It would have to be faster than 218GB/sec to get better than 4.5 times improvement over a CPU accessing DDR4.
ASUS R9 390 is 8 GB at 384 GB/s for $300 used on Amazon. I just donât know how good the GPU cores are going to be at making use of the bandwidth. The only hope for CPUs might be if the Equihash algorithm is inherently electricity-expensive for the cores. That way it will not be long before it cost GPUs and CPUs the same $1 per coin in electricity (if they are made at the same chip resolution like 22 nm).
1 Like
If memory bandwidth is the only factor then speeds of up to 33 times improvement over a regular CPU accessing DDR4 are possible with some top of line 1024GB/s GPUs available today. Depending on the role that memory latency plays on GPUs that will probably decrease to 20x, 15x or 13x. Even then an open source GPU miner will be appreciated for a fair launch if results are higher than 9x.
Not on the short term, but later on, at around or below the 9x, GPU electricity costs will start to play a role and may not be sustainable in the long term.
It is therefore encouraged that people currently using and testing GPU miners publish their results. It will greatly benefit Zcash launch.
1 Like
Wow, things have come a long way since last I looked at GPUs
Yeah just divide the RAM access speed of the GPU by that of your comparison memory and you get the relative (theoretical) maximum speed up.
So it sounds like more than a 4.5 times speed up over CPU with DDR4 is possible.
That does change things.
1 Like
RAM bandwidth is only half the factor as the benchmarks show. The CPU is the other half. Few reading this have maxed out their RAM bandwidth. Only someone with 8 cores (not 8 intel hyperthreads) at 3 GHz can really max out DDR3 1600 MHz on this routine. Iâm hoping the Equihash algorithm will prevent GPUs from being more than 2x as efficient, and might be only 1x. First things first: the sort routine for Zcashâs CPU miner is probably far from being optimized.
I donât want to come off as not allowing private mining operations to profit. I just think we could easily find a space where zeropond makes a killing on their operation, and the community gets to compete on roughly equal terms. I doubt with the popularity of Zcash (at least measuring the interest so far) will in any way limit how much money Zeropondâs profitability.
I for one will be buying a contract or several, just because I donât want to setup a farm myself.
Most datacenters have a no-camera policy to protect their customersâ sensitive setups, but we are visiting our probable DC tomorrow and weâll get something on camera.
Yes we will do a demonstration on testnet. We need to figure out the best way to set that up and prove something. Maybe weâll announce a demo time in advance, then film us running a single GPU against testnet, something like that. Another other suggestions?
I think we are all interested in setup tech specs, and if possible a comparison to a common say i7 or Xeon CPU. Pricing as well of course.
About the time warp attack: correct me if Iâm wrong, but the attacker needs to have a multiple of the network hash rate to pull it off, right? Or else they donât solve many blocks during the warp? If we have a big strong network hash rate right from Genesis, doesnât this mitigate the problem?
1 Like
The following procedure would convince me.
Verifiably Demonstrating Hashing Power
(1) Pick a (bitcoin) block number some time in the future (say, bitcoin block number 431000).
(2) When that (bitcoin) block is submitted to the bitcoin network, let d denote its block hash. Let p be the string âzeropondâ.
Begin computing Equihash on your mining equipment, but instead of using H( I || V || xi ) as per normal, use H( d || p || V || xi ).
Instead of computing the difficulty as H( I || V || x1 || x2 || . . . || x2^k ) (as per normal), instead compute it as H( d || p || V || x1 || x2 || . . . || x2^k ).
Keep track of the solution you find with best difficulty you are able achieve.
(3) After a number of hours (as many as you want, but at least four), let D denote the solution with the best difficulty you were able to achieve.
OP_RETURN the SHA1 hash of D to the bitcoin blockchain. (Be sure to use a large miner fee so it is included in earliest block possible).
(4) Report back here and show us:
(a) the solution, D , that you found along with
(b) the txid of the transaction you used in your OP_RETURN.
