7 ms·

I would argue that it is not a trivial upgrade of the capacity of the hardware (8 GB Ram, 2-3 Ghz x86_64) but the optimization and work on the underlying algori

I would argue that it is not a trivial upgrade of the capacity of the hardware (8 GB Ram, 2-3 Ghz x86_64) but the optimization and work on the underlying algorithms and hardware. Furthermore, the cost associated with introducing the new products might not be currently sustainable. An ASIC can perform some of the computational tasks you mentioned (such as image recognition) with much better thermal/performance levels; it might be hard to make it a commercial success.

yourapostasy 10y ago
These applications absolutely depend upon more algorithmic and software engineering work, and it definitely is not a matter of just adding more hardware; if it was, I'd have the rack or two (or more?) of hardware it could take today. It's hard to predict markets, but in the Open Source world there are at least primordial forms of every application I mentioned, or at least the elements of them. It's fun to tinker with them at the very least, while the hardware keeps improving. It's like back in the 80's it was neat to imagine GB-scale disks with ubiquitous, continuously-indexed search across all files on the system. We could even implement it, and tinker with it on a toy scale for practical speeds, or on a system-wide scale at slow demo speeds. It took better hardware before that feature was really practical. I'm wondering if ASIC-based products will come out for "settled" niches that large swathes of the workstation market agree upon for a kind of standardized co-processing set of sockets/slots in workstations, and workstation-embedded FPGA's as software-reprogrammable become available for niches that are not as settled, but not quite amenable to GPU/CPU processing. I don't know enough yet about hardware at that level to discern what field of application could possibly penetrate mainstream computing that would need those devices at such a low level, though, that couldn't be addressed with PCIe cards today, if not through CUDA/OpenCL for accelerators/co-processors and general-purpose CPUs.

yourapostasy 10y ago

These applications absolutely depend upon more algorithmic and software engineering work, and it definitely is not a matter of just adding more hardware; if it was, I'd have the rack or two (or more?) of hardware it could take today. It's hard to predict markets, but in the Open Source world there are at least primordial forms of every application I mentioned, or at least the elements of them. It's fun to tinker with them at the very least, while the hardware keeps improving. It's like back in the 80's it was neat to imagine GB-scale disks with ubiquitous, continuously-indexed search across all files on the system. We could even implement it, and tinker with it on a toy scale for practical speeds, or on a system-wide scale at slow demo speeds. It took better hardware before that feature was really practical. I'm wondering if ASIC-based products will come out for "settled" niches that large swathes of the workstation market agree upon for a kind of standardized co-processing set of sockets/slots in workstations, and workstation-embedded FPGA's as software-reprogrammable become available for niches that are not as settled, but not quite amenable to GPU/CPU processing. I don't know enough yet about hardware at that level to discern what field of application could possibly penetrate mainstream computing that would need those devices at such a low level, though, that couldn't be addressed with PCIe cards today, if not through CUDA/OpenCL for accelerators/co-processors and general-purpose CPUs.