4 ms·
Well I got down to the part where the author said that AMD never threatened Intel in the data center market and I closed the tab. AMD won entire generations of
by jeffbee 6y ago
Well I got down to the part where the author said that AMD never threatened Intel in the data center market and I closed the tab. AMD won entire generations of data center orders while Intel was flailing with Itanium and NetBurst.
- mhh__ 6y agoSpeaking of Itanium, if the x86 dam has truly burst, I'd much rather see something more like the Itanium than RISC-V. Something new. It's a shame the Mill is so secretive, actually, they're design is rather nice.
- sitkack 6y agoOne of RISC-V's main goals is to be boring and extensible. Think if it as the control-plane core, or the EFI for a larger system. You would take RISC-V and use it drive your novel VLIW processor.
- mhh__ 6y agoHow? RISC-V will have to have memory model, for example, which will define some at least effective execution model. If you turn RISC-V into not RISC-V you might as well just start from scratch.
- garethrowlands 6y agoPretty sure you can't take RISC-V and use it to drive a Mill.
- raverbashing 6y agoNah I think the Itanic concept is dead in the water VLIW works (especially in the way it was done in Itanium - IIRC) when either your workload is too predictable or maybe if your compiler manages to be one order of magnitude smarter than it is today (even with llvm, etc) It seems even M1 prefers to reorder scalar operations than work with SIMD ops in some cases (this is one of its processors)
- mhh__ 6y agoItanium is dead but VLIW as a concept is still interesting to me. If you look at uops executed per port benchmarks you can see that CPUs are far from all seeing eyes.
- hajile 6y agoAMD and Nvidia both used VLIW in the past and both moved away because they couldn't get it to run efficiently. If embarrassingly parallel problems can't execute efficiently on VLIW architectures, I somehow doubt that CPUs will either. The final versions of Itanic started adopting all the branch predictors and trappings from more traditional chips. The problem is that loops theoretically cannot be completely predicted at compile time (the halting problem). Modern OoO CPUs are basically hardware JITs that change execution paths and patterns on the fly based on previous behavior. This (at least at present) seems to get much better data resulting in much better real-world performance compared to what the compiler sees.
- garethrowlands 6y agoMill is claimed to run general purpose code well, unlike Itanic and VLIW in general. Are you claiming Mill would be like Itanium?
- tyingq 6y agoThere's Russia's Elbrus VLIW chip. https://www.anandtech.com/show/15823/russias-elbrus-8cb-microarchitecture-8core-vliw-on-tsmc-28nm https://www.anandtech.com/show/15823/russias-elbrus-8cb-micr...
- secondcoming 6y agoGCP only offer Epyc CPUs in some regions. None of those regions are ones we use! Gah! Can someone update us on where AWS offer them, if at all?
- jeffbee 6y agoIf you think about how these providers deploy a cloud facility, it makes sense that the offerings in a given place are relatively static. The whole network design, thermal/mechanical design, and floor plan is built with certain assumptions and they can't just go in and rack up some new machines. It evolves pretty slowly and when a facility gets a new machine it is because they refresh the whole thing, or a large subset of it. That said, the EPYC machine type is available in 12 zones of four different regions in the US, which isn't bad.
- vinay_ys 6y agoUsually you would have some number of enclosed aisles of racks make up a deployment pod. You can usually customize machine configuration within a deployment pod while staying within the electrical and thermal envelope of the aisle and without changing the number of core-spine to pod-spine network links. You could potentially build out a data hall but not fully fill it with aisles. As demand starts to trend up you can forecast two quarters into future and do the build-outs with just one quarter lead time. I would expect very large operators to have perfected this supply chain song and dance very well.
- jeffbee 6y agoThey have perfected it, just not in the manner that you are suggesting.