AMD Dual-Core Performance Revealed

Timmus writes "In two separate articles, FiringSquad takes a look at the performance of AMD's dual-core Opteron CPU. The first article examines the performance of dual-core in scientific computing applications (MATLAB and LS-DYNA) as well as digital photography, while the second story focuses on the performance of dual-core Opteron paired against Intel's dual-core Pentium Extreme Edition in video encoding, Cinebench, and a few other applications. The performance improvements are pretty impressive in multi-threaded applications that take advantage of the technology."

Comments

YESSSS

[sehryan]

I am running one right now, which is why I got first post!

Re:YESSSS

[antifoidulus]

Unfortunately, it looks like the GNAA has a beowulf cluster of these and beat you to it.....

full article mirrors

[winkydink]

here and here.

OK then.

[millennial]

So we have:
scientific computing applications (MATLAB and LS-DYNA)
digital photography
video encoding
Cinebench and
"a few other applications".
So what about the average user? Will the college kid who just needs to type their papers, the parents who want to do their taxes, the gamers who want to play high-end stuff, etc. get any sort of boost from this?

Re:OK then.

[antifoidulus]

Why would they need a "boost"? These are expensive and obviously aimed at high end users. You can already get sub $1k laptops that really do all the stuff you described, so why would they buy a dual core desktop system?
If you are using a dual core system to run word either a) you have WAY too much money, or b) the code bloat at Microsoft has REALLY gotten out of hand......

Re:OK then.

[Chris Burke]

Currently, 50 processes. The two highest (memory and VM wise) are Thunderbird which is using (60mb of main memory) and Firefox which is using 55mb of main memory.

Point conceeded. Some OSS software chews up the memory, and FireFoo are major culprits.

Though I'd hope to hell Visual Studio is way down the list. It's just an IDE! It has a GUI and a text editor. All the memory-chewing hard work is done in the compiler back end. With that comparison, my Emacs session is 6MB.

Nearly all popular linux distributions now come on more than one CD (even if you ignore the source code) and the default installations are WAY bigger than that of Windows XP.

Of course they are -- they include reams of free software! Nobody would complain about the large size of Windows installations if that installation came with practically every piece of software you would ever need! Even a 'default' install that doesn't install everything still has vast swaths of software from compilers to office suites to web browsers to web servers to image manipulation to whatever.

Who could possibly complain about getting more free stuff, even if it takes another CD or two or three to fit it? Consuming disk space for useful things is fine. Windows installs are considered bloated because the size increases but the perception is that you're not actually getting more stuff. Honestly -- what comes with the XP install these days?

Chance for someone to karma whore...

[Amiga Lover]

OK. Anyone have a quick simple explanation of why Dual Core over Dual CPU motherboard? are there inherent advantages to dual CPUs so close together?

Re:Chance for someone to karma whore...

[BlueCodeWarrior]

Imagine two dual core CPUs plugged into a dual CPU mobo.

Pop an erection yet?

Re:Chance for someone to karma whore...

[Anonymous Coward]

Less heat, less space, less energy requirements, eventually less money because there is only one chip.

Re:Chance for someone to karma whore...

[Moraelin]

Probably the biggest advantage is that it's cheaper. (Although if by much, that remains to be seen.)

Plus, AMD's promise was something like being able to double the number of CPUs without having to buy a new motherboard. Though how much saving that will be (I expect AMD to price these pretty high), and whether it will mean that you're stuck with much slower cores to keep the TDP limits, that remains to be seen.

There are other possibilities for improvement, such as using a shared cache and IMC instead of just throwing two cores together and going over HT like on a dual CPU system. But AMD hasn't yet done that.

Re:Chance for someone to karma whore...

[shawnce]

Yes.

One example... dual core (true dual core) CPU have the ability to exchange data between the cores at faster rates and more importantly with less latency then when having to exchange data between CPUs on a dual CPU system. This can improve SMP flow.

Another example... good dual core implementations will utilize some form of cache unification to allow better bulk sharing of data between cores while still allow high-levels of independent cache activity (the IBM's Power5 is a good example of this).

Re:Chance for someone to karma whore...

[hobuddy]

1) Cost.
Since there need only be half as many sockets, the motherboard can be smaller, less complicated, and therefore less expensive. This is especially true in the case of single-socket motherboards, which are usually 50-60% as expensive as their dual-socket brethren. AMD has sweetened the cost savings even further by arranging it so that most single-socket motherboards already in use with a single-core CPU can accomodate a dual-core CPU after just a BIOS flash.

2) More efficient interconnection between the cores.
This advantage currently applies to AMD's design but not Intel's. As explained here, "As you can see, AMD didn't simply glue a pair of K8 cores together on a single piece of silicon. They've actually done some integration work at a very basic level, so that the two CPU cores can act together more effectively. Each of the K8 cores has its own, independent L2 cache onboard, but the two cores share a common system request queue. They also share a dual-channel DDR memory controller and a set of HyperTransport links to the outside world."

