, the 2nd H/s was not calculated correctly, and one is 1.3 GHz. Also the 2nd to last one clearly had a problem, either from the statistics or his setup. So I get 7 out of 11 (removing the atom and 1.3 GHz from comparison) are 0.044 H/s or higher. When someone says “CPU” I think a 2013 desktop is fair, and that’s only because so many have “slow” new laptops that will be about the same.
It’s better but I am not “happy” with 0.1 H/s. People will understand marketing hype, but I believe you and Tim normally operate at higher level than standard marketing morals. Technically you have 0.1 H/s in your data as I explained above. But if you trust Daira and I, you should see you must have had a statistical error, or the laptop has a problem.
I will also be much happier if you get at least 1 GPU on the testnet for the next 2 days before your sales launch. This is partly for my own investment decision for your services.
There are still other issues that unavoidably require trust in your company that are large enough to overwhelm the above in importance. But I want to make sure the data I have is Kosher: I want to know you have a clear path to profit. I want to know it is at least theoretically a better option than buying more CPUs ($700 for 0.6 H/s, total).
@austin-williams I am comparing 1 core to 4 core by the results on beta testnet, not the benchmark.
1 Like
@Andrei GPU is not 20x faster, please scroll up through this thread to see actual data before making assumptions.
1 Like
To be fair, the effort requested was completely trivial, and necessary in order to demonstrate that Tim Olsen’s old email account was controlled by the same person or people as the Bitcoin forum account. (Obviously, an email from that account could just be forged, so a demonstration that you can read an incoming email is needed.)
1 Like
@zawy
Regarding the benchmark results, if you take zcash’s own benchmark on there, Xeon @ 3.2 Ghz with 32 GB RAM, the h/s is 0.0372 per core. Using your 2.8 multiplier for 4 cores, we get 0.104 h/s. The same is basically true for Voluntary’s i5 and Lego’s i7 results.
The laptop I was using was actually a tower, by the way, with 32 GB RAM and an i7. Tim just didn’t know which computer I ran it on when I gave him my results. It ran four cores with 30 solveequihash solutions and came out with 0.9 h/s. I will try 100 runs.
Regarding the public test, we have agreed to run one and will before orders open. Tim and I are out most of the day Monday (visiting datacenters again, a few meetings, and an interview), and we still have a lot of coding to do for our launch. But the test will come Monday or Tuesday.
1 Like
I did not explain myself clearly in my estimate of your electrical expenses. I was complaining that you are not charging enough. Instead of 2x your electrical costs, I wanted to see 3x or 4x. At 5x I would say it was too expensive. At 2x it makes me wonder about your numbers and the necessary trust issues that overwhelm the details I’m describing.
You would be well ahead of anyone else and conservative in describing to your clients what they are getting if you did what I am estimating you need to do and even what I’d like to see you do: double your current prices while increasing your definition of CPU to 0.125 H/s. This would give you a gross of about 3x your electrical expenses. This works out to $750 a month for 40 CPU equivalents.
Keep in mind on the benchmarks some people were mining at the same time they were running their benchmarks.
Regarding Zcash’s Xeon results, the data on the wiki is in error. You can see their real results are 0.043 H/s. Xeons are always slow for some reason in these benchmarks, but Zcash apparently has a 3.2 GHz version which should not be so slow, even for a Xeon. Here’s Zcash’s Xeon: https://speed.z.cash/changes/
Voluntary’s CPU you are citing uses only 15 Watts. His electrical costs are 1/3 yours on a H/s basis. Wahlberg and janb are other reasonable CPUs that are slower than 0.044 H/s, but they’re also 15 W CPUs. They have sacrificed speed for low power laptops. Electrical costs are a factor as much as initial expenses. My desktops are a better comparison because our electrical costs are about the same.
Concerning lego’s results, you did not read what I said carefully: the H/s column on the wiki is wrong. He got 0.044 H/s.
So everyone is basically 0.044 H/s or better, except for low power laptops who have 1/3 your electrical costs.
[edited]
If you look at the timeline view on speed.z.cash (sorry, a direct link won’t work because of a forum limitation), you can clearly see that the error bars and averages are all over the place between runs. This is because, since the parameter change in z9, the variability is such that only 10 solution runs are definitely not sufficient to give meaningful results. (Also note that the part of the algorithm that got more significant to the total run-time has not had as much optimisation attention.)
I believe most of the results on the wiki are only using 10 solution runs.
2 Likes
If you defined your CPU’s as laptop CPUs, then I could not complain.
2 x power is still insanely profitable and under the assumption that they hash for themselves as well I am sure they can make a killing and feel good about themselves at a fair pricing level.
Pricing based on their cost also gives them a great response should their service still turn out to not be profitable due to low zcash price.
I think it’s great they they have stayed away from crazy zcash assumption pricing or some odd tiered model based on Bitcoin or other unknown.