Bitcoin Forum

Can someone explain to me why is Rx 580 doing about 32MH same as Gtx 1070 on ETH or ETC, but rx 580 is doing about 300sols and gtx 1070 about 450sols on ZEC? I really dont get it, they are both gddr5, they have the same memory speed and memory bandwidth, only difference is core speed, gtx have 1920 cores at speed 1650, rx have 2300 cores at 1350 which is about 7% faster on gtx, but gtx is getting about 50% more sols on ZEC.

GTX 1070 is a way better card then the RX 580.
But in Etheruem AMD cards are better so that's why you don't see the difference.
But Nvidia cards are stronger in Zcash, so that's the reason why your RX580 won't compete with a 1070.

You just repeated what he said...I think he wanted to know WHY :D

One of guys who write minig software could probably explain it best
(claymore, wolf etc)

Is about modding i think, AMD share a lot, in counterpart NVDIA prefer to keep everything for NVIDIA. I suppose AMD get a lot more of work time from all this miner dev, in result AMD can versus NVIDIA, but i suspect if dev miner put equivalent work on NVIDIA optimisation ( but as i think is more complicated a lot of nvidia env is closed ).

Diffirence is on memory rail  ,example Gtx 1080 256 bit vs gtx 1080ti 352 bit.

Haha exactly. You are right I should ask wolf or claymore about this, because it really doesn't make sense.

You are right  there is difference between gtx 1080 and 1080ti, but rx 580 and gtx 1070 have the same memory bus, in another words they are the same 256bit cards. Edit: and have the same memory size 8GB vs 8GB

1070 gtx is much more powerful on the compute side , also the cards access  memory in a different way .  more over the 580 not modded is much slower than a 1070 .... 1070 in ETHEREUM uses half of the energy of a 580  or 2/3 depends  on the oc ...  1070 gtx is overall a much better card

Before mining craze, 1070 costed as two RX580's, so real question is if 1070 gtx is better than TWO RX580s

I dont care about the price in this scenario, im interested in why is gtx doing more sols on zec than rx, but on eth those two cards do the same speed, if cards access memory in a different way the question still exist, why does it only effect zec. On white paper cards are almost exactly the same.

GTX 1070:
Cores : TMUs : ROPs1920 : 120 : 64
Base Clock1506 MHz
Boost Clock1683 MHz
Memory Clock (Effective)2002 (8008) MHz
Computing Power (FP32)5,783 GFLOPS
Memory Size8192 MB GDDR5
Memory Bus Width256-bit
Memory Bandwidth256.26 GB/s

Rx 580:
Cores : TMUs : ROPs2304 : 144 : 32
Base Clock1256 MHz
Boost Clock1340 MHz
Memory Clock (Effective)2000 (8000) MHz
Computing Power (FP32)5,788 GFLOPS
Memory Size8192 MB GDDR5
Memory Bus Width256-bit
Memory Bandwidth256 GB/s

Zec algorithm uses memory very little and only GPU performance matters. You can as well down-clock your memory and see no difference in zec sols.

That ^^^ should explain why 1070 does zec better than RX580.

They have very different architecture, even though specs look very similar they have huge differences.
Also AMD drivers are shit  8)

Even old R9 280X does it better in ZEC with 340...also, Nvidia drivers are shit
thats why twice as expensive card barely do the same in ETH  8)

Hahaha
Ok they are build on a different architecture i get that (vega and polaris are too), still doesnt explain why is one card doing so great in one coin but not in another (both having the same spec on paper), either gtx is way under performing in ETH or rx is way under performing in ZEC. I would never ask if we were talking about 5-10% but having the same speed in eth, while in ZEC gtx have 50% more sols is odd, gtx then should get about 48MH in ETH or rx about 16MH.

BTW believe it or not R9 280X have faster memory than RX580

No, slower memory...but wider bus...384 vs 256 bit, so better memory throughput

However, people tried to explain that Nvidia is (with same memory speed) faster than RX580
because ZEC depends on core speed, and not memory speed...contradictions on all sides of story :)

I think it is more to do compute core structure. The 280x is similar speed to 570.

Yea you are right, slower memory but wider bus, Memory Bandwidth 288 GB/s vs 256.
IMO I dont think that zec is more core speed oriented, because r9 280x have lower core speed than rx580, and gtx have about 7% more core speed than rx 580 which again doesn't explain 50% more sols in ZEC.

Hey @Wolf0 can you pliz help us out here, we need a pro tip.

I think not only core speed matter, with nvidia you have cuda technology what is done by nvidia for nvidia, it is well optimised. On amd you have opencl what is created by apple with amd, intel, nvidia, qualcomm cooperation, it is not optimsed for amd, it is for all. If you compare opencl on amd and nvidia then amd would win, but zcash on cuda it is different matter..

I thought claymore on Nvidia uses CUDA for ETH, too (at least, I see two CUDA directories
installed with it)...so what happened with CUDA magic on ETH?

Amd memory access is better (specially when you can mod memory straps), that why eth is better on amd, core performance is better optimised with cuda (and nvidia have more speed), that why zcash is better on nvidia.

...or should we say, AMD cards have better memory performance under 2GB but have troubles reaching 2GB+ regions. Unless they really going to fix it in their drivers, I'd say overall nvidia has better memory perf.

Now we are getting somewhere, ETH is more memory speed oriented, of-course core speed is important too, but in ETH memory speed is "more important" with better memory access from miners, amd is doing more MH than gtx, while (Branko was right) in ZEC core speed is "more important" (even on paper gtx have faster core speed) with (better optimization on) cuda gtx is better.

AMD has confirm that they did fixed dag problem.

I must have missed that news. Any independent confirmations?

It would be nice to see some segregation in mining hashpower in the near future. Say, all red cards will be pointed at ETH and all green will mine ZEC.

I think it was mentioned by claymore in his thread,

https://bitcointalk.org/index.php?topic=1433925.msg20491019#msg20491019

I don't think that driver is still released but its probable that it will in the next days.

Looks like good news indeed. I wonder what did it take for them to fix the problem? Or was it just a bug in the driver?

The 2GB DAG boundary issue reminds me of the 32-bit addressing problem and limitations of an Int32 which maxes out at exactly 2GB.

I have some RX580 and the maximum I can make with them is 26 Mh/s.

I have updated memory timings, overclocked memory / underclocked gpu, but nothing abore 26 Mh/s.

I have one 1080, in Zcash I can get 500-540 sols/s (overclocked) using Nvidia native software.

1080's price is almost twice as the rx580. I'm not sure if it's worthy. Maybe getting 2x rx580 is better.