Discussion Intel Binary Optimization Tool thread

[inquiss]

So using the GPU driver analogy, if a GPU driver makes a GPU look better in a gaming benchmark, is that bad?

Again, this is not my wheelhouse whatsoever, so if the analogy of a GPU driver is not anywhere near representative, then I stand corrected but would like to understand how IBOT is different.

This is a driver that only works to improve framerates for the benchmark portion of the game but then doesn't boost performance when you play the actual game.

 

[Saylick]

This is a driver that only works to improve framerates for the benchmark portion of the game but then doesn't boost performance when you play the actual game.

Yikes, then that's bad. Deceptive, even.

 

[511]

In real world apps, not unfair. It’s good.

In benchmarks, bad

Well except for Geekbench iBOT is just for games so real world apps

 

[511]

This is a driver that only works to improve framerates for the benchmark portion of the game but then doesn't boost performance when you play the actual game.

Are you sure though cause than it's straight up cheating.

 

[igor_kavinski]

but would like to understand how IBOT is different.

Like the GB developer showed, Intel executed an extra 17 billion AVX2 instructions via iBOT. If AMD executed those too, then Intel wouldn't look as fast.

Just waiting for the LOL moment when someone cracks iBOT and makes it work on AMD too.

 

[gdansk]

Home

Contribute to intel/thin-layout-optimizer development by creating an account on GitHub.

github.com

Possibly an outgrowth of that Intel project. But it seems to be making it more difficult to add general optimizations because it doesn't disassemble like BOLT.

 

[igor_kavinski]

^^^ That's a HUGE repo!

 

[Nothingness]

Home

Contribute to intel/thin-layout-optimizer development by creating an account on GitHub.

github.com

Possibly an outgrowth of that Intel project. But it seems to be making it more difficult to add general optimizations because it doesn't disassemble like BOLT.

With a major twist: it doesn't only change code layout, it also "recompiles" some of the functions.

 

[NTMBK]

I look forward to our first bugs when an "optimized" binary has subtly different behaviour to what we actually shipped 🙃

 

[Doug S]

I look forward to our first bugs when an "optimized" binary has subtly different behaviour to what we actually shipped 🙃

I wonder if some OEM will sue Intel for modifying the behavior of their software without their consent? Whether or not there's subtly different behavior, it is one more variable OEMs have to take into account when providing software support to their customers. I bet one of the first things you'll get asked is "do you have IBOT enabled?" if you say yes they'll say "please disable it and see if you can reproduce this behavior".

I wonder if Intel is going to provide a way for OEMs to opt out? Like if John Poole didn't want Geekbench modified in this way, is there any way (short of a lawsuit) to prevent Intel from doing so? Can you mark a binary in some way "do not IBOT me"?

 

[igor_kavinski]

I wonder if Intel is going to provide a way for OEMs to opt out? Like if John Poole didn't want Geekbench modified in this way, is there any way (short of a lawsuit) to prevent Intel from doing so? Can you mark a binary in some way "do not IBOT me"?

They should simply opensource it and then let the community take over. Then they are basically off the hook because a bunch of anonymous, virtually untraceable randos did the tuning and whoever uses those EXEs understands the risks.

 

[511]

They should simply opensource it and then let the community take over. Then they are basically off the hook because a bunch of anonymous, virtually untraceable randos did the tuning and whoever uses those EXEs understands the risks.

Well that ain't going to happen AMD will simply run away with that which they don't want

 

[igor_kavinski]

Well that ain't going to happen AMD will simply run away with that which they don't want

AMD won't blatantly copy that. But the community optimizing binaries for AMD CPUs is a possibility. Again, it won't have a huge impact because I don't think reviewers will use community binaries for benchmarks.

 

[511]

AMD won't blatantly copy that. But the community optimizing binaries for AMD CPUs is a possibility. Again, it won't have a huge impact because I don't think reviewers will use community binaries for benchmarks.

You need skill to copy AMD continues to have software skill issue even now it would be someone from community

 

[igor_kavinski]

By the way, @511 thanks for pointing out in another thread that Nova Lake is a 12-wide core. That absolutely demands newly compiled executables for extracting maximum ILP from the new architecture. iBOT will be absolutely essential to fighting a potential 6.5 GHz Zen 6 part. Though sadly, only a few battles will be won and Intel may again lose the war.

 

[gdansk]

That absolutely demands newly compiled executables for extracting maximum ILP from the new architecture

Not really. The point of a big frontend is to exploit more existing ILP.

If they have to recompile software for it, it's worse than every other braniac in the industry.