No offense Carmen, but some serious research is needed if you're planning to include this in a Bitcoin podcast and you can't answer that question.

Research != making a new forum post to ask about it, either.  You're in the right spot though, this place is a trove of information.  Just start reading and see what people are using.

IMO, the Radeon HD 5850 and 5870 seem to be the sweet spots for simple price vs performance vs power usage, too bad they're impossible to find new.

5970s on the other hand are amazing hashers, but $$.  Same deal with 6990s.

I would bump the mem clocks up for general usability, 300mhz is a bit slow for graphics performance.  The only loss is slightly higher power usage, maybe 10W or so.

I bought a batch of 5 Sapphire 5830 'Xtreme's a while back, they all clock a bit differently.

Rig 1:  ASrock Extreme3,  750W + 480W PSU, 2x 6950 in the 16x slots, 2x 5830 in the 1x slots.
The 5830s are rock solid at 1010/300 for ~320MH/s each.

Rig 2:  ASrock Extreme3,  750W + 480W PSU, 2x 6950 in the 16x slots, 2x 5830 in the 1x slots.
The 5830s lock everything up at anything > 930/300 for ~292MH/s each.

The 6950s are clocked the same in both boards, 885/760.

Another rig has two 'older' Sapphire 5830s, they're unhappy > 940/300, and yet another single-card rig has one of the Xtremes stable at 970/300.

The stable clock rates seem to be dependent on the motherboard more than anything else for me - if I swap the cards between the two ASrock boards, Rig 1 is still happy at 1010/300, and Rig 2 still locks up > 940/300.

This tells me the 5830s all probably have a bit of headroom still, but the motherboards can't push 'em that hard without getting a bit squirrely.

It's half looking at what other people have done and half experimentation to find values that actually work well for your card.

Use the other examples you've seen, try each, and find the numbers that give you the best hashrates.  Every card is different - some overclock well, others not at all, etc, you need to experiment.

That said, lowering your memory speeds reduces power consumption, so try to find the lowest value that's stable on your card and doesn't degrade your hashrates.

Right?  And I just blew my funds on a new motherboard and a 6870.  Hindsight, you're a killer.

Then again I would've needed the new mobo and a PSU for one of these guys, so better to be $300 down instead of $1000

Well, after a little more research it turns out this probably does not matter much.

In applications with significant memory usage, clock alignment might have some effect on throughput, as the GPU may be stuck waiting on the next memory register to fill.  This is partially why boosting the memory clock gives you better framerates in games.  Lining up the clocks would help ensure a just-filled register will be available immediately.

In our case, where memory bandwidth / throughput requirements are tiny, we can stack up a bunch of instructions and don't have to worry about starving the GPU of work.  This means any desynchronization is going to have very little to no effect on overall hashrate.  The GPU spends significantly more time on processing the current instruction set than reading in the next.

I believe the goal then is just 'find stable numbers you think are pretty'

I should probably clarify that what I'm looking for are speeds where there's an actual alignment of GPU clock ticks to Memory clock ticks, not the clock speeds that happen to work well for you. 

Finding such a number could allow us to find a 'peak efficiency' where the GPU / MEM are never waiting for the other to swing, potentially increasing GPU load and average hashrates.

Logic would dictate that any integral ratio would work - 3/1 ratios would be 3 GPU cycles for each 1 memory cycle, for example, but hardware doesn't always line up like that.

My hunch is a 3/1 or 4/1 ratio, but 'pretty accurate' isn't the same as actually accurate

I'd like to think we'll find for each card model or family there's a best ratio to use.  On my 5830s  999/333 seemed very stable, but still fluxuated a bit (+/- 5MH/s).  There are a lot of variables here though - that could also be due to worksize or latency requesting the next job, etc.

Mining MH/s may not be the best measurement for what I'm looking for..

Hey all,

I'm just wondering if anyone has done any experimentation to find a proper GPU / MEM clock ratio for these radeon cards..

I know that the clocks are decoupled and can be set individually, but for streamlined operation (making sure the GPU isn't waiting for a memory tick or vice-versa) it seems like there should be a clock ratio that gives best performance.  I'm not super familiar with GPU architecture, so I may be way off base, if anyone actually knows better please clarify.

For example, if there was a 3/1 clock ratio for GPU to MEM, 999/333 might perform better than, say, 1000/333 due to misaligned cycles.

If there's any merit to this idea, it may only show up in long-term averages; it's quite possible there'd be no obvious effect untill the cycles were seriously misaligned, which might happen in only 1/100 cycles.

Also note these numbers are all bogus, I didn't do any real math

Also note it's likely any effect here would end up adding up to < 1MH/s, but it could also be the difference between a nice, stable hashing speed vs one that bounces around +/- 10MH/s

So, has anyone done such research?  If you've only experimented with different memory clocks, but kept track, that'd be useful info as well.  With enough data we might glean useful results.

lmao

On these cards, the only direction the fans blow is 'into the heatsink'.  The exhaust comes out whichever path has the least resistance, which by default happens to be the open 'rear' of the card rather than out the grill.  A powerful fan can reverse this, but probably at an efficiency reduction.

