AMD's Athlon 64 and Athlon 64 FX Vs. Intel's Pentium 4 Extreme Edition. We showcase the latest CPUs from AMD and Intel and take a look at what they will deliver.
SiSoft Sandra is used for this test and it tests memory bandwidth synthetically. This means that it utilizes the bus to see just how much bandwidth it can use at one time.
If this graph doesn't firmly preach the values of an on-die memory controller, I don't know what would. The AthlonFX and Pentium 4 are both utilizing a Dual Channel DDR400 memory bus. You must keep in mind here that the AthlonFX is at a big disadvantage due to very relaxed memory timings it's dealing with. At the time of testing, getting CAS2 registered DDR400 was an impossibility for us. Corsair and Kingston both have announced registered DDR400 DIMMs today that will allow for more aggressive tweaking.
In the case of the Athlon64 3200+, we did not see as pronounced a difference. In the case of VIA Vs. NVIDIA on the chipset front, you'll see that the chipset is no longer a defining factor in memory bandwidth, as the memory controller is now on the CPU die. VIA, NVIDIA, and their chipsets are no longer the go-between when it comes to memory bandwidth. Quite frankly, I'm very glad to see this. AMD should be the one controlling it, in our opinion. Will Intel be next to do this? It would make sense if we do see it come to pass.
While this 2001 test has traditionally been a benchmark that really showed off the power of the AthlonXP, the Pentium 4 has been gaining ground recently and bringing more of a contest to this data set. Here we finally see what the Athlon64 series of CPUs bring to the table by watching them lead the Pentium 4 by over 20%. A commanding victory. You might also be surprised by the small impact that the DDR333 had when compared to the DDR400 scores on the Athlon64s. Certainly in this benchmark, it was all about the CPU clock and the instructions carried out per clock.
The extra cache in our P4 EE also made a difference in scoring in this business application, although that isn't the market that Intel is seeking with this new CPU. From this comparison it would be safe to assume that the extra on-die cache positively impacted the Athlon64 scores as well, but IPC would seem to be the major contributing factor to the stellar scores.
Notice the scoring delta between the K8T800 and nForce3 scores with the 3200+. We will discuss this below.
In this more up to date test that uses newer software, we don't see as big a difference in the scores though the scaling is almost identical. The criteria that we discussed above can certainly be applied here as well. One data set that does stand out is the AthlonFX compared to itself. The anomaly of the DDR333 score being higher than the DDR400 score repeated itself over and over again in our testing. So it was not a "fluke", but I really don't have any insight as to why.
Again, notice the K8T800/nForce3 difference in score.
In our older version of this content creation benchmark, we have traditionally seen the Pentium 4 take a commanding lead. Both the Athlon64s are showing a great amount of dominance over the Pentium 4 and the P4 EE. This time we do see the DDR333 and DDR400 scores pull a bit further apart, but not as much as one might think. The differences in the scores are still within the margin of error.
Once again we see the nForce3 board take a healthy lead over the K8T800 board.
Again, we see the same scaling as we did in the previous content creation benchmark. I think we have finally reached the time to let our older benchmarks go, as they are getting a bit redundant.
Above we saw the nForce3 150 board from Shuttle easily outstrip the VIA K8T800 board from MSI in every single benchmark. At first we were a bit unsure, but the picture becomes quite clear once you see the full data sets from the entire review and have had some time to spend with all of the hardware. It's our opinion that we are seeing the score differences because of the superior IDE implementation by NVIDIA. NVIDIA has had issues with their IDE drivers for well over a year now with their nForce2 line, and some of those problems have followed over to the nForce3. With that said, NVIDIA has devoted huge resources to fixing these issues and we think that the IDE driver of NVIDIA has simply been tweaked to perform much better than the VIA IDE implementation.
There is always another side of the story though. Currently we can crash every nForce3 board we have possession of, both retail and engineering samples, when running extreme IDE traffic. We are also hearing that this is why some Taiwanese board companies are going with the VIA solution instead of the NVIDIA nForce3. We've received new drivers here recently, but we haven't had time to test them out to see if the stability has gotten any better. We will surely be looking at this soon, as we are testing retail nForce3 150 and Pro boards.