20%? Considering that the raw power scaling of Espresso is 6 times Broadway, I'd say that puts Espresso a little bit ahead of the last gen CPU.
How did you reach to this number? If we take the statement "one Xenon core is 20% faster than Broadway at GP" as a base, it's 1243 / 729 / 1.2 = 1.4208, or IOW a single Expresso core should be 42% faster than a single Xenon core at GP.
Did you actually search? 750CL/Broadway and Espresso all feature 2-way SIMD known as paired singles.
It might not be too troubling. If it this is the case, it could just be a firmware away from being improved (not the game code, but the system level's network performance). Also consider that netcode can make a huge difference, it's possible the the netcode just wasn't meshing well with the networking on the WiiU (whereas it wasn't a problem on other platforms).
I was referring to the bolded number in your post I quoted. Didn't you post that in response to freezamite's last post?
Expresso surely has paired singles. Whether it has some SIMD beyond that is not known, but the chances are low.
What do you mean by raw power scaling?
No.
Most PPCs have had 64bit fpus since the 603 days. Actually, it's one of the reasons for Gekko to put those wide FPRs to good use.
Ok, I just saw his edit. My bad.
From here:blu said:
I'm aware of the source of the 20% in your post. I was asking where krizzx's 20% were coming from as his pre-edit post allowed misinterpretation (which I fell for, apparently) in the sense of '20% is what an Expresso core should do over a Xenon core at GP'.
If I had seen the question mark from the start I wouldn't have misinterpreted your post. The question mark would have been a sufficient indication you're referring to the same number the other poster had already given a context to.
Thanks, Donnie. I'm adding to it as I come across specimen. Funny fact: the Dothan actually got on the list by accident - I was asking around for an Atom netbook to test on, and a friend offered his. Turned out the netbook had a Dothan ; )
Yep, that should do - thanks. I'll build a core2-tuned version tonight and post the asm listing along with a gcc command line to produce the binary. You'll have the pleasure to do the bunch of runs ; ) Just make sure the test machine is not in some power-save mode where it could throttle the CPU.
gcc -o testvect_intrinsic -x assembler path/to/downloaded/asm/listing -lrt -lstdc++echo "1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 1" | ./testvect_intrinsicYou have been slightly misled. Extrapolating the results from the test shows an Espresso core should perform equally (on the test) to a 1.6GHz Bobcat, not to a 1.6GHz Jaguar. Jaguar should have twice as wide SIMD fp unit compared to Bobcat, and that should show in the test. Now, how much better Jaguar will be compared to Bobcat remains to be seen.
Thank you, lightchris. I'll add those to the list. And yes, one has to be very careful about the clock speeds - all new CPUs have some form or another of clock boost. One has to know what that boost is and when that happens.
First of all make sure you are dealing with the Nehalem code (A: autovectorized; B: manual emit) as some of the other listings are not even x86. Then use the 'download' link at the top of the page instead copy-pasting the code. If you still cannot build it, paste the error output here and we'll figure it out.
sudo gcc -o testvect_intrinsic -x assembler sse3_nehalem.txt -lrt -lstdo++Indeed. That Nehalem code is 64bit (x86_64). I can build it for 32bit (the Dothan code for instance is 32bit) but that will likely affect the results negatively (increase register pressure, particularly in the manual emit case). Ok, we need a 64bit OS for the C2D. Thank you for the effort nevertheless!
It should be perfectly fine.
By all means! Just be careful of boost clocks. Thanks!
That's because I made a typo on the command line I posted :/ It's not -lstdo++, it's -lstdc++