Sapphire's coolers work well for stand-alone cards, but when you've got a mess of them and need to actually try controlling the airflow, man they become a big PITA.

Mike's right, there isn't any limitation on the number of PCIe devices outside of the number of PCIe slots you've got.

The danger is in overloading the power rails and traces.  The motherboard will often let you pull a lot more power via the PCIe slots than the 24pin ATX connector can handle, and you end up melting your 12V wires.

So, you can run more than 4 devices without powered PCIe extenders, it's just not a good idea after 4 cards.  This is especially important in relation to the dual-GPU cards, they generally suck a lot more power from the PCIe slots than single GPU cards, up to the 75W available limit if the manufacturer has any care for specs.

I thought about this too.  I was considering a very similar setup but with isolated GPU 'slots' - basically a box around each GPU.

You can cut a slot in the lexan/acrylic/whatever for the PCIe connector to poke through, but otherwise it keeps each GPU from sucking down it's neighbor's heat, while efficiently exhausting it's own.

Such a thing would work pretty well even in a 'traditional' closed case as long as you can push enough air through the GPU channels.

At stock clocks, I loose about 400MH/s between all the cards on my rigs and my power consumption drops about 150W total.

Sorry, but overclocking like mad is still the way to go.

It's simply a matter of ROI, and even though miners have scooped up hardware left and right, it's only been specific hardware lines with specific features (the desirable ones you mention), and at the same time we're not willing to pay too much for them.

Their bread and butter is still the considerably larger gaming community who were happy, even eager to pay $350 for a 5850 when they were new, and the even larger 'low-end' market picking up GPUs that we wouldn't consider worth the shipping fee.

The simple fact is it's not currently worth it to develop a line of cards tailored to bitcoin mining.  Not to mention that revenue stream could dry up in an instant if bitcoin happened to fail hard.

Why the OEMs aren't scrambling to reduce the prices on otherwise slow moving 69XX series cards, or pumping the market with tried-and-true 58XX series cards, I don't understand.  They've already machined their tools to build them, gone through a few iterations to get things just right, and are likely high into profit on these cards.

Produce a lot more of them and they'll still sell faster than they can get to the shelves, and they don't even have to reduce their pricing model.  While I don't expect to see mining-specific cards, this is a no-brainer.

Then again, seeing Newegg selling $100 5830s, when they've been disappearing in minutes for $30 more, feels a little like market saturation testing by the OEM.  Perhaps Sapphire will heed my advice.

Don't underestimate the bandwidth involved here, even though it's relatively low for mining compared to gaming.  Up to 64Gbit for a 16x slot according to PCIe 2.0 specs.  In reality the GPUs can't throw a stone anywhere near that number, but it's something to be aware of when choosing your IC.   I don't know much about how the bandwidth is split between lanes, perhaps it's possible to bypass the IC for these data channels and let the motherboard handle it.

I'd assume you'd have to re-broadcast the clock signals to the connected devices, I think you're going to run into some latency problems here that will either be a major roadblock, or contribute to a lot of invalid hashes at least.

Note that i'm not trying to be a pessimist, just thinking this isn't going to be an easy task.

It'd help but it likely wouldn't be enough. 

Remember a single GPU is going to be dumping ~200W into the fluid continuously, that's a lot of heat in a fishtank.

Submersion will work well if you treat it more like traditional water-cooling.  Hermetically seal the whole thing (have fun replacing components!), and pump the fluid through a large radiator with fans.

That'll keep evaporation down and your lungs relatively free of mineral oil, while maintaining a relatively cool temperature.

It's probably not worth the trouble really.

There are expansion boards that can do this, as well as a couple of external PCIe enclosures, however they're all rather expensive.

Unfortunately it's not a plug-and-go matter of splicing in new connectors and sharing the bandwidth, it requires some logic ICs to handle multiplexing the signals to all the connected devices properly.

Most of that kind of thing is aimed at the high-end server crowd right now as the average user doesn't need or want to split out a 16x slot.  Someone producing such a thing cheaply would go over fairly well in this crowd.

SuperTramp, at that price i'd buy it just to have it around

Mouse, google "kill a watt", best ~$20 USD purchase you'll make for your miners.

Now just to be a dick, I'll scrutinize some more!


Olive oil??

You must've seen the mess coming

Most crazy enthusiasts use mineral oil instead - it's a hell of a lot cheaper too.


ETHANOL?!

You must've seen the explision and flames coming    Electricity + flammable liquids = inevitable disaster.

Sure, submerged in ethanol is probably fine as, like gasoline (petrol) it's the gasses that are actually combustible rather than the liquid.  Still, there are far better options.

I think you're going to have trouble selling any components that have been submerged in anything other than plain old air.