Slashdot Mirror
Current Processors Tested With Linux
Happy-Jollies writes "The team at LinuxHardware.org have kept us up-to-date on the latest in processors for quite some time now and they're at it again. With the latest release of the Pentium 4 'Prescott,' many Linux users will be deciding where to spend their money. LinuxHardware.org's round-up takes a look at the Prescott, Northwood, Extreme Edition, and the AMD Athlon 64."
Actually, there's some fairly hefty core changes involved in making the integer pipeline 31 stages instead of 20. Almost enough to account for most, if not all, of the additional transistors in this thing.
I'd actually be suprised if the AMD64 extensions were in this chip...most of what I've been hearing/seeing has Tejas being the earliest we could see them.
I will always use AMD for my home PC mainly because of the price. IMHO Intel is over priced, you're paying for the name not the product/quality.
Don't forget the extra pipe stages... those count for a lot. I can't say how many transistors, but it should make a dent in the numbers.
A 1 MB L2 cache will be at least 50 million transistors, assuming 6T/bit (1024 * 1024 * 8 * 6), not including sense amps, decoders, tags, coherence, predecode bits, etc.
Been 'Dotted already??? Jeez, that was fast.
Don't let the fact that they proclaim the Athlon 64 to be the 'clear looser' in the SPECViewperf discipline because of the poor results in the DX-08 and DRV-09 tests fool you.
Those scores are probably not representative of the true performance of the AMD processor but rather of the early stages of optimization of the Linux support for the Nvidia NForce 3-150 chipset. The very same weakness was observed in the past in other reviews that used Specviewperf on Windows platforms, such as this one from THG. Subsequent versions of the Nvidia drivers have since brought noticeable improvements.
The AMD scores would likely have been much more competitive if a motherboard based on another Athlon 64 chipset like the VIA K8T800 would've been used for this review.
Unfortunatly AMD not being 100% compatable with the standard causes problems.
Good old FUD -- when Windows crashes on an Intel box, it's because of Microsoft. When Windows crashes on a AMD box, it's because the AMD "isn't 100% compatible". Uh huh. I'll buy either AMD or Intel without a hesitation (I'm not a fanboy) where one is the clear winner, but I still see FUD when someone spouts it.
Here's a clue for you -- neither AMD or Intel are x86 processors, but rather both have a converter that converts x86 to their own internal microcode. The x86 standard that they're converting is EXTREMELY well known and standardized (I mean -- other people have to write the software that runs on it). AMD is 100% compatible with core x86.
Intel also has much more R&D into making a chip that won't fry itself if it gets too hot. AMD? Naw, they'd rather make you buy a new chip than bother to make it prevent a problem before it occurs.
Tom, is that you? Firstly, Intel and AMD chips are comparable in heat levels (actually the Prescott sets all new levels for heat generation, making Intel the clear winner in the egg cooker category), but secondly you are correct -- Intel has been more fervent in putting heat protection on their chips, though it is arguably over-engineering. It's like having guidewires "just in case" around all of your high rises. It's generally unnecessary as, apart from Tom's Hardware PR stunts, heatsinks don't fall off processors in regular use. My car doesn't have a special fuel purge if I decide to fill it full of ketchup.
"Prescott, Northwood, Extreme Edition, and the AMD Athlon 64."
In all honesty--unless you absolutely need 8GB of memory--there's little difference between these processors in terms of performance. They're all more or less in the same ballpark. Sometimes faster, sometimes slower, depending on the benchmark. None of them is a huge breakout CPU performance-wise.
Here's what's different:
PRICE: There's a lot more than a few percent variation in price.
WATTS: In exchange for your 5-15% speed boost, note that you're getting more than a 15% increase in power usage.
That doesn't make sense. Yellow Dog is an Apple Authorized Reseller who will ship you a brand-spankin new G5 or XServe with Yellow Dog Linux running on it and under warranty from Apple.
They really should have used a Yellow Dog box or named it an x86 shootout.
Arrogance is Confidence which lacks integrity. -- me
Simple reason: MS Windows does not run on a PPC.
The owls are not what they seem
Do you know what a lathe? Maybe you're thinking of a planer, because I can think of no reason why spinning a processor at several thousand rpm's would help answer the question....
A lathe spins an object about a point very fast allowing you to work it into a cylindrical shape.
A planer shaves a layer off the top (or bottom) of an object allowing you to quickly work it down to a desired thickness...
