AMD Kaveri OC On Planet Neptune

[9enesis]

yet....some kind of benchmarks of i5-2500 vs some AMD eng sample( implied as steamroller)

http://browser.primatelabs.com/geekbench3/compare/226070?baseline=223722

 

[PPB]

I'm not really surprised that you look down on everyone. Regardless, I wasn't going to call fake because there are people around here that have sensitive feelings when it comes to AMD.

How considerate. The only one you are convincing with that excuse is yourself

 

[Homeles]

How considerate. The only one you are convincing with that excuse is yourself

Whatever makes you feel better about yourself.

 

[inf64]

I have found something for you guys, posted it on SA and here for now . Hope you like it. The scores are not out of this world like previous time, so this is looking like a legit screenshot. Magic date is January 14.
http://www.gdm.or.jp/voices/2013/1230/56219

Enjoy!

PS It takes 4.2Ghz PD to match 3.7Ghz SR in C15 . 7850K runs C15 at 3.7Ghz, it doesn't turbo up in this workload (at least not in the run that you see in the image above ).

 

[monstercameron]

darn I can't wait! 2w until launch 4w until availablilty?

 

[inf64]

Who knows, it should be like that but anything is possible.

 

[Ajay]

PS It takes 4.2Ghz PD to match 3.7Ghz SR in C15 . 7850K runs C15 at 3.7Ghz, it doesn't turbo up in this workload (at least not in the run that you see in the image above ).

Ugh, and still slower than an Intel mobile i5 @ 1.70 GHz
Maybe C15 performance is really dependent on L3$. Thanks for the post though.

 

[inf64]

Mobile i5 @ 1.7Ghz? You do know that mobile i5 doesn't run this benchmark at 1.7Ghz, right?
Mobile i5 also has 2x more SIMD resources per core and costs many times more.

 

[PPB]

- Gets a kaveri ES
- benches it on the superpi of this era.... Errr on CB 15

Dat logic. Here's for waiting for benches that at least represent real world workloads.

 

[SiliconWars]

Is that lower IPC than Jaguar? Dear me, hard to believe it's THAT bad tbh.

 

[inf64]

Is that lower IPC than Jaguar? Dear me, hard to believe it's THAT bad tbh.

Nope I don't think so. It's actually ~8% higher than PD in this particular test.

 

[SiliconWars]

I thought Kabini @ 1.5 GHz was getting ~0.4 in CB 11.5 single thread.

 

[inf64]

This is Cinebench R15. I don't know how R15 scores compare to R11.5, but I'm pretty sure they are not comparable. We only have R15 score so far. It would help if we knew Kabini's R15 score.

 

[SiliconWars]

Ahah, gotcha cheers.

 

[inf64]

I found something regarding Kabini here:
http://www.noticias3d.com/articulo.asp?idarticulo=1883&pag=8

1Ghz dual core part, scores 38pts in muti cpu portion of R15. So to get the same core and clock count one should multiply this with 3.7 and then with 2- bets case scenario for Kabini assuming perfect core and clock scaling (this will never happen in reality though, so real scores would likely be ~5% lower)
38x3.7x2=281.2
Turns out SR with its 2 FP units is ~11% faster than 4 Jaguar cores, both running at the same clock of course and assuming Kabini scales perfectly with core and clock count. Note that Jaguar cannot clock past 2.5Ghz let alone ~3.7Ghz range.

 

[PPB]

Ahah, gotcha cheers.

Anyways, to continue the line, sr has a trimmed FP compared to PD, achieving 8% increase on pure FP code with fewer resources is a good thing (trimming an already "thin" FPU is totally debatable if we left HSA out of the equation, tho).

I would kill for some vray benches, or anything that resembles a real case scenario.

 

[inf64]

Well C15 is not "pure" FP code but at least there is no regression in IPC.
Something tuned for bdver3 would probably show >30% improvement (my guess). Time for some linux benchmarks to show up.

 

[SPBHM]

PS It takes 4.2Ghz PD to match 3.7Ghz SR in C15 . 7850K runs C15 at 3.7Ghz, it doesn't turbo up in this workload (at least not in the run that you see in the image above ).

what about the MP ratio compared to Richland/Trinity? (need to disable turbo I guess to be fair)

as a quick reference, my i3 2100 scores 106 for ST but only 266 for MT (that's dual core SB at 3.1GHz)

Kaveri is not going to look good against haswell i3 when it comes to CPU performance, but I think that's no surprise, the IGP performance should be more interesting to look at.

 

[avtek21]

Cpuz of Kaveri memory clock is not set up right with less than half clock of A10-6800k. Recent reviewers reversed conclusions after A10 setup errors were fixed.

 

[mikk]

Kaveri is slower than Richland in Cinebench R15 if this is legit.

 

[inf64]

Kaveri is slower than Richland in Cinebench R15 if this is legit.

6800K @ stock scores ~322pts. That's when the chip runs at ~4.2Ghz. Kaveri runs at 3.7Ghz in this test and scores ~311pts.
So yes, 6800K will be a bit faster, some ~3.5%.

 

[Erenhardt]

I though ES runs at 3.5 GHz? Is is retail sample?

 

[Homeles]

Dat logic. Here's for waiting for benches that at least represent real world workloads.

The point of these benchmarks is to emulate real world workloads.

Well C15 is not "pure" FP code but at least there is no regression in IPC.
Something tuned for bdver3 would probably show >30% improvement (my guess). Time for some linux benchmarks to show up.

Isn't Cinebench integer-heavy?

 

[Ajay]

Mobile i5 @ 1.7Ghz? You do know that mobile i5 doesn't run this benchmark at 1.7Ghz, right?
Mobile i5 also has 2x more SIMD resources per core and costs many times more.

Meaning it would get spanked by a desktop i5, or even a desktop i3, by an even more substantial margin :whiste:

Color me disappointed so far, but we'll see what happens when we get a real review. I really hope it shows much better than this.

 

[ShintaiDK]

Isn't Cinebench integer-heavy?

Yep.