Intel's Linux Software Optimizations Continue Paying Off Big Time For Xeon Emerald Rapids

[onlyLinuxLuvUBack]

What about Steam on Intel's Linux?
Do games get any benefits?

it's like 8,000 usd for 1 xeon and gigabyte motherboard.
Michael Maybe there should be a part 2 article with 12th gen i5 vs 13th gen i5 vs 14th gen i5 for us poor people.

[dimko]

it's like 8,000 usd for 1 xeon and gigabyte motherboard.
Michael Maybe there should be a part 2 article with 12th gen i5 vs 13th gen i5 vs 14th gen i5 for us poor people.

Yeah... I shit the bed with this one.

[varikonniemi]

quoting is bugged on this platform...

[varikonniemi]

Well, the 2x loss only happened in a single benchmark and it led by a fairly wide margin in almost every other benchmark.

maybe not release software that leads to 50% reduction in performance?

[AdrianBc]

it's like 8,000 usd for 1 xeon and gigabyte motherboard.
Michael Maybe there should be a part 2 article with 12th gen i5 vs 13th gen i5 vs 14th gen i5 for us poor people.

While most Intel Xeon CPUs have ridiculously high prices when sold at retail, there are a few SKUs with decent prices.

For instance the Emerald Rapids CPU with 28 cores Xeon Gold 5512U has a recommended customer price of $1230.

This is a much better performance per dollar than any Epyc or Threadripper CPU (because unfortunately AMD has raised a lot its prices for server/workstation CPUs).

The champion for performance per dollar remains AMD Ryzen 9 7950X, but if you also want a lot of memory in a single box, starting from 128 GB (7950X with 128 GB is much slower than with 64 GB), then a 28-core Xeon 5512U is the best choice and the price for a DIY workstation or server remains decent (e.g. up to $3000 for a complete system), even 2 or 3 times less than when buying something equivalent from a big vendor and less than e.g. buying a much slower mobile workstation laptop.

[yump]

That is not a clue. Performance versus ondemand or schedutil will not result in benchmark results that eclipse the others by 3x.

MIchael's already done tests comparing the governors, performance doesn't win by massive margins. It has better results, but it always carries higher power draw and temperatures.
Incidentally because this an Intel processor, it is instead using intel_pstate which will override the governor and instead offer only powersave or performance. Michael's also already done tests on this and has shown that the intel_pstate powersave governor only loses by about 9% while at the same lowering CPU temperature by 10 degrees on average and saving about 60-80W on average in his test too.

You better watch out for your copium tanks, breathing in any harder might cause them to implode.

This is a dual-socket machine. The intel_pstate powersave governor uses hardware-controlled pstates. The CPU in one socket has no idea what the serially-dependent load is on the machine as a whole. The machine is enormously wider than non-file-parallel video encoders can utilize, and they are notoriously sensitive to CPU frequency scaling.

The results here are an obvious outcome of punching intel_pstate powersave right in its glass jaw.