.05 BTC Reward for answering question about ZCash mining

[rdnkjdi]

I'm going to ask this question again & looking for an answer from someone who knows what they're talking about (it shouldn't take you more than 5 or 10 min).

Per this thread - https://bitcointalk.org/index.php?topic=1438066.msg14704298#msg14704298

http://The bottleneck is going to be sorting - so it will be transfer speed between the ram and the processor and the processor speed.  Not the amount of RAM that will be the bottleneck?

And on the official ZCash forums there's this -

Note that Equihash is not intended to be "GPU-resistant"; only to limit the relative advantage of a GPU over a CPU. There is some discussion of the relative efficiency of CPUs and GPUs for parallel sorting, which is designed to be the main performance bottleneck, on page 10 (section VI part b)) of the Equihash paper20.

Now my question is - what is "sorting" going to look like hardware wise?  What's going to be the most efficient / give the most hash from the list of the following?  Memory size (8GB > 4GB), memory speed (Nano > 390X), or card processing power? Any ideas on power usage / hash ratio?

Which of these combinations is likely to perform better & by how much?

6 X R9 280X 3GB
3.5 X R9 390
1.5 R9 Nano

I'm looking for an answer from Wolf0, Claymore or Tromp

[1715328407]

1715328407

[1715328407]

1715328407

[joblo]

I'm going to ask this question again & looking for an answer from someone who knows what they're talking about (it shouldn't take you more than 5 or 10 min).

Per this thread - https://bitcointalk.org/index.php?topic=1438066.msg14704298#msg14704298

http://The bottleneck is going to be sorting - so it will be transfer speed between the ram and the processor and the processor speed.  Not the amount of RAM that will be the bottleneck?

And on the official ZCash forums there's this -

Note that Equihash is not intended to be "GPU-resistant"; only to limit the relative advantage of a GPU over a CPU. There is some discussion of the relative efficiency of CPUs and GPUs for parallel sorting, which is designed to be the main performance bottleneck, on page 10 (section VI part b)) of the Equihash paper20.

Now my question is - what is "sorting" going to look like hardware wise?  What's going to be the most efficient / give the most hash from the list of the following?  Memory size (8GB > 4GB), memory speed (Nano > 390X), or card processing power? Any ideas on power usage / hash ratio?

Which of these combinations is likely to perform better & by how much?

6 X R9 280X 3GB
3.5 X R9 390
1.5 R9 Nano

I'm looking for an answer from Wolf0, Claymore or Tromp

Sorting is the first real programming assignment for any comp-sci student.

What it looks like HW wise is shuffling data around in memory until it is sorted. The amount of memmory is not realy a factor
but the cache size is. Sorting involves random accesses to memory. The bigger the cache the more likely an access will
hit the cache. A worst case scenario is that each access misses the cache and has to
fetch the data from memory. That's where the second bottle neck comes into play, memory bandwidth, how fast can you
get the data from memory to the cache.

This seems to be a similar approach to HOdl except that HOdl doesn't sort but search. Searching can be done sequentially
to take advantage of the cache. As previously mentioned sorting is mostly random.

As long as there is enough RAM to contain the entire sort set the amount of RAM doesn't matter. Cache size and mem bandwidth
are the main limiting factors.

[adaseb]

When will this coin be released? Like 6 months from now?

[Ayers]

it say that you can only mine them on testnet right? so there is no real mining for now, why bothering with that coin? also new gpu are coming it's really better to wait and see because you are wasting money by building a new machine now

[rdnkjdi]

I'm going to ask this question again & looking for an answer from someone who knows what they're talking about (it shouldn't take you more than 5 or 10 min).

Per this thread - https://bitcointalk.org/index.php?topic=1438066.msg14704298#msg14704298

http://The bottleneck is going to be sorting - so it will be transfer speed between the ram and the processor and the processor speed.  Not the amount of RAM that will be the bottleneck?

And on the official ZCash forums there's this -

Note that Equihash is not intended to be "GPU-resistant"; only to limit the relative advantage of a GPU over a CPU. There is some discussion of the relative efficiency of CPUs and GPUs for parallel sorting, which is designed to be the main performance bottleneck, on page 10 (section VI part b)) of the Equihash paper20.

Now my question is - what is "sorting" going to look like hardware wise?  What's going to be the most efficient / give the most hash from the list of the following?  Memory size (8GB > 4GB), memory speed (Nano > 390X), or card processing power? Any ideas on power usage / hash ratio?

