Monday, 14 May 2012

More Trinity Benchmarks! IPC on-par with Phenom II, ~43% better single core performance than Llano! (K 10.5)


Leaked Benchmarks! More to Come!

A10-4600m vs A8-3500m


+38% higher single core CPU performance. 


+23% higher multi core CPU performance. 
Remember 4 physical cores gives a scaling benefit over 2 modules.


+29% higher multi core CPU performance


+ 43% higher multi core CPU/GPU performance



+41% higher CPU/GPU performance


+220% higher CPU/GPU performance




Looks like Trinity is ready to hit Intel where it hurts! IPC looks great! This is great news for Vishera (FX 8350). Remember old bulldozer arch has significantly worse IPC than Llano. 

NOTE: Compare the single threaded performance to the Phenom II X4 940 in cinebench 10 found here
We see at 3 Ghz (even before turbo), the Phenom II X4 940 (scores 2626) actually falls behind the performance of a single core from the A10 mobile part (2792) also at 3 Ghz (with turbo). IPC is looking ahead of K10.5! (by over +6% in this case)

The only benefit older arch has is better scaling. Per core performance, however, is looking ahead of phenom II. This loss in scaling is more than made up for with higher clocks. Comparing a 6C/6T Phenom II X6 @ 4.2 Ghz to a 3M/6T Piledriver CPU @ 4.9 Ghz will be an even match when all cores are used, but for a single core, the Piledriver CPU will edge ahead of the Phenom II by a significant amount (~20%). 
(Gaming benchmarks will be looking much better for Piledriver) 

With current bulldozer arch (in the case of the FX 8150), both scaling and single core performance are lacking when compared to older phenom II processors, with only a max clock advantage. At least AMD has made up for one of these pitfalls. We will see just how much scaling increases with the newer architecture when official benchmark reviews are out tomorrow! I am guessing only ~5% or so. Overclockability will also increase roughly 5% when compared to the FX 8150. 

Phenom II X4 4C/4T - 3.96 x Scaling , 1.00 single core performance @ 3Ghz, hits 4.3 Ghz
FX Bulldozer 2M/4T - 3.28 x Scaling, 0.75 single core performance @ 3 Ghz, hits 4.9 Ghz
FX Piledriver 2M/4T  - 3.44 x Scaling, 0.97 single core performance @ 3 Ghz, hits 5.1 Ghz
Intel i5 3570k- 4C/4T - 3.56 x Scaling, 1.25 single core performance @ 3 Ghz, hits 4.9 Ghz

This of course this is a rough estimate, but you get the idea! 

For the 4M/8T FX 8350, I estimate @ 5.1 Ghz it will match Ivy Bridge i5 3570k single core performance @ 4.3 Ghz. But will scale roughly 85% higher when all threads are used. 

My cost estimate for the FX 8350 will be ~300 USD. 

EDIT: Given officially released clock-speeds (3.2 Ghz not 3.0 Ghz) IPC is looking on par or slightly below K10.5. 



3 comments:

  1. So if i read that right PD would be 6% IPC increase on Phenom II, per clock per core, which would be about 15-20% increase over BD

    This is for Mobile Chips, how do we know that will translate to the desktop chips?

    And you mentioned "official benchmarks" what do they entail and what chips are used?

    Or do i have to wait and see? :)

    ReplyDelete
  2. Why did you pick a Phenom II X4 940?
    Please,...955 to 980 with DDR3-1600...jeez. Those CPUs hit more like 3.7 GHz anyway instead of 4.3, the ones doing 4.3 were the AM3+ C3 stepping ones. Likewise, there are Phenom II X4s that can actually run 100% load for extended periods on air cooling at 4.3-4.4, when Bulldozer will be at about 4.8 for cherry chips.

    6% IPC increase over 940 is like saying "just behind Phenom II X4" in IPC anyway.

    ReplyDelete
  3. Unfortunately the number don't seem to add up anyway.

    Firstly the A10-4600 runs turbo @ 3.2Ghz, not 3Ghz.

    Add to that anandbench has the x4 940 single threaded performance at 3387 that's more than the bench in the link here @ 2626.

    add it all up and Trinity is about 10% down on Phenom II

    ReplyDelete