AMD Kaveri vs. Intel Skylake With The Latest Linux/Mesa Open-Source Drivers

[pal666]

No. I meant the following: I did take a look at the Mesa source code and (by measurement) found a piece of code that is slow and appears to be optimizable.

you can get any peace of code, find slow part and then nothing prevents it from appearing optimizable to you

[haplo602]

He has plenty of Intel laptops to compare against.

can you link one where he does ?

[linuxgeex]

This benchmark is fundamentally flawed in two ways.

First, from the get-go, the clocks and timings for the memory modules are not disclosed. IGP performance is directly proportional to the bandwidth of the memory modules, so omitting this information invalidates the benchmark immediately.

Second, the games tested are CPU-bound, not GPU-bound, so this isn't even a graphics performance benchmark. It's a CPU benchmark.

phoronix, Michael I am very disappointed, you are slipping.

[Michael]

This benchmark is fundamentally flawed in two ways.

First, from the get-go, the clocks and timings for the memory modules are not disclosed. IGP performance is directly proportional to the bandwidth of the memory modules, so omitting this information invalidates the benchmark immediately.

Second, the games tested are CPU-bound, not GPU-bound, so this isn't even a graphics performance benchmark. It's a CPU benchmark.

phoronix, Michael I am very disappointed, you are slipping.

It can always be assumed that the memory modules are of the rated frequencies for each particular CPU and using the given number of channels available. I've mentioned it before and can pretty much always be assumed that way unless explicitly otherwise mentioned. It's not displayed in each article due to dmidecode not working unless root. PTS will report said information when it's run as root, but not when it's run as a normal user, and not aware of any other way on Linux to obtain the RAM information outside of root dmidecode.

[linuxgeex]

It can always be assumed that the memory modules are of the rated frequencies for each particular CPU and using the given number of channels available. I've mentioned it before and can pretty much always be assumed that way unless explicitly otherwise mentioned. It's not displayed in each article due to dmidecode not working unless root. PTS will report said information when it's run as root, but not when it's run as a normal user, and not aware of any other way on Linux to obtain the RAM information outside of root dmidecode.

Then feel free to disclose that in the article: "PTS ran the suite 10x to be thorough, but the numbers are +-50% depending on what RAM I left in the machines."

I'm sorry but this is just plain sloppy. Tell your visitors what exactly "of the rated frequencies for each particular CPU" actually means. It means anywhere from DDR3 1600 to 2400+. That's a massive amount. It will completely throw the results, but your professionalism doesn't drive you to control or report it. Interesting.

[bridgman]

I believe Michael meant "the maximum rated frequencies for each particular CPU".

[linuxgeex]

I believe Michael meant "the maximum rated frequencies for each particular CPU".

I'd like to take that interpretation. I really would. However there's a few things preventing me. First, Michael doesn't equip every machine in his stack with enthusiast-level hardware. There's a variety of reasons for that, ranging from running what he gets as review samples to running what he bought at the time, before higher spec parts were available, and lastly, the lack of a bottomless wallet. So saying "of the rated frequencies for each particular CPU" is extremely vague.

Respectable hardware review sites like TGH and AnandTech always report the frequencies of the parts.

All I'm saying here is that Michael seems to think a vague comment about a general policy is good enough for his viewers.

He has a lower standard. I am pointing it out. You are free to dislike that I am pointing it out, but it is a factual statement until Michael adopts a higher standard.

[bridgman]

I don't care what you point out

I was just thinking that you might be interpreting a typo (missing "maximum" in the post to which you were responding) as a lack of policy when Michael's statements suggested otherwise.