Hitler manages a FPGA design company
773 views5/17/2013
0 Comments
Be the first to leave a comment.
โธShow captions (49)
| 00:00 - 00:03 | Here are the timing reports from the 82 P&R runs we did |
| 00:04 - 00:05 | We have reached 720Mbps here, |
| 00:05 - 00:07 | 731 here, and 726 here. |
| 00:08 - 00:12 | From what we can tell, there are huge delays on the memory bus, |
| 00:12 - 00:15 | but beyond that, it's not clear what is going on. |
| 00:17 - 00:19 | Switch to GDDR5 SDRAM. |
| 00:19 - 00:21 | The NVIDIA folks can run it over 5Gbps/pin. |
| 00:24 - 00:26 | Sir... |
| 00:27 - 00:28 | The... |
| 00:31 - 00:33 | The maximum IO clock frequency is 1250MHz. |
| 00:34 - 00:36 | The PLL buffer does not go over that. |
| 00:53 - 00:58 | Everyone who programs in CUDA, get out now. |
| 01:13 - 01:15 | What the hell are they thinking? |
| 01:15 - 01:17 | Selling those busted shavings of shit |
| 01:18 - 01:23 | as if you could do any serious computation on them? |
| 01:25 - 01:28 | What do they think is going to happen? |
| 01:29 - 01:31 | They're fucked! |
| 01:31 - 01:34 | Look what happened to the Slowtan-6. |
| 01:34 - 01:37 | Crumbling from routing delays over 10 ns |
| 01:37 - 01:40 | Half the features crippled with 300% PVT variations |
| 01:40 - 01:42 | Sir, it's a low-cost FPGA. |
| 01:42 - 01:46 | Is that any excuse for software using the Wind/U toolkit? |
| 01:46 - 01:48 | Sir, we have hired 32 technicians to make sure it runs on our machines. |
| 01:48 - 01:52 | We can get it to work, but at what price? |
| 01:53 - 01:54 | Days wasted installing libstdc++5 and Motif |
| 01:56 - 01:57 | Randomly broken bitstreams after hours of compilation |
| 01:57 - 02:00 | Months tracking down simulation/synthesis mismatches |
| 02:00 - 02:03 | And all that for a design that is hopelessly slow? |
| 02:04 - 02:08 | Are the customers going to pay for that? |
| 02:08 - 02:13 | Meanwhile even the FORTRAN folks have it easy |
| 02:14 - 02:16 | Do you imagine them with a compiler |
| 02:17 - 02:21 | with an execution time increased a hundredfold? |
| 02:27 - 02:29 | which would half the time produce a binary that crashes |
| 02:30 - 02:34 | or when it runs on your board, does not on the customers' |
| 02:34 - 02:36 | or just because the temperature changed by one fucking degree? |
| 02:41 - 02:42 | Too much of this... |
| 02:43 - 02:47 | Too much of this running in circles. |
| 02:48 - 02:53 | Just throw FPGAs out the door and let the customers have a decent computer. |
| 02:54 - 02:56 | Yeah. That'll work. |
| 02:56 - 02:59 | Who do they think they're fooling? |
| 03:00 - 03:02 | The whole thing is a sham! |
| 03:04 - 03:07 | It's okay Anna, we can work for Intel. |
| 03:14 - 03:16 | The customers will have to do with GPUs and PCs |
| 03:19 - 03:23 | They'll feel satisfied with the results |
| 03:25 - 03:26 | I admit defeat. |
| 03:31 - 03:33 | We will make websites and iPhone apps now. |
| 03:40 - 03:46 | Instagram is already operating well in the Photography space |
| 03:46 - 03:49 | but perhaps the same basic ideas can be applied in the Knowledge space. |
| 03:53 - 03:56 | That's where we'll come in. |