4 ms·

> and it would seem to be in Intel's interest to encourage more research in to how to take full advantage of them. This seems a little overstated -- we're talk

> and it would seem to be in Intel's interest to encourage more research in to how to take full advantage of them. This seems a little overstated -- we're talking about CPUs that are widely publicly available for around $300 USD, not some big iron.

nkurz 11y ago
I don't think it's overstated: Intel would benefit from published research that plays to the strength of their newer processors, and there is a lot of published CS research based on older processors because the researchers don't have access to the newer machines. The CPU is several hundred dollars, but since it requires a different motherboard and DDR4 RAM, it would be over $1000 to put together a test system to meet our needs. The norm is that it just doesn't happen, and that people publish using a single machine that they replace every couple years. Consider the plethora of articles published for the last couple years of showing the very modest improvements in processor performance running existing software. Then consider the value of just a couple high profile cases of people showing 2x-4x gains on newer processors using algorithms that target them. I think that value would be much higher than the cost of providing access, and that the absence of such articles is because the hassle and cost of testing on newer processors is still quite high. Consider the innovative project I just posted here: http://cglab.ca/~morin/misc/arraylayout-v2/ http://cglab.ca/~morin/misc/arraylayout-v2/ Great project, legitimate research, but no numbers for any Broadwell or Skylake processors. Or Power8 for that matter.
- jonah 11y ago
  Some time back I was working at a company that made very popular cutting edge graphics software. Some of the key developers had beige-box machines labeled "Property of Intel" and "Confidential". Inside were Intel motherboards populated with pre-release Pentium III CPUs - also etched "Intel Confidential". These were Katmai machines which the engineers were using to implement SSE instructions[1] in the software. So, yes, Intel does supply hardware to key developers. I have no idea what the criteria is, but next time, talk to your connections and see if your company or research group qualifies. [1] https://en.wikipedia.org/wiki/Streaming_SIMD_Extensions https://en.wikipedia.org/wiki/Streaming_SIMD_Extensions
- petercooper 11y ago
  If the value really is there, that seems like a reason for Intel to have a program where they give hardware to researchers. I mean, they do this for reviewers who clearly generate value for the company, so why not CS researchers?
paulsutter 11y ago
A comparative benchmark necessarily requires access to a range of processors, on a routine but rare basis. It really is inconvenient to buy and manage them all. For starters, where would I put them? More software developers would take advantage of the special features in processors, if Intel provided easier access. I've been curious to try features and make comparisons, but it's just a couple steps too much hassle.

nkurz 11y ago

I don't think it's overstated: Intel would benefit from published research that plays to the strength of their newer processors, and there is a lot of published CS research based on older processors because the researchers don't have access to the newer machines. The CPU is several hundred dollars, but since it requires a different motherboard and DDR4 RAM, it would be over $1000 to put together a test system to meet our needs. The norm is that it just doesn't happen, and that people publish using a single machine that they replace every couple years. Consider the plethora of articles published for the last couple years of showing the very modest improvements in processor performance running existing software. Then consider the value of just a couple high profile cases of people showing 2x-4x gains on newer processors using algorithms that target them. I think that value would be much higher than the cost of providing access, and that the absence of such articles is because the hassle and cost of testing on newer processors is still quite high. Consider the innovative project I just posted here: http://cglab.ca/~morin/misc/arraylayout-v2/ http://cglab.ca/~morin/misc/arraylayout-v2/ Great project, legitimate research, but no numbers for any Broadwell or Skylake processors. Or Power8 for that matter.

jonah 11y ago
Some time back I was working at a company that made very popular cutting edge graphics software. Some of the key developers had beige-box machines labeled "Property of Intel" and "Confidential". Inside were Intel motherboards populated with pre-release Pentium III CPUs - also etched "Intel Confidential". These were Katmai machines which the engineers were using to implement SSE instructions[1] in the software. So, yes, Intel does supply hardware to key developers. I have no idea what the criteria is, but next time, talk to your connections and see if your company or research group qualifies. [1] https://en.wikipedia.org/wiki/Streaming_SIMD_Extensions https://en.wikipedia.org/wiki/Streaming_SIMD_Extensions
petercooper 11y ago
If the value really is there, that seems like a reason for Intel to have a program where they give hardware to researchers. I mean, they do this for reviewers who clearly generate value for the company, so why not CS researchers?

paulsutter 11y ago

A comparative benchmark necessarily requires access to a range of processors, on a routine but rare basis. It really is inconvenient to buy and manage them all. For starters, where would I put them? More software developers would take advantage of the special features in processors, if Intel provided easier access. I've been curious to try features and make comparisons, but it's just a couple steps too much hassle.