Wednesday, 11 April 2012

More Trinity Benchmarks

Trinity die shot

Here are some more Trinity Benchmarks. It looks like IPC increases by +10-20% for these tests when compared to the FX 8150, but it appears in these tests, IPC still does not catch up with the older K10.5 architecture.

In single core / threaded performance, the 2 module / 4 thread AMD A10-5800k @ 4.2 Ghz stock compared to a four core i5 2500k @ 3.7 Ghz is ~10.9% slower in FPUperformance test, and ~13.3% slower in the Integer performance test.

EDIT: Assuming the above ratio's are correct, and a scaling benefit of 10% (taken from cinebench 11.5) from the FX 8150, a 5.0 Ghz FX 8350 Will perform roughly 57% better in Int. Math, and 60% in FPU than an i5 2500k @ 5.0 Ghz when all threads can be used.

The mobile variant for trinity the quad-core AMD A8-4500m @ 2.2 Ghz (EDIT: 2.6 ghz TurboCore 3.0)when compared to the quad-core intel i7 2920xm @ 3.5 Ghz is only a mear ~20.0% slower in the FPU test, and ~31% slower in the integer math test. But remember the intel i7 has a huge (EDIT: 34.6%) clock advantage!!) Just look at the FP / Int per Ghz on the i7 2920xm compared to the A8-4500m.

If you think thats impressive, just wait for the mobile A10-4600m @ 3.2 Ghz TurboCore 3.0! ^^  I suspect if these clocks are correct, that the A10 will be a serious mobile ivy bridge competitor.

ADDED: Following this trend, we see a 2M/4T Trinity 17w ultrathin platform part @ 1.8 Ghz Turbo will perform significantly better than current i5 or i7 ULV processors.

Not only will AMD offer superior GPU horsepower with these new chips, but early CPU benchmarks impy Piledriver really is what Bulldozer should have been to really be the intel killer CPU people hyped it to be. The IPC difference between K10.5 and piledriver is ~5%, which will vastly outweighed by the higher clocks, especially for mobile parts.



  1. as I told you, piledriver should be a sucess, if the APUs are ~10% slower than the best quad cores from intel, then vishera should be much faster, and it should be quite as fast as mid-range IVB-E

    as for the mobile benchmarks, that only shows us that trinity will smoke anything intel has to offer both in CPU performance as well as in GPU performance at every price segment even next gen if intel's clock is 1.59x the AMD's clock then per clock, AMD should smoke intel, (and AMD should be able to reach intel's clocks), even though that is intel's extreme chip

    one thing I'd like to note, sernox on xbitlabs said that AMD vishera (late PD) has a much better IPC than AMD trinity (early PD) for example vishera should do 4 MOV (the most used Assembly instruction and accounts for more than 30% of the instruction usage) instructions per clock although trinity should do only 2

    the reason the IPC for llano may seem more is that trinity can be clocked higher, and since Performance = IPC * base clock speed * Multiplier, peformance should be quite better despite slightly lower IPC, another thing clocks don't really change performance at equal rates meaning the performance gain (in points) from 2.2->2.4 != 3.2->3.4

    still waiting for an AMD comeback, I wish those benchmarks are true

  2. another thing I'd like to note is that we're here talking about CPU performance, not GPU performance, that's where AMD easily smokes intel

    but I guess intel will cheat and bribe vendors not to sell AMD systems

    AMD will lose money, won't be able to release another competitive product and intel will have a monopoly, maybe rory could do something about it