Which of these combinations is likely to perform better & by how much?

6 X R9 280X 3GB
3.5 X R9 390
1.5 R9 Nano

I'm looking for an answer from Wolf0, Claymore or Tromp

Sorting is the first real programming assignment for any comp-sci student.

What it looks like HW wise is shuffling data around in memory until it is sorted. The amount of memmory is not realy a factor
but the cache size is. Sorting involves random accesses to memory. The bigger the cache the more likely an access will
hit the cache. A worst case scenario is that each access misses the cache and has to
fetch the data from memory. That's where the second bottle neck comes into play, memory bandwidth, how fast can you
get the data from memory to the cache.

This seems to be a similar approach to HOdl except that HOdl doesn't sort but search. Searching can be done sequentially
to take advantage of the cache. As previously mentioned sorting is mostly random.

As long as there is enough RAM to contain the entire sort set the amount of RAM doesn't matter. Cache size and mem bandwidth
are the main limiting factors.

Give me your btc address - & thank you for the explanation. 

http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/4 - Essentially if cache winds up being the primary bottleneck then Fiji based cards (2mb l2 cache) could be up to 2X fast as Hawaii architecture (1Mb cache) and Polaris may thump all of them with 6mb cache.

http://techreport.com/news/29616/amd-will-introduce-two-polaris-gpus-this-year

[joblo]

I'm going to ask this question again & looking for an answer from someone who knows what they're talking about (it shouldn't take you more than 5 or 10 min).

Per this thread - https://bitcointalk.org/index.php?topic=1438066.msg14704298#msg14704298

http://The bottleneck is going to be sorting - so it will be transfer speed between the ram and the processor and the processor speed.  Not the amount of RAM that will be the bottleneck?

And on the official ZCash forums there's this -

Note that Equihash is not intended to be "GPU-resistant"; only to limit the relative advantage of a GPU over a CPU. There is some discussion of the relative efficiency of CPUs and GPUs for parallel sorting, which is designed to be the main performance bottleneck, on page 10 (section VI part b)) of the Equihash paper20.

Now my question is - what is "sorting" going to look like hardware wise?  What's going to be the most efficient / give the most hash from the list of the following?  Memory size (8GB > 4GB), memory speed (Nano > 390X), or card processing power? Any ideas on power usage / hash ratio?

Which of these combinations is likely to perform better & by how much?

6 X R9 280X 3GB
3.5 X R9 390
1.5 R9 Nano

I'm looking for an answer from Wolf0, Claymore or Tromp

Sorting is the first real programming assignment for any comp-sci student.

What it looks like HW wise is shuffling data around in memory until it is sorted. The amount of memmory is not realy a factor
but the cache size is. Sorting involves random accesses to memory. The bigger the cache the more likely an access will
hit the cache. A worst case scenario is that each access misses the cache and has to
fetch the data from memory. That's where the second bottle neck comes into play, memory bandwidth, how fast can you
get the data from memory to the cache.

This seems to be a similar approach to HOdl except that HOdl doesn't sort but search. Searching can be done sequentially
to take advantage of the cache. As previously mentioned sorting is mostly random.

As long as there is enough RAM to contain the entire sort set the amount of RAM doesn't matter. Cache size and mem bandwidth
are the main limiting factors.

Give me your btc address - & thank you for the explanation. 

http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/4 - Essentially if cache winds up being the primary bottleneck then Fiji based cards (2mb l2 cache) could be up to 2X fast as Hawaii architecture (1Mb cache) and Polaris may thump all of them with 6mb cache.

http://techreport.com/news/29616/amd-will-introduce-two-polaris-gpus-this-year

I was reading a bit of the white paper and there may be techniques to optimize the sort based on cache size. The list could be
split into smalller chunks that fit in the cache. Once all the chunks have been sorted they can be merged into one big sorted list.
There may already exist such an algorithm but I haven't looked at sorting algorithms since that first assignment in school.
The only trick would be tuning it for the cache size of the CPU being used.

I didn't really respond for the purpose of claiming your bounty but my BTC addr is in my sig.

[rdnkjdi]

I know - intelligent people have better things to do with their time than reply to bounties.  But still - I'm grateful & like to stick to my word.  Not sure how I missed your sig.

[joblo]

I know - intelligent people have better things to do with their time than reply to bounties.  But still - I'm grateful & like to stick to my word.  Not sure how I missed your sig.

Rceived, many thanks.