From my understanding, the AMD=64bit. Supporting 32bits is a 'perc' so you can still run 32bit apps. Theoretically you cannot compare these two. We don't have a usable 64bit OS yet, and we also don't have 64bit benchmarking apps? Hell, We don't even have problem free motherboards yet, plus it,s BRAND NEW technology. I really don't see the point of this review. This benchmark actually means nothing , except that 32bit performance on this 64bit chip is not as good as Intel.
I think most people who buy AMD know that AMD does not always beat in sheer performance, but an OC'd AMD unit is cheap and matches (and sometimes beats) intel's products which are priced much higher. I think at this stage AMD is ahead, since they actually do have 64bit currently available. Intels CEO still mentions 64bit is not required for desktops right now, what crap is that. In my opinion, Desktops (esp gamers and extreme users) will prolly end up more powerfull than servers.
I agree on the 32bit front, a stock Intel will have a performance lead over a stock AMD - but OC the AMD and you're bound to find for the price you pay for an Intel, you could have save it and gone for AMD.
I agree, for Serious Server (multithreading) apps, Intel is the way to go, I actually recommended Intels for our Work Servers, but for gaming, AMD rox the sox off Intel.
In conclusion, I would honestly like to see a 64bit benchmarking app come out that can test 64bit capabilities soon. I'm sure intel's new offering will be an interesting one - we have to wait for this before we actually compare the AMD to anything at the moment.
The changes from 32-bit to 64-bit are not that huge. The only really important change is that it now can address more than 4 GB, in a single thread, without switching segments. See, now matter how much you praise 64-bit, a 32-bit processor can address just as much memory, only with a few detours.
The second change AMD made was obviously to extend the general purpose registers from 32-bit to 64-bit. In a way, you might think that this doubles the data being processed, but it doesn't. Really most of the time, a 32-bit word is enough to store any integer value you want. For example a loop counter really doesn't have to go beyond four billion. In cases where a bigger range is needed, we already have MMX and SSE which have 64-bit and 128-bit registers for several years now. I would even say that longer general-purpose registers make the processor slower because you have to wait longer for all those extra transistors.
The third thing they change is more revolutionary. Instead of keeping 8 general-purpose, 8 MMX and 8 SSE registers, they extended them all with 8 more registers. The good thing about this is that the double amount of data can stay in very fast registers so less memory accesses have to be made. Eight registers really always have been quite low and has been x86's flaw from the start. However, we do pay a price for that. AMD uses a prefix byte for all instructions that use this extended register set. Since many 32-bit instructions are only two or three instructions long, this is a significant increase in program length. Implicitely it requires a higher memory bandwidth to keep the pipelines filled.
Last but not least, AMD included the memory controller on the die. Although this has a good impact on performance, it has nothing directly to do with 32-bit to 64-bit transition so we can't make a fair comparison here.
But all in all, I do think you can fairly compare AMD with Intel. The changes that have been made are really not that huge. It is still x86, only just a variant. If there's going to be a performance increase, it will be from the extended register set and integrated memory controller, not the actual 64-bit addressing or longer registers. So why wouldn't the comparison be fair?
You will see very misleading information the next few months! AMD will show charts of the performance of programs optimized for their chip, showing 200-300% performance increase compared to Intel. But then a little later you'll see Intel showing MMX and SSE optimized applications that run equally well or better on their own chips.
I just wanted to note the importance of hand-optimized applications. I myself work a lot with assembly and have seen increases of over 300% in performance compared to what my compiler produces (it can't use MMX nor SSE). So a fair comparison would be both AMD and Intel optimize the same program for their own chips and compare those.
One big advantage of AMD though is that compilers can easily use the eight extra registers and 64-bit variables, while MMX and SSE hardly can be used by the compiler. Even vectorizing compilers do a poor job at using SIMD instructions and need many hints from the programmer to work more optimally.
So, if AMD compares 32-bit with their 64-bit programs compiled by the same respectable compiler (e.g. Microsoft Visual Studio .NET 2003 Professional or the GCC compiler for Linux), and not optimized manually, then we do have a fair comparison. And I'm fairly sure that in this case AMD will win most benchmarks!
Yeah, this is what I was talking about - If you have applications that are optimised esp for AMD, or intel - they will exceed with flying colours. I think if you were to include tests in the benchmark application to actually test the new features, I'm almost certain the AMD will be number 1 for sure, since the Intel does not have any of this currently. When the Prescott arrives, it will most certainly be the faster processor. Just because THG used benchmarks that are not taking the new improvements into account, don't really mean they are accurate.
These small changes drastically effect performace if you actually use them, sorta like what difference a 2MB cache does instead of a 1MB cache. Bit of a silly comparison, but you get the idea!