TalkChess.com

Graham Banks wrote:

APassionForCriminalJustic wrote:.......AMD is garbage. Komodo is 100 percent developed with Intel CPUs.

I don't think that statement is fair on two fronts, firstly because it is a big putdown to those using AMD and secondly because engine authors should realise that many engine users will use AMD architecture, just as some have more modern computers than others.

I think it has been proven previously that Komodo and other engines don't suffer through running on AMD?

Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

Wolfgang wrote:

Graham Banks wrote:

APassionForCriminalJustic wrote:.......AMD is garbage. Komodo is 100 percent developed with Intel CPUs.

I don't think that statement is fair on two fronts, firstly because it is a big putdown to those using AMD and secondly because engine authors should realise that many engine users will use AMD architecture, just as some have more modern computers than others.

I think it has been proven previously that Komodo and other engines don't suffer through running on AMD?

Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

A 16-core Opteron is probably about equivalent to a 4-6 core i7. I wouldn't put much stock in this as an equivalent test to TCEC conditions.

Head to head, I might take your Haswell 4770K overclocked vs the Opteron in this test.

Wolfgang wrote: Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

I just ignore comments like that as a rule. I'm perfectly comfortable with my choice of hardware, its strengths and limitations, and the various factors that I weighed up in purchasing it. I don't care what anyone else thinks of it.

Modern Times wrote:

Wolfgang wrote: Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

I just ignore comments like that as a rule. I'm perfectly comfortable with my choice of hardware, its strengths and limitations, and the various factors that I weighed up in purchasing it. I don't care what anyone else thinks of it.

Modern Times wrote:...
I just ignore comments like that as a rule.

the best one can do!

Modern Times wrote:

Wolfgang wrote: Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

I just ignore comments like that as a rule. I'm perfectly comfortable with my choice of hardware, its strengths and limitations, and the various factors that I weighed up in purchasing it. I don't care what anyone else thinks of it.

Hi Ray,

Over on fishtest, a suggestion is being raised that AMD chips indeed behave quite differently; that is, even once you control for NPS, details of the architecture make the relative performance of intel and AMD quite different (stronger in some workloads, weaker in others).

I find this suggestion quite fascinating. It is not yet established in any concrete sense, though I'm sharing it since it may go a good way to explain discrepancies between AMD and intel testing. If true, this issue is quite sensitive since komodo is developed on intel (i.e., it may even be why some people match K9.3's expected improvement, and others do not).

Again, I have only anecdotal information on this. Perhaps someone with both an intel and an AMD chip can run a pair of tournaments, first on one chip and then on the other, where (for instance) the TC is chosen to give SF equal NPS on each.

fenchel wrote:

Modern Times wrote:

Wolfgang wrote: Hi Graham,

don't care what this "great expert" says. Even my i7-4770k is "garbage" as he wrote here, http://talkchess.com/forum/viewtopic.ph ... &start=160

If an engine would really suffer significantly from an architecture, it's the engine which is garbage, not the machine...

I just ignore comments like that as a rule. I'm perfectly comfortable with my choice of hardware, its strengths and limitations, and the various factors that I weighed up in purchasing it. I don't care what anyone else thinks of it.

Hi Ray,

Over on fishtest, a suggestion is being raised that AMD chips indeed behave quite differently; that is, even once you control for NPS, details of the architecture make the relative performance of intel and AMD quite different (stronger in some workloads, weaker in others).

I find this suggestion quite fascinating. It is not yet established in any concrete sense, though I'm sharing it since it may go a good way to explain discrepancies between AMD and intel testing. If true, this issue is quite sensitive since komodo is developed on intel (i.e., it may even be why some people match K9.3's expected improvement, and others do not).

Again, I have only anecdotal information on this. Perhaps someone with both an intel and an AMD chip can run a pair of tournaments, first on one chip and then on the other, where (for instance) the TC is chosen to give SF equal NPS on each.

Actually, Ray, I believe you yourself have even commented on this in the past! So I apologize for sharing information you know better than I.

Even so, maybe it'd be interesting if someone performance some concrete comparisons of this type and the discussions can be pointed to when people hate on different setups.

fenchel wrote: Over on fishtest, a suggestion is being raised that AMD chips indeed behave quite differently; that is, even once you control for NPS, details of the architecture make the relative performance of intel and AMD quite different (stronger in some workloads, weaker in others).

Different generations of Intel chip may well behave quite differently from each other, and also different generations of AMD as well. So it vastly more complex than saying "Intel may behave differently from AMD". A prime example with AMD is the more conventional Phenom II X6 architecture vs the Piledriver - very very different CPUs. Far more different probably than any "Intel" vs "AMD" comparison. With AMD I generally prefer the X6 for chess for consistency of results.

I don't believe there has ever been any conclusive proof of ratings differences with current engines on ratings lists. Any difference is probably so small it is hidden in the statistical margins of error. But in fishtest where you release a change based on a gain of just a couple of Elo, then you may have a problem.

Modern Times wrote:

fenchel wrote: Over on fishtest, a suggestion is being raised that AMD chips indeed behave quite differently; that is, even once you control for NPS, details of the architecture make the relative performance of intel and AMD quite different (stronger in some workloads, weaker in others).

Different generations of Intel chip may well behave quite differently from each other, and also different generations of AMD as well. So it vastly more complex than saying "Intel may behave differently from AMD". A prime example with AMD is the more conventional Phenom II X6 architecture vs the Piledriver - very very different CPUs. Far more different probably than any "Intel" vs "AMD" comparison. With AMD I generally prefer the X6 for chess for consistency of results.

I don't believe there has ever been any conclusive proof of ratings differences with current engines on ratings lists. Any difference is probably so small it is hidden in the statistical margins of error. But in fishtest where you release a change based on a gain of just a couple of Elo, then you may have a problem.

Thanks for the comments! I see now that my post was quite naive..