After reading the TechReport article I linked to above, it looks to me like AMD is way ahead in the dual core market in all of the areas that count: better backward-compatibility, better cache coherency, and lower heat.

one step closer to the day

[Prince Vegeta SSJ4]

where I can encode mpeg2 DVD (maybe it will be HD-DVD by then), rip & copy a DVD, Download a huge torrent, and Play UT with a respectable framerate.

1 Dual Core vs Dual CPUs?

[uofitorn]

Wouldn't a better benchmark be to compare a dual core setup to a similarly configured dual processor workstation?

The simple future

[caryw]

Why stop at dual core?
Once a way to link multiple cores of a CPU is firmly implemented scaling the chip to 4, 8, or even 32768 cores should be relatively easy.
With chip dies getting smaller and smaller the only real reasons not to continue this multi-core scaling would be physical space and power usage.
Perhaps they could scale multiple cores vertically instead of just making the chip wider and longer.
And perhaps the cores could only be "turned on" when called for instead of using up juice all the time.

Interesting look at the future of chips.
Sony's Playstation 3 is using a "cell processor" or similar multi-core design that has already been covered here in the past.

Arstechnica article on the cell processor here.
--
NoVA Underground: Fairfax County, Loudoun County, Arlington, Price William chat and local forums

Re:The simple future

[Detritus]

It wouldn't work. Why do you think we have processors with two or three levels of cache? There is a serious speed/bandwidth mismatch between the processor and the main memory system. There are ways of increasing main memory bandwidth, but they are very expensive. There's no point in adding more processors if they are going to spend 95% of their time stalled, waiting for cache lines to be filled.

Re:The simple future

[mobiux]

I am waiting for intel's 32768 core processor.

iirc, the p4ee dual core puts out 225 watts at full power.
That would make the chip putting out roughly 3686400 watts, or 3.686 megawatts.

It's too cold on this planet anyway.

Re:The simple future

[jd]

IIRC, the latest generation of Sun UltraSPARCs has 6 CPU cores. An alternative approach is to have "virtual" cores - have a stack of registers and pools of computational elements. This does require some extra element of sophistication, to share out resources, but if you have two programs with very different CPU needs, both programs should run faster. Also, if you have fewer programs than there are virtual cores, but instruction parallelization can be performed, you still get a speedup.

The idea of turning off parts of the CPU would work, if you have a large enough cache. What you would need to do is prefetch all possible paths far enough ahead that you could turn on any deactivated part of the CPU before the instruction needed to be executed. You then have an independent "monitor" processor (an MPU?) which purely scans the cache and turns off all elements on the CPU that aren't needed within the lifetime of any of the contents of the cache.

Another poster noted the bandwidth issue between processor and main memory. That is certainly a problem, but one that may be fixable. One way is to sped up memory (and the bus). The other is to look at ways of reducing the amount that needs to be transferred, by putting some of the CPU in memory. (The technique is called "Processor-In-Memory", and has been around for about 10-15 years.)

As for gamers (from TFA)

[ahsile]

Notice the lack of an Athlon 64 FX version of AMDs dual-core strategy. For the time being, its recognized that games are exclusively written for single-threaded operation and as such run better on single-threaded processors at elevated frequencies. Thus, the FX series marches on at 2.6GHz for now. ... so for games, keep to your single core CPU.

Re:As for gamers (from TFA)

[millennial]

I suppose that makes sense. The question this raises, though, is whether there are any games designed to work better on hyperthreaded/multiprocessor systems.

Re:As for gamers (from TFA)

[phoenix.bam!]

Definitely not on the hyperthreaded system. Hyperthreading is only useful is you have 2 or more low demand threads. The benefit of hyperthreading disappears when 1 process needs 99% of the cpu, like many games. This can even be deterimental as your game will never be able to use the entire processor.

Re:As for gamers (from TFA)

[DigitumDei]

Depends I guess. I know I don't have the luxury of keeping my gaming machine seperate from all other applications I use, so my gaming machine is also my work machine and it tends to have a lot of stuff running at any given time. Now when playing shooter games I often notice a sudden drop in fps when some service or other decides it needs to do something. A dual core machine would be a lot less prone to this I guess.

Also, from the article. "And although the company says dual-core isn't for gamers quite yet, perhaps it is, only in a different usage model. Alan Dang and I were discussing processor benchmarking moving forward and he came up with the idea that we don't run compute-intensive tasks in the background today because we think they can't be done. However, if a dual-core processor enables a DVD encode while you're playing Half-Life 2: Deathmatch, there's a good chance that the way we think about demanding tasks may change. Even though games aren't currently threaded, the background processes a dual-core processor enables may very well catapult the technology into favor with game enthusiasts."

