There is a lot of bad data in this table. Pretty much all of the Radeons have incorrect Mh/s/$ values (eg. Radeon HD 6990, 800 Mh/s, $623 -> 1.28 Mh/s/$... but the table lists 1.46 Mh/s/$).

The prices are wrong too (Radeon HD 6990 sells for more than $623).

The wattages are wrong to (Radeon HD 6990 o/c'd to mine at 800 Mh/s will consume MUCH more than 346 Watt).

Etc.

Schleicher: very funny

Edit: I was privately contacted to explain my reaction to his post... Schleicher's post is obviously untrue. There are no such performance improvements in the GTX 680. I assume he was being humorous.

Your math is almost right, except this:

The Kepler SMX can execute 2x as many instruction as a 48-shader Fermi SM, but it can execute 3x as many instructions as a 32-shader Fermi SM (which the GTX 580 is an example of). Therefore do not multiply by 2, but by 3.

You forgot to divide the end result by 2 because, as you pointed out, the number of SMXs in the GTX 680 is half the number of SMs in the GTX 580.

Finally, a GTX 580 can mine at 149 Mh/s with a properly fine-tuned miner, not 140 Mh/s. (An entry in the wiki claims this with rpcminer-cuda.exe set to use a high aggression level parameter.)

Correcting these 3 mistakes give a GTX 680 estimation of: 149*3*1058/772/2 = 306 Mh/s... which matches exactly my first estimation (300-315 Mh/s).

The value of these benchmarks to predict Bitcoin performance is null. SiSoft publishes almost no information as to what actually their GPU code is doing. For all we know, they are probably memory-bound instead of ALU-bound. And the huge discrepancy disfavoring AES-128 on AMD, and SHA-256 on Nvidia suggests their code is poorly optimized in these respective cases and/or exposes flaws in the OpenCL compilers... There is no reason for SHA-256 hashing to be that slow on Nvidia (even if GTX 680 has no BFI_INT and no int rotate, it should be at least about half as fast as HD 7970). Conversely, there is no reason for AES-128 to be that slow on AMD.

Vbs, if this interpretation was true, then GTX 680 would be a 6180 single precision GFLOPS chip, but it is actually only 3090 SP GFLOPS...

Perhaps talking about GFLOPS is an easier way to convince you... In all recent Nvidia and AMD GPUs, the number of 32-bit integer instructions executable per clock is linearly proportional to the number of single precision floating point instructions executable per clock. (This was not true with the old GT200 because Nvidia counted GFLOPS assuming mul+mad or 3 fp operations/clock, anyway I digress.) Per Nvidia's published figures:

GTX 460 = 907 SP GFLOPS
GTX 680 = 3090 SP GFLOPS
And 3090/907 = 3.4x (or about the same 4x ratio I mentioned earlier)

Therefore GTX 680 would only mine 3.4x faster than the GTX 460 (again assuming no BFI_INT and int rotate).

What Nvidia mean by "twice the number of instructions per clock" is that a whole SM/SMX executes twice as many instructions per core clock:  a GTX 460's SM executes 48 (# of shaders) * 2 (shader clock is 2x the core clock) = 96 instructions per core clock; a GTX 680's SMX executes 192 (# of shaders) * 1 (shader clock same as core clock) = 192 instructions per core clock.

Yes. Two doublings. Number of cores quadrupled in an SM (now SMX). From 48 to 192. But I already take this into account in my calculation (192 cores in an SMX * 8 SMX = 1536 cores, which I base my numbers on). I think you are picking up isolated sentences without understanding the whole picture of how the GPU works.

Another way of seeing it, if I may grossly simplify, is that this Anandtech article says the GTX 680 approximately quadrupled the performance of a GTX 460 (which is approximately true: 4x the # of cores per SM/SMX, about same number of SM/SMX 7 vs 8, and the same shader clock 1006 vs 1350 MHz (+/- 30%)). GTX 460 mines at ~70 Mh/s, therefore GTX 680 would mine at ~280 Mh/s (+/- 30%), which is again consistent with my first estimation of 300-315 Mhash/s (assuming no BFI_INT or int rotate). This is a gross approximation but you get the idea.

Having 4 warp schedulers instead of 2 does not mean that you can execute twice the number of instructions per clock. It just means you have 4 warps of threads which take turn to be executed instead of 2.

In other words, divide your 576 Mh/s number by two: 288 Mh/s, which is within 4% of my first prediction (300-315 Mh/s).

Nvidia GTX 680 has 1536 ALUs running at 1006-1058 MHz. There are a few unknowns because AFAIK there is still no public information whether the Kepler architecture has a native instruction doing (a & b) | (~a & c) (aka BFI_INT on AMD GPUs, useful to implement maj() and ch() in SHA-256), and a native integer rotate instruction. Nonetheless, it is easy to calculate the upper bound of mining performance because it is an embarrassingly parallel workload, and we know how many instructions are required in these different scenarios:

• 300-315 Mh/s assuming the Kepler microarchitecture still lacks BFI_INT and int rotate
• 375-395 Mh/s if Nvidia added only int rotate
• 450-475 Mh/s if Nvidia implemented both BFI_INT and int rotate natively

This would still be worse than AMD's comparable HD 7970 card at stock clocks (554 Mh/s); but definitely a step in the right direction for Nvidia who finally recognized that having (much) more ALUs at a (slightly) lower clock is better than fewer ALUs at a higher clock.

