The entire benchmark "doesn't really need to be". It's just beneficial. It's purpose is to collect data that we can then use to draw conclusions from. And the more data it collects the better it serves that purpose and the less assumptions we have to make when extrapolating. Please do not say that it's fulfilled its purpose or that any further data wouldn't be helpful to anyone because that's just plain wrong. I'm compiling a new cpu hierarchy for dolphin and I can tell you that it would greatly benefit from having more available data. Particularly from "exotic" (uncommon) processors.
kodiack Wrote:The benchmark was originally for finding out which processors would be capable of emulating the Wii's CPU at or above full speed.
That only applies to POV ray though. Any other application is going to have a different workload and therefore have different host machine cpu requirements to emulate the Wii cpu at fullspeed. For example the i5 4670K easily beats the wii cpu in this benchmark yet it still cannot run many games as fast as a real wii can.
kodiack Wrote:but beyond that there really isn't too much more to say.
Do you really think that haswells great performance (which is something we already knew about before this benchmark) was the only useful thing anyone could possibly conclude from a benchmark like this? We still know very little about what exactly makes a cpu good for dolphin. Or how the various components of a cpu affect dolphins performance. To reach these conclusions we need more data. I am particularly interested in studying the effects of cache subsystem properties on dolphins performance.
I guess I worded that poorly. There's more that can be discovered, although the chart doesn't need to be updated for every last CPU model out there. Showing both an i3-4330 and an i3-4340, for example, wouldn't be particularly useful since the 4340 is basically going to be x% faster than the 4330 in virtually all possible cases. Likewise, showing how much of an improvement a processor gets from simple overclocking isn't particularly exciting either since it's going to just be near-linear scaling.
I'm interested to see what sort of results you get from those "exotic" CPUs. Are they really high-end stuff, or just CPUs that may have had more of a niche role in the market? Is there anything that you think will be able to compete with Haswell, or are you more curious to see if those processors can simply get adequate performance?
Kodiack Wrote:the chart doesn't need to be updated for every last CPU model out there.
Agreed but that's not the same thing as not updating it at all. Which is what you were trying to rebut.
Kodiack Wrote:Showing both an i3-4330 and an i3-4340, for example, wouldn't be particularly useful since the 4340 is basically going to be x% faster than the 4330 in virtually all possible cases. Likewise, showing how much of an improvement a processor gets from simple overclocking isn't particularly exciting either since it's going to just be near-linear scaling.
Actually measuring scaling can be very useful. It helps identify bottlenecks in different components. And it's less linear than you might think.
Kodiack Wrote:I'm interested to see what sort of results you get from those "exotic" CPUs.
I'm not the one doing the testing. I don't have a warehouse full of different cpus at my disposal (unfortunately). I stated that I want to see more of these results in the benchmark.
Kodiack Wrote:Are they really high-end stuff, or just CPUs that may have had more of a niche role in the market?
Since most people buy the best processors they can for dolphin they tend to be lower end stuff. And of course they're niche. That's the whole point.
Kodiack Wrote:Is there anything that you think will be able to compete with Haswell,
Not yet no.
Kodiack Wrote:or are you more curious to see if those processors can simply get adequate performance?
I think you misunderstand why I want this data. Please reread this part of my post:
NaturalViolence Wrote:We still know very little about what exactly makes a cpu good for dolphin. Or how the various components of a cpu affect dolphins performance. To reach these conclusions we need more data. I am particularly interested in studying the effects of cache subsystem properties on dolphins performance.
Oh, yes. I promised to benchmark with different CPU-NB and HT speeds, didn't I? I'm sorry about that, got preoccupied with a session of random video games, and I wasn't actually free yesterday, even though I gave time to post on the forums.
I'll get right to it, and, again, I apologize about the two-day delay.
(04-03-2014, 05:39 AM)NaturalViolence Wrote: [ -> ]Kodiack Wrote:the chart doesn't need to be updated for every last CPU model out there.
Agreed but that's not the same thing as not updating it at all. Which is what you were trying to rebut.
I've told the same thing to several people already: if you want to make your own results spreadsheet (with blackjack and hookers), go ahead instead of complaining about the current one. But it's always easier to complain rather than doing anything.
Also, it's always appreciated when I take ~30min-1h to go through old results and integrate them into the spreadsheet only to get told by people that I "don't update it at all". Guess I won't bother then, have fun.
I'm closing that thread, actually. Thanks to all the people who provided useful results, and fuck to all the people who spend their life complaining about other people's work without doing anything. Go make your own thing.