K10: meet (Beverly) Krusher to the 10th degree
By Theo Valich: Tuesday 28 August 2007, 18:09
LAST WEEK in Leipzig my kit was nicked, but before that happened we asked AMD if it would let us run memory benchmark scores on a system there. The reps gave us the company line and declined, but can still disclose the benchmark scores of K10 here and now.
If you were wondering why AMD was hiding the scores of K10 so secretly, there were two reasons. The first might be that the CPU sucks badly and after AMD comes out, Intel's lads can start celebrating the death of AMD. On the other hand, there the was clear and present danger of the K10 significantly beating not just the current Conroe/Kentsfield generation, but easily out besting Wolfdale/Yorkfield. This statement warrants at least three hatemails from Intel's R&D lads, but all that we will disclose here are results we have in our possession. The pics are gone with my stolen laptop, though.
The particular piece of equipment that was briefly in our possession allowed us to run 3DMark06, the most important to overclockers worldwide, and Everest 4.0, our favourite memory benchmark.
The particular processor was none other than a single socket Barcelona or Agena FX, call it what you will. The reference motherboard containing RD790 chipset packed two HD 2900XT cards, and the memory installed was Corsair’s Dominator PC2-9136C5D, or the same ones we have been using ever since they came out. There was a Raptor hard drive, and that was about that. OCZ will like the fact that PP&C Quaddie CrossFire PSU was installed in the system.
Windows Me II (Vista) Ultimate was installed as the operating system of choice. For some odd reason, the 32-bit version was installed in a system with 4GB of memory, needless to say the system detected 3.24GB, and benchmarketing commenced.
When running at 2.5GHz, the 3DMark06 score ended at 23.768, so we were thrilled to see such a good score coming from two 512MB cards. This showed the clear potential of this four core processor marchitecture, but the helter skelter ride happened after we overclocked the processor to 3.0GHz.
When clocked at 3.0 GHz and equipped with two overclocked HD2900XT cards in CrossFire, Agena FX or single-core Barcelona smashed an index of 30,000 3DMarks 06. Yes, you’ve read it right - the barrier of 30,000 was passed to, barely, 30.031. We know that there are different systems floating around, and we have no doubt that leaks will start appearing after this article goes live. Knowing what was the overclock of Kentsfield processor, Intel Yorkfield has quite a big job to do. One thing is clear, though. Intel needs Nehalem, and it needs it badly.
To be clear with you, dear readers, two HD2900XT cards were overclocked to 830MHz for the core and 900MHz for the memory, but ominous Catalyst 7.7 drivers were used.
Memory wise, it is particularly interesting to see benchmark scores of DDR2-800 on AMD’s Athlon 64 X2 or now this Phenom FX something. When running in dual channel mode, a theoretical bandwidth of 12.8 GB/s can be reached, but that is a pipe dream - of course.
In the real world, our Intel Kentie 2.93GHz test bed will score around 7.5 GB/s, Athlon 64 X2 5000+ shoots around 8.3-8.8 GB/s, depending on the memory latency, and now Phenom scores even better – pretty close to a five figure range. But regardless of missing the 10 GB/s mark, it is still faster than any DDR3 memory on an Intel system, regardless of the clock achieved by the DDR3 memory. If you put the memory on 1066 MHz, 11GB/s bandwidth was smashed with read, write and copy tests and that was by quite some margin.
We expect the results to go officially live prior to Barcelona launch in September. µ
Mas enquanto não aparecer pics e validações não acredito...
