Slashdot Mirror
Dual-Core Shoot Out - Intel vs. AMD
sebFlyte writes "The Intel vs AMD battle of the benchmarks continues. ZDNet is running its rather comprehensive-looking guide to a side-by-side test of Intel and AMD's dual-core desktop chips, the Athlon 64 X2 3200+ and the Pentium D 820. They look at pure performance, as well as the difference it makes to apps you might use on the desktop. In the end, AMD comes out as the winner. From the article: 'AMD currently offers the most attractive dual core option. The Athlon 64 X2 3800+ may cost $87 more than its Intel counterpart, the Pentium D 820, but the AMD chip is a much better performer. It also uses considerably less power.'"
The best price/performance deal is the $146 AMD 3000 chip.
It is an amazing little bugger that can git er done with ease but does not cost and arm + leg.
If you "get" pointers add me as a friend (116)!
Coral Cache Link of the entire article in Print View
Why is it so hot? Where am I going? What am I doing in this handbasket?
Isn't this like the fourth time we've seen a Xeon-vs-AMD benchmark on the front page? It's old news.
The problem with the Xeons is they're totally throttled. The Xeon was like a V-6 engine under a VW carburetor; the dual-core Xeon is like a big-block V8 under the same carburetor.
The AMDs have better access to RAM and better (independent) cross-CPU communication. The dual-core Xeons were clearly rushed to market to answer AMD's offering, before Intel could get their own memory-access ducks in a row.
zdnet is usually fairly good, but not this time.
A kernel compiled for a single CPU is faster than a kernel compiled for multipe CPU's, even when you only have one CPU. This is why OpenBSD has two kernels: 1) one cpu and 2) multpiple CPU's. The main developer of DragonBSD said that his preference is single CPU, performance wise (I'll leave that as a Google exercise).
Huh? No reason you can't use Intel's compiler for AMD64 it if you like the code it generates (AMD64 supports SSE/SSE2).
/ eng/compilers/clin/220007.htm
Intel themselves even point out that their compiler supports AMD.
http://www.intel.com/cd/software/products/asmo-na
Incidently gcc 4.0 does automatic loop vectorization using SSE/SSE2, so I wouldn't dismiss it too quickly either.
You mean it wouldn't perform well on real-world floating point applications then? Like SPEC fp 2000?
Spec fp 2000 results
Oh, I dunno, I think those AMD results look pretty good...
And 3800+ vs 820:
3 800/index.x?pg=13
http://techreport.com/reviews/2005q3/athlon64-x2-
"The system based on the X2 3800+ draws less power at idle and under load than anything here but the single-core A64 3800+. Under load, the Pentium D 840-based rig draws 292W at the wall socket, while the X2 3800+ system draws 166W. And the X2 3800+ outperforms the Pentium D 840 more often than not. The performance-per-watt picture on the X2 3800+ is impressive indeed."
Load: 166W vs 251W (85W difference)
Idle: 117W vs 160W (43W difference)
You're talking about the infamous THG video. No, Athlon 64 doesn't suffer from that. (downloading videos from THG may need registration) See also this.
The thing that bothers me about all these reviews is they fail to mention that the Intel processors need (more expensive) DDR2 memory versus DDR for AMD. If one is going to compare prices of the processors, the cost of the faster memory required by should be included in the price of their processor. Also note, that when AMD comes out with Socket M2 processors, which support DDR2, then they should benchmark even faster.
AMD hasn't consistenly offered better CPUs at a better price.
Back when they were lagging in the performance race, with the early XP line (Palomino) versus the P4 (Northwood), AMD was trying hard just to keep up. They priced their processors typically 20% below equivilant-performing Intel processors.
AMD also had a pathetic platform for the server space, which consisted of (at most) a 2-way Athlon MP system utilizing a single 266MHz bus. The only chipset available, the AMD 761MP, wasn't exactly a top performer.
Keep in mind: back in 2001, AMD had ZERO server presence. Now they own 10% of the server market, and they are a popular choice for supercomputing cluster projects. Most of that has come since the release of the Opteron. You can only grow so fast...10%, given the Opteron has only been out for 2.5 years, is quite impressive.
These days, AMD's average selling price has gone way up, and their sales have also been growing impressively, which is why the company has posted their first profitable quarters in years. And AMD is poised to grow dramatically in the next few years as Fab 35 ramps up, more than doubling their current production capacity.
Man is the animal that laughs.
And occasionally whores for Karma.
Why wait?
The term 'Hemi' is short for 'hemispherical', which describes the shape of the cylinder head's combustion chamber. From Wikipedia:
"Anyone that has ever gotten an idea based on any of my work and done something better with it-good for you."--J.Carmack
I don't think it's so much that Intel's chips are throttled. The X2 3800+ uses PowerNow, and so it too throttles back and drops voltage down when idle.
I believe the real story goes back to earlier architecture choices Intel and AMD made with the P4 and original Athlon chips.
AMD decided for SMP ops it would be better to design the link between CPU and chipset with a star topology, so each CPU gets its own comm lines.
Intel chose to stick to the simpler bus topology for SMP ops, which reduced motherboard complexity and simplified chip design.
This still applies into the dual-core arena, although everything is moved on die. Intel's cores talk to the system on a bus, meaning that under load bus contention creates a bottleneck as each core tries to move data around. On the AMD platform, each core can talk independently, and contention issues are minimized.
Worse still, some Intel chip designs examined on tomshardware show they tom's calls: double-core, where two separate cpu cores are mounted on the same chip carrier, rather than having two cores as one die. This means that cheaper CPUs can be made because the dies are smaller, but each die much be matched so that they'll run well together under a common heat spreader. And again, their sharing data lines.
I'm too lazy to break out reference URL's for your examination, so I may have some factual errors, but I belive the general idea here is accurate.
USNG: 14TPU4605