Haswell vs Kaveri

That is quite old info, though, it's possible that it was true for the configurations present at the time, but doesn't necessarily mean it's still true for all configurations?

 

They're making FM2+ backward compatible which probably means that it doesn't sport that many new pins and that Kaveri was shoehorned into a dual channel configuration for compatibility. This might explain the lowered performance numbers and the delay.

 

AMD A10-7850K APU
AD785KXBJABOX

174$
http://www.amazon.com/AMD-AD785KXBJABOX-A10-7850K-APU/dp/B00H7Z7YMI

 

Will a Core i3 from Broadwell have more graphics performance than Kaveri?

 

A full implementation utilizing Crystalwell would have a very good chance IMO. Even without Crystalwell it should be close.

 

Well the question was about "Core i3". As such I doubt you'd see anything higher than GT2 without Crystalwell. (Doesn't necessarily mean though it can't beat Kaveri, especially for the mobile chips where achievable performance is purely a function of power efficiency. Not much idea though how Kaveri nor Broadwell fair there but needless to say Broadwell has a HUGE process advantage.)

 

Cinebench R15 score from Kaveri: http://www.gdm.or.jp/voices/2013/1230/56219

 

The efficiency of the CB15 score is slightly above what an 8-core Piledriver performs at best, despite the lack of L3 cache and the shared RAM interface with the IGP core.

Sadly, it won't be soon enough to find how the new architecture re-spin manages, as a proper desktop SKU.

 

Core scaling between single and multi core score is way way better than on old Piledriver APU's. This is definitely a better single-tasker compared to older A10-5xxx and A10-6xxx CPU's. My A10-5800K gets around 2.9 to 3.05 MP scaling and that is with 200MHz higher turbo in the BIOS.

 

Single-thread score isn't good, it's slower than Richland. A10-6800k scores 100 points.

 

On another forum someone suggested that while FP performance has dropped by 20% from Piledriver, they got INT performance up by 30%, so it all depends on the load you throw at it, Cinebench is AFAIK FP-heavy

 

It's 88. I just got 85 from my old Core 2 Duo E8400 3.00GHz, with a couple of tiny background tasks. I sure hope Cinebench is not representative of Steamroller's overall performance.

 

C2-45nm has more INT resources per core and 3x as much L2 per core compared to AMD's modules.

The only reason why Core 2 is not competitive anymore to newer AMD and Intel parts in several workloads is the FSB bottleneck. The cores themselves are still better than anything AMD has to offer in that regard.

 

http://forums.overclockers.com.au/showthread.php?t=1103810


328/101

 

101 for 4.4GHz

the Kaveri leaked result was at 3.7GHz I think (no turbo?)

 

Someone said it was with Turbo.

 

The turbo max clock was 3.9GHz for this sample, I think.

 

Turbo clocks up to 4.0 Ghz and since this is a retail sample 4.0 Ghz is more likely.

 

Considering that my Deneb@3.9GHz (which is virtually ancient in CPU terms) just did 402/103 on a system that has't been rebooted in ages, those CPU scores aren't all too impressive.

Hope they've gotten the overall efficiency up over Richland, though, as I've been waiting for a SFF PC with a bit more graphics grunt than the Intel offerings (>=HD 6570 DDR3).

 

Huge dump from intel on haswell's GPU, including a 1000 page EU opcode dump:

https://01.org/linuxgraphics/documentation/2013-intel-core-processor-family