For the lazy...

[Anonymous Coward]

...and for those who don't want to flip through pages and pages of flash banner ads:

Scientific Computing

MATLAB: Though the script includes a moderate amount of matrix math, it doesn't seem like much of it is parallelized. Our recommendation from two years ago still stands - for most Matlab users, the fastest performance will come with a single Athlon64 line.

LS-DYNA: I will bench the CPUs using two classic tests, a 3-vehicle collision and a single front-collision. The 3-vehicle collision takes more than 24 hours to complete - we do not have these numbers ready for this round of articles.

Digital Imaging

Capture One: With Capture One only supporting two CPU threads, the dual-core Opteron's lower clockspeed is a disadvantage.

Bibble: It took only 4 minutes to complete with the 2x Dual Core Opteron 275. 4 minutes! That's 4.2MB/sec of processing time - a 2x Dual-Core Opteron 275 can process RAW images about as fast as it takes to copy them from to your computer using a standard-grade USB 2.0 CF card reader!

Noise Ninja: On the slower Opteron 246, the fastest results were had with 4 threads, but on the faster CPUs, 8 threads was better.

Video

After Effects: Since the decoding of WMV-HD does not seem to take advantage of both CPUs, the performance gain from the Dual-Core AMD Opterons is virtually absent.

Anandtech article on same subject

[bersl2]

http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2397

All I can say is...

[FlyByPC]

...LONG LIVE COMPETITION!

I wish both AMD and Intel well. All the better for us. Lower prices and better performance.

Wow!

[truesaer]

This is pretty disasterous for Intel. The game benchmarks show significant performance penalties for dual core chips, as expected. Intel launched its dual core specifically as an Extreme Edition for games.

On other benchmarks the AMD dual core gets 10-20% better performance! SiSoft Sandra is an exception, where there is a mixed bag between the two processors.

This pretty much verifies for me that Intel did a seriously rushed cludge to get this thing out the door. The only reason I can think of to target this to gamers is that no OEMs would want to buy them for server or desktop use, so you have to target people who like the latest technology even if it isn't that great.

AMD on the other hand seems to have a pretty good product here. I can't wait until the desktop versions come out.

Anyone realize that suddenly all P4's disappear?

[denominateur]

Now, it's struck me as very peculiar that the benchmarks where the dual dual core setup from AMD really shines leave out any comparison whatsoever to the Intel dual-core offering. This begs the question whether the person doing this review is a journalist or a marketing represenatative of AMD.

"We did not have time to evaluate the Intel platform with the Intel MKL, the P4 3.0GHz is an older reference measurement." is a very cheap excuse and indicates either lazyness or bribes on the side of AMD... I hate hardware review sites!

In other news ...

[Luscious868]

Dell still won't sell servers with them ....

Anandtech has some cost comparisons/benchmarks

[IronChefMorimoto]

Anandtech has an AMD dual core Opteron and Athlon64 X2 article that might compliment the original poster's story pretty well. It has a sh*tload of benchmarks:

http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2397

I really wish they wouldn't do gaming benchmarks with an Opteron in stories like these. Just because the Opteron used has similar specs to the dekstop processor that hasn't been released doesn't necessarily mean that the gaming benchmarks are all that useful. Just my 2 cents.

It'll be interesting to see how soon prices fall for these AMD processors (server and desktop) when they go mainstream. Read the cost comparisons for these badboys in the article.

Finally, I'm glad that Anand decided to demonstrate that the new AMDs will be backwards compatible with Socket 939 motherboards WITH BIOS revisions. Intel's dual core processors don't offer that luxury, from what I read in the article.

IronChefMorimoto

Amateurs

[Jah-Wren Ryel]

How can anyone take an article seriously when the very first sentence just screams, "AMATEUR!!" like this one does:

Intel may very well go down in history as the first processor manufacturer with a dual-core solution, if only by three days.

IBM Power4, Power5
HP PA-8800
Sun Sparc IV

All full-fledged dual-core processors shipping long before Intel -- HP's been shipping for over a year and IBM's already well in to their 2nd generation of dual core processors with Power5.

Sure, you can excuse the author with some hand-waving about x86 context only or whatever. But if they really knew what they were talking about, they would have said it that way - or at least a competent editor would have corrected it. If these guys can't even get the trivial stuff right, how can anyone trust them to get the real technical details right?

Re:Amateurs

[leoc]

Another thing that pisses me off is that he tests these 64 bit CPU's with 32 bit Windows, claiming that Linux is "hardly mainstream".

What a load of crap.

These dual core chips are PERFECT for high performance NON-GAMER Linux systems, and yet these guys disregard the most mature and stable 64 bit platform to run game benchmarks on 32 bit windows.