hashcat
advanced password recovery

We got our first shipment of GTX Titan X GPUs today! Been playing around with them for the past 8 hours, and holy fuck, this card does not disappoint!

https://twitter.com/jmgosney/status/590670001021054978

• 60% faster than AMD Radeon R9 290X at stock clocks – up to 85% faster with +225 Mhz OC.
  
• Measured maximum 273W draw from the wall (100% load, 100% fan, +225 Mhz OC). PSU is 86% efficient, so card itself is drawing ~ 234W. That's about 17% less than the R9 290X.
  
• First single GPU to break 35 GH/s on NTLM. Can actually break 40 GH/s with +275 Mhz OC, but it's not stable for all algorithms at this clockrate.
  
• Max observed GPU temperature is 64C at 100% fan (22.7C ambient temp, no external air flow)
  
• Hands-down the best GPU you can buy for Hashcat right now!
  

Benchmarks -- Stock Clocks



Benchmarks -- +225 Mhz OC



We're waiting on more chassis to arrive to test this in the Sagitta Brutalis. We should have 8x Titan X benchmarks posted by the end of the week, but the single GPU benchmarks show that a single Sagitta Brutalis with 8x Titan X will be nearly as fast as our old 25-GPU VCL cluster combined!

Stay tuned...

nice!
how you connect 8 gpu in one computer?

Congratulations, Jeremi!

If the core clock during cracking process is indeed 1.215 GHz-ish, then my predictions in this thread were true.
On some algorithms, however, Titan X is performing better than expected (LM +76% vs 980).
NV did a good job by ditching FP64 and focusing solely on int performance.
Oh, and I expected Titan X to be hotter.

(04-22-2015, 07:12 AM)bobva Wrote: how you connect 8 gpu in one computer?

You need a big computer I mentioned a specific model in my post; Google it.

(04-22-2015, 07:45 AM)Rolf Wrote: If the core clock during cracking process is indeed 1.215 GHz-ish, then my predictions in this thread were true.

As you know, Maxwell has very aggressive dynamic frequency scaling, so the cards rarely run at a consistent clock rate. But yes, at "stock clocks" (i.e., not setting the Clock Offset), the clock mainly hover around 1127 ~ 1216 Mhz.

(04-22-2015, 07:45 AM)Rolf Wrote: On some algorithms, however, Titan X is performing better than expected (LM +76% vs 980).

Hmm, I didn't even notice that. That is indeed a healthy gain. I suspect this might have something to do with the larger memory bus since DES makes heavy use of s-boxes.

(04-22-2015, 07:45 AM)Rolf Wrote: Oh, and I expected Titan X to be hotter.

Everyone did, and I do not know why! It doesn't draw enough power for it to be as warm as everyone said it was... Maybe if you leave the fan at 20% you can heat it up?

i google it and not found...

(04-22-2015, 08:52 AM)bobva Wrote: i google it and not found...

You should definitely improve your Google foo. It's the very first entry you get when you search the terms. For those people who wants to know if they can afford it:

Base configuration price: 18,499 USD

Comparing stock Titan X with overclocked GTX 980

Summary: Titan X at stock clocks is ~30% faster on most algorithms, the only exception is scrypt, which is slower by ~33%.
This is expected, since scrypt kernel of 980 is highly optimized, whereas the titan x kernel is very generic.

@Jeremi: Thanks! And I actually wanted to ask you yesterday if there are any news on the Titan front

@Rolf: great work, these comparisons are gold, we actually should have something like a database, although keeping everything in a contant testing enviroment is a pain in the darkzone.

Quote:

(04-22-2015, 07:45 AM)Rolf Wrote: Oh, and I expected Titan X to be hotter.

Everyone did, and I do not know why! It doesn't draw enough power for it to be as warm as everyone said it was... Maybe if you leave the fan at 20% you can heat it up?

Maybe there's an explanation for that. Here's a picture from a well known Hardware site, showing the back of a Titan X under full load:




As you can see the memory chips are the hottest part of the board with over 100°C. Since hashcat, especially in benchmark mode, doesn't need much memory, leaving the 12GB of the Titan X merely unused. Since years cards with GDDR5 RAM have the ability to slow/clock down their memory to save power. So the conclusion would be:

1. easy use on memory clocks most memory chips down using less energy
2. since memory chips are the hottest parts, the card runs cooler now

The Titan X has 16 more chips than the GTX 980, so there you go.

But that's just my 2ct.

Well damn, and I thought I was going to be the bad kid on the block with the 8 980s I just got....

Guess I just have to start making the case for an upgrade already on next years budget.

The upcoming 980Ti will use the same chip, has only 6GB of RAM (still very fine for Hashcat) and speed should be at least the same if not higher.

Next year a new kid will rock up the place named Pascal.

But I'm a bit disappointed that the Titan X is only 30% faster than the GTX 980. Do the kernels need a specific upgrade or do they run on different clock speeds? I mean the ACTUAL clock speed, not some maybe boost rate.