5970 2GB vs 5970 4GB???

[gigabytecoin]

Can anybody comment as to their Mhash/S rates for either of these two cards?

Are they the exact same?

Does the size of the memory on a card matter at all for the hashing calculations we are performing?

[true]

Bitcoin mining doesn't need these huge amounts of RAM. With no other differences, either should give the same hashrate.

[gigabytecoin]

true, thanks!

Can anybody else attest to this assumption?

I have the cash and the desire to purchase the fastest hashing board that I can.

Has anybody purchased the 5970 4GB? What is your average hashrate?

[marcus_of_augustus]

Anybody had a go with the older 5970 1 GB MB mem boards?

http://www.i-love-pc.com/ProductDetail.asp?ProductNO=3169&utm_source=myshopping&utm_medium=cpc&ref=myshopping

[true]

Anybody had a go with the older 5970 1 MB mem boards?

There were 1MB boards? Maybe the code fits in that space, but you might have to run at 640x480 8bpp or less to reduce the framebuffer size and make room...

That said, as long as core clockspeeds are the same, there won't be a hashrate difference between 1GB, 2GB or 4GB boards. The data being hashed is very small, and the hashes (except those with difficulty at least 1) aren't stored. There's no reason to concern yourself with RAM size unless you intend to play games with very high AA or high resolutions.

[urizane]

1 GB 5970's do not exist.  However, 2 GB boards have 1 GB attached to each individual GPU.  With the internal crossfire on the 5970, Windows would report the card having 1 GB of framebuffer, but only because crossfire connected GPUs must replicate their framebuffers.  The same can be said for nVidia SLI (GTX 295 1792MB boards have 896MB on each GPU).

That said, as long as core clockspeeds are the same, there won't be a hashrate difference between 1GB, 2GB or 4GB boards. The data being hashed is very small, and the hashes (except those with difficulty at least 1) aren't stored. There's no reason to concern yourself with RAM size unless you intend to play games with very high AA or high resolutions.

I'm using a CUDA miner on an nVidia GPU and the framebuffer usage tends to go up by only 30 MB.  I imagine ATI GPUs will be at least somewhere near that on an OpenCL miner, so getting a 4 GB board wouldn't help mining and may not even be worth it unless you game with multiple monitors simultaneosly (Eyefinity) or run a single 2560x1600 panel in games.

[Grinder]

The only thing more memory will have an impact on when you're mining is your electricity bill.

[urizane]

The only thing more memory will have an impact on when you're mining is your electricity bill.

Yes, but when a video card switches from 128 MB chips to 256 MB chips, typically they use higher density chips.  The difference in power is more like 5W.  It's miniscule compared to the 300W the card is rated for at stock clocks.  However, if gaming isn't a concern, the extra cash spent on getting the 4GB card in the first place wouldn't make sense anyway.

[Grinder]

Yes, but when a video card switches from 128 MB chips to 256 MB chips, typically they use higher density chips.  The difference in power is more like 5W.

It's probably a lot more than that. http://www.overclockersclub.com/reviews/sapphire_toxic_hd5970/18.htm

[urizane]

Yes, but when a video card switches from 128 MB chips to 256 MB chips, typically they use higher density chips.  The difference in power is more like 5W.

It's probably a lot more than that. http://www.overclockersclub.com/reviews/sapphire_toxic_hd5970/18.htm

On the first pair of graphs, they're comparing a Sapphire Toxic 4 GB (BTW, the Toxic line means it's overclocked to 900 MHz) to a Sapphire 2 GB card (which is not overclocked by default).  The GPU's themselves make up most of the difference in power.

On the second pair of graphs, they've overclocked both sets of cards again.  The 5970 4 GB card is overclocked to 966 MHz and the 2 GB card is overclocked to 890 MHz.  If you look at the standard 5970 2 GB card when it's overclocked vs the Sapphire Toxic 5970 4 GB (which itself is already overclocked to 900 MHz), the power is similar (actually slightly higher for the 2GB card overclocked).

Read page 4 of that same article to see what I'm talking about.  The difference in wattage is not based on RAM, it's the GPUs.

Also, the final paragraph on page 18 (the page you linked to) states this fact, as well.

EDIT: I took one more look.  The only appreciable difference is idle wattage.  Load wattage is too similar to make a difference on the electric bill.  (654W for the 2 GB card at 890 MHz vs 648W for the 4 GB card at 900 MHz)

[Grinder]

Should have noticed that. Guess I'll have to agree that it's very similar, then.

[urizane]

Yeah, RAM chips have very little in the way of logic, so power draw at the RAM chip is very little.  Memory controllers, on the other hand, can consume great amounts of power at super high clock rates.  For example, my nVidia GTX 460 at load clocks the RAM at 1000 MHz and one power state below that is 162MHz.  Once it drops to that power state with no load on the GPU before or after that transition, my system's overall power consumption drops roughly 25W per card.  Chalk that up to misappropriation of fast (and therefore leaky) transistors in places where it would not be proper.  The GTX 560 Ti apparently makes up for this by some amount, but I don't have the numbers on hand.  I don't think AMD ever had this problem, seeing as how the 5870 uses a 1200 MHz clock for memory.

EDIT: I should point out that there are actually 4 power states for the overclocked GTX 460s that I have.  (1) 810 core/1000 memory, (2) 405 core/1000 memory, (3) 405 core/162 memory, (4) 51 core/67.5 memory
It's the transition between (2) and (3) that drops 25W off of each card, indicating that the memory clock is the cause for the drop.  Since the chips themselves aren't much to power, the memory controller must be the issue.

[snedie]

I'm running two 5970's at 900Mhz clock and get a steady 1.3Ghash/s, so you could expect 6-700Mhash from a single card if overclocked.

At stock speeds I get 1Ghash/s.