Folding Forum

In summary:

• The ocore is a streaming core that represents the next big architectural move for F@h. The ocore currently doesn't have a front-end and the backend is under development. It currently runs in the command-line in Linux on AMD and Nvidia cards.
• FahCore 17 is the current generation of the F@h GPU cores and is the preferred core for researchers. It works with existing F@h infrastructure, but doesn't have an AMD Linux build yet because it was buggy when we originally alpha tested FahCore 17.

The question was raised on #fah and there were opinions on both sides. On one hand, proteneer said "until more people on AMD confirm i'm holding out a bit longer" and that he wants to get the "ocore working first", but on the other hand there's been a lot of people on one team who frequently ask about an AMD Linux version of core17, many of whom have powerful cards. There isn't enough development power to push both, hence I'm launching a poll to gauge opinions. Please weigh the pros and cons of both sides, and then cast your vote above.

Jesse_V wrote:In summary:

• The ocore is a streaming core that represents the next big architectural move for F@h. The ocore currently doesn't have a front-end and the backend is under development. It currently runs in the command-line in Linux on AMD and Nvidia cards.
• FahCore 17 is the current generation of the F@h GPU cores and is the preferred core for researchers. It works with existing F@h infrastructure, but doesn't have an AMD Linux build yet because it was buggy when we originally alpha tested FahCore 17.

The question was raised on #fah and there were opinions on both sides. On one hand, proteneer said "until more people on AMD confirm i'm holding out a bit longer" and that he wants to get the "ocore working first", but on the other hand there's been a lot of people on one team who frequently ask about an AMD Linux version of core17, many of whom have powerful cards. There isn't enough development power to push both, hence I'm launching a poll to gauge opinions. Please weigh the pros and cons of both sides, and then cast your vote above.

The ocore is no more than proof of concept. Not even alpha level, as it doesn't even award points yet. Sure, this is the future of fah, but it is a very long way from done. Too long.

Finish the AMD fahcore. It's main stream. It awards points. It will be productive for fah much sooner. And it will have more users on the first day of release than ocores will have in the next 3 months.

Not really a choice at all.

Mod edit to get the poll explanation back at the top of the thread.

Am not sure I have enough information to decide, if the development time frames are similar then the Ocore wins, but if Ocore will take significantly longer then our friends on Linix with AMD perhaps would benefit the project with their powerful cards able to do current projects. Unsure this is a decision donors can make without knowing what the overall benefit will be. How many Linux AMDs are really out there to be tapped? Suspect quite a few.

I really agree with you 7im, good point.

kiore wrote:Am not sure I have enough information to decide, if the development time frames are similar then the Ocore wins, but if Ocore will take significantly longer then our friends on Linix with AMD perhaps would benefit the project with their powerful cards able to do current projects. Unsure this is a decision donors can make without knowing what the overall benefit will be. How many Linux AMDs are really out there to be tapped? Suspect quite a few.

The ocore needs a lot of infrastructure behind it, but the Linux AMD core is, as far as I know, fairly ready to go. The ocore runs on AMD cards in Linux, suggesting that a core17 build would run properly too.

Everything that I've read indicates that there's a lot of AMD folders in Linux who have chosen to run Windows but would prefer to Linux because of this, or they are sitting in Linux with idle AMD cards. Either way, there's a lot of people waiting.

Looks like there was some issues with voting. I've made some changes, so if you were affected please try again. Voters can also change their vote after the fact now if they wish.

The old poll for some reason expired and couldn't be voted on, so I recreated it here instead.

For some reason I didn't see 7im's post before mine, perhaps a merge occurred, but agree totally that main stream, if doable, the better option for now.
We now have 2 polls?

<LasersHurt> Well, this is a curecoin centric reason, but limux + AMD is the most common folding setup I am aware of
<eennaam> my point is legal issues
<eennaam> i cant run windows everywhere
<LasersHurt> The linux machines make great dedicated workers. That's how they took over scrypt mining so easily, cram a bunch of them into a very light OS and off you roll
<LasersHurt> And its free, so have at building rigs til you pass out
<LasersHurt> So if they are looking for nigh industrial folding power, not a bad idea to bring those folks on board

kiore wrote:For some reason I didn't see 7im's post before mine, perhaps a merge occurred, but agree totally that main stream, if doable, the better option for now.
We now have 2 polls?

No, I merged them into one (The previous poll is now locked), which is why 7im's post is now ahead of yours.

<Ricorei> the ocore or core17, as long as i can fold, i won't argue for one or another
...
<tempnick> I'm not on AMD nor on Linux for folding, but for the sake of more computing power, I think it's a better choice to get it on AMD Linux, because many AMD mining rigs are on Linux because multiple AMD cards don't mix well with Windows

Thing is... just how much compute power would AMD cards running on Linux bring?
The drivers are still horrid.

I am only a small miner, but I myself have 12 AMD cards - only 1 is running on Windows, the others are under Linux. I know other larger miners with 350 cards or more. Most of the literature on the internet for running headless miners was to run under Linux - with the majority choosing AMD because of the high hashing power over the older nVidia.
My option at present may be to sell my AMD cards and replace with nVidia, but my choice would be 750Ti - which unfortunately has reported problems with Core17 - so I am in a no-win situation - my electricity costs are the critical factor, in Australia - paying 26c per KWh so reducing power consumption is very important.

Are the drivers still causing people problems? I've been running the ocore with 2 7970s on Ubuntu 14.04 (with Cat 14.4) with zero issues.

SolidSteel144 wrote:Thing is... just how much compute power would AMD cards running on Linux bring?
The drivers are still horrid.

Before CureCoin, still a lot more than are running the ocore. After CureCoin, well, go check the team stats. It's a lot more! And that's without the AMD card running.

The one post above with 12 GPUs is probably more than run ocores.

kiore wrote:Am not sure I have enough information to decide, if the development time frames are similar then the Ocore wins, but if Ocore will take significantly longer then our friends on Linix with AMD perhaps would benefit the project with their powerful cards able to do current projects. Unsure this is a decision donors can make without knowing what the overall benefit will be. How many Linux AMDs are really out there to be tapped? Suspect quite a few.

Any prediction of development time frames has to contain room for lots of unpredictable things being discovered so getting "enough information" is impossible. Nevertheless, consider just the major bullet points.

AMD Core_17 for Linux"
* Port the working OpenCL core to work with the latest AMD drivers.
* Test as necessary and fix any problems discovered.

Ocore:
* Port both of the working Linux cores (GPU and CPU) to work with Windows and with Windows drivers.
* Test as necessary and fix any problems discovered.
* Examine the existing streaming infrastructure for bugs and evaluate growth needs/potential to make it production-ready.
* Create a plan for streaming Points that's fair to everyone.
* Examine the points crediting process for WU-based points and (if necessary) adapt for streaming credits.
* Devise a tracking methodology to answer questions like: "I ran ocore-(GPU/CPU) from [date-time1] to [date-time2] and didn't get the expected credit."
* Evaluate the advantages/disadvantages of a local log file and create that structure if appropriate.
* Decide if ocore will be limited only to the command-line interface and (if appropriate) port it to work with a GPU front-end.