Per the NDA lifted today, this was a blatant error. The ALUs will run at 1006-1058 MHz, which should allow this card to mine at an upper bound of 450-470 Mh/s (80-85% the speed of a HD 7970.) This is assuming of course thar Nvidia added a BFI_INT-like instruction to the architecture, which is not certain. If not, performance would be much lower...

Edit: for more accurate perf estimations see https://bitcointalk.org/index.php?topic=73627.0

Newegg listed "Shader clock: 2012MHz".

Interesting... 1536 ALUs at 2012MHz would give it more processing power than a HD 6990. Either the GTX 680 is going to be the fastest GPU for mining, or it is a blatant mistake from Newegg or the microarchitecture has undisclosed limitations that would prevent exploiting all this apparent power.

Hal sent me two archives of version 0.1.0 (a rar, and a tgz, both of the same source tree). I uploaded them to my site:

http://www.zorinaq.com/pub/bitcoin-0.1.0.rar
http://www.zorinaq.com/pub/bitcoin-0.1.0.tgz

SHA1:
ec9ed4ccbc990eceb922ff0c4d71d1ad466990dd  bitcoin-0.1.0.rar
35f83eaa334e0e447ceea77a7cc955a4ccdd1a1d  bitcoin-0.1.0.tgz

MD5:
91e2dfa2af043eabbb38964cbf368500  bitcoin-0.1.0.rar
dca1095f053a0c2dc90b19c92bd1ec00  bitcoin-0.1.0.tgz


All, please mirror them!

I certainly feel I was properly compensated.

Today, I would either do it the same way, or do it using the Kickstarter model like suggested on this thread. For those not familiar with it: http://en.wikipedia.org/wiki/Kickstarter
In 2 words: funds are pledged by potential customers. If a funding target is not reached by a certain date, money is returned to those who funded. Else the funds go to the seller who can finish producing and releasing the product.

I am the guy.

What happened to me? Well for a while I was selling my BFI_INT-enabled GPU miner. My strategy to prevent it from being leaked and pirated (I think the only strategy that can work) was to price it quite high (400 BTC which was about $200 at the time), so that the buyers would feel they had a valuable, exclusive product that they would not want to leak. AFAIK it worked and was never leaked. At the time I was 10-15% faster than the other open source miners.

Then the open source ones progressively caught up, so I discounted the price down to 300, then 250 BTC. And eventually I stopped selling it altogether.

It is about maximum utilization, and efficiency.

It does not matter if you buy only 1 rig. But the larger your scale, the more you should care about these details.

I would rather spend 1 or 2 grands in mining hardware than in installing extra circuits because the BFL designers chose an awkward wattage that prevents me from utilizing my circuits to their maximum capacity.

I am pretty sure 20A is more common than 15A, at least in all US residences built in the last few decades. My 30-year old apartment building has 20A circuits in the units. My office has 20A circuits. My previous apt had 20A circuits.

Anyway, perhaps the mini-rig should be sized for 640W. One could run 2 per 15A circuit with 160W headroom (eg. in an old residence where the circuit might be shared by a lightbulb). Or 3 per 20A circuit with no headroom (for those operating them on dedicated circuits).

BFL-Engineer, I would recommend to size the mini-rig either for 960W, or 1920W.
This way, either 1 or 2 could be placed on a standard 120V-20A circuit, while drawing no more than 80% of its rated capacity per the National Electric Code (1920W).

As discussed, they are almost certainly using a cheap source of old-gen 65nm FPGAs: https://bitcointalk.org/index.php?topic=66314.msg769355#msg769355

FLOPS is floating point operations per second. LINPACK FLOPS are just meaningful in the context of the TOP500 supercomputer rank because this organization decided to use this benchmark to establish the rank. A supercomputer might achieve "x" FLOPS on linear equations with LINPACK, "y" FLOPS when doing protein folding, and "z" FLOPS when doing some other work. The "x" FLOPS value of the LINPACK benchmark is no more significant than "y" or "z".

All are a fraction of "t", the theoretical peak FLOPS of the hardware (which can often be reached within 1-2% with a useless loop of multiply-add instructions). This theoretical peak can be used to predict the performance of SHA256-based Bitcoin mining because it scales linearly with the peak theoretical integer performance of a chip, which is itself directly related to its peak theoretical FLOPS performance by a fixed ratio. For example, there is exactly a 1:4 ratio between the number of integer and double precision floating point instructions that an HD 69xx series GPU can execute.

I would love it: m.bevand at gmail.com
Take your time, it is not urgent.