Why This Works
(1) Including the bitcoin block hash, d , proves to me when (at the earliest) you started your demonstration. This way I know you arenât âfast-forwardingâ.
(2) Including p (the string âzeropondâ) indicates to me that you (or someone cooperating with you) is actually in control of the hashing power being demonstrated.
(3) OP_RETURNING your result, D, acts as a timestamp so we can verify exactly when your demonstration ended.
(4) In total that gives us a way to irrefutably verify how long your demonstration lasted (say X bitcoin blocks). We also know the difficulty you you were able to achieve during that time (because youâll have produced and timestamped D). From that we can compute an accurate estimate of the hashing power under your control.
Naturally you wonât be able do this after Sprout launches (because youâll want all your hashing power actually mining). But then, after sprout, there will be enough hashing power on the zcash blockchain that we wonât have to worry about Zeropond âfast-forwardingâ. So, after Sprout launches, you can demonstrate your hashing power by adding a footprint to all the blocks you find.
EDIT: Fixed errors and formatting.
1 Like
Yes, to do the time warp attack you would need 2x network hashrate or more for it to make sense. And that your uses can buy time in increments less than an hour. Yes, if you enable high hash rate at launch with constant-on clients, it helps mitigate time warp attacks. We need about 2,000 CPU-equivalents to mitigate the already-known threat. In their 3rd difficulty algorithm, digisheld was still talking about 5x and 10x attacks.
Couldnât we just mine to a predetermined address instead? Seems easier than the OP_RETURN stuff for proving that itâs us.
What if we did like a livestream video where we use a single system to boost testnet diff and mine some blocks to our address?
1 Like
He only needs to mine the beta and tell us when he started. Thatâs the best way to test it. To make it a quick test, he should activate 20 instances for 4 hours. If he has 3x my old DDR3 CPUs per âinstanceâ (say, an account someone would purchase from them) the difficulty should double in 2 hours. But I would look at the solve times to get a better measure. Telling us which blocks were his would confirm there was no other big change in the network, and allow a more accurate measure.
Wouldânt a simple benchmark 10 run be sufficient ? If we know the hardware and itâs livestreamed ?
Hi,
Will it be possible to get in with creditcard?
Im in holiday next 7 days. Bad timing :-),
Thanks.
He has his own GPU miner so I donât know if it can tap into the benchmark. He could leave 1 instance on for a day and weâd have a good idea.
Iâm concerned that the GPU devs are not releasing an estimate of their capability. Iâm afraid this means theyâre at the 10x mark or higher. They will not tell because then it would alert other investors and devs of the possibility and thereby spur more research and investment and thereby cut into their profit since they are very likely taking something like a 20% cut of the coins mined so that they remain on board for future changes and as a payoff to keep them silent and not release other code. Likewise, Tim Olson telling us his capability might be like ringing the dinner bell for competitors if he discloses too soon if he can do 6x or greater.
It also risks Zcash changing the algorithm to make CPU more competitive. I think they should go to 1.5 GB per thread and 10 minute mining (if needed due to a security risk for slow mining) to reduce the GPU advantage. Instead of 10 threads in a 8 GB GPU, they would only be able to do 5, while an 8 GB CPU can still do 4 threads.
.
2 Likes
I also think 1.5Gb or 2Gb would be more adequate
Theyâre under pressure to release and to not make major changes in beta. The CPU dream might be over. Just a few days ago Nick Sabo pointed out in twitter someone might have solved the PoS problem, which can overturn bitcoin et alâs philosophy of âmight is rightâ (PoW)
https://eprint.iacr.org/2016/889.pdf
So if Zcash delays, they might make another overhaul that changes everything.
Thatâs the thing. He could very easily just lie and say âI started this morningâ when actually he started last week. Then, heâll release a weekâs worth of blocks in one day and say âlook how much mining power we have! We mined all those blocks in one day!â
If weâre going to use that method then we need a verifiable timestamp on the first block they mine.