They who would give up an essential liberty for temporary security, deserve neither liberty nor security
Although the Athlon64/Opteron and AthlonXP processors either match or surpass the Pentium 4 line, the one application that the Pentium 4 processors still excel at is in multimedia editing and encoding. Most people don't look at Hyperthreading favorably, yet with Hyperthreading enabled you get an over 30% performance boost in DV25 to MPEG-2 transcoding for video with everything else equal. That's not to say that the Athlon64/Opterons are not useful, but this is one area where they still excel. Considering that transcoding is the major bottleneck for the folks who edit video and create DVDs, there is a good reason to go for a Pentium 4. 30% of your time on a 2 hour video is a lot of time lost if you're trying to crank out videos. Not to mention that most of Matrox's editing cards are almost always approved for Intel hardware to be stable but not for all AMD hardware, particularly motherboards.
If, however, you're a gamer and aren't doing much media encoding or do it casually, then I would recommend an Athlon64.
One of the reasons to do use the new amd 64 bit chips is that since you'll have to recompile your software for it anyways to use the 64 bit mode efficiently, AMD has been kind enough to add some extra registers to the CPU. As the x86 has traditionally been a bit starved registerwise, this is a very nice addition and by itself can account for a 20 - 30% increase in speed. Compilers have more registers available to them which results in less 'slow' accesses of reading and writing to memory.
This does not enumerate the transistors, but you have a nice picture. http://www.hothardware.com/reviews/images/Prescott 32_34ee/prescott_die_8in.jpg
The additional cache and pipelines add many transistors.
Where is the AMD's answer to Intel's compiler?
I can't find the link, but I remember hearing once that code optimized for the Pentium actually had a larger speed improvement on the Athlon than it did on the Pentium. I don't recall which models were under test or what their clock speeds were, but it was a few years ago now. It wasn't publicized much, of course, because it bloodied Intel's nose..
The point is that AMD CPUs perform optimized and unoptimized tasks more efficiently (wrt clock speed) than Intel CPUs. While Intel has focused on exceedingly high clockspeed (witness the 31 stage pipeline..) and compiler improvements (IA-64/VLIW anyone?), AMD has gone for parallelism and efficiency in utilization.
AMD pioneered vector math with 3DNow (Correct me if I'm mistaken, but I'm pretty sure 3DNow predates MMX...) but, instead of Intel using AMD's technology, they developed MMX in response. I believe code optimized for either extension saw about the same improvement. However, as market leader, Intel led the market where they wanted. And since more developers coded for MMX than 3DNow, AMD added MMX. In doing so, AMD's chips started benefitting just as much or more from Intel's extensions.
Historically, its been Intel introducing the extensions, and AMD quietly implementing them in response. The x86-64 extensions are shaping up to be the first example of an AMD-pioneered extension that Intel will implement!
As for Intel vs. GCC compilers, GCC is a very flexible cross compiler that targets many different architectures. Intel's compilers are very architecture specific (hint: they target only Intel's architectures). Apples and oranges. Like Java and the CLR; their design goals are worlds apart.
As a side note, MMX.
AMD licensed MMX and then added on top of it...
This article provides background.
---- Move SIG...For great justice!
be warned - to get the boosts yr going to have to start at a low stage install and this results in a looong install time. on my imac it took (wait for it) a week to install. and updates take forever too. but, as long as you read up on your use flags and set them well you'll get a fast core system.
2 1337 4 u!
The "not 100% compatible" complaint is about ten years old.
My experience, buying a 200 MHz K6 and an 800 MHz Duron, has been good with AMD chips.
The price/performance ratio has always been more attractive for AMD relative to Intel.
I will say that once I had a problem with a K6-III that would lock up after an hour or so - turned out the cheapo CPU cooling fan wasn't moving.
AMD produces fine chips, but their reputation as a "budget" CPU means they get tied into other components of mediocre quality in order to keep the price in the basement. That helps sales at the low end, but actually hurts sales at the high end. MyCorp, like many, pretty much buys Intel only hardware, paying more than they need to for x86 instructions and for systems integration, just to keep the potential support hassles down to a minimum.
If AMD made sure that a separate model line, such as the Opteron, were only used in MoBos and systems with high MTBF on all the parts they might have a better chance of getting more corporate business. But it will still take time to overcome the image of "CPU Backwards R Us".
"Provided by the management for your protection."
First of all, no latches... all these chips are clocked very tightly and timing propogation will get ruined by latches. They'll be flip flops or buffers.
Secondly, its likely that they've subdivided their stages so small it will require extra logic to accomodate more stages. Keep in mind the typical computer example uses FIVE pipleline stages (from MIPS processors), and we're at THIRTY ONE.
Most of these transistors I would bet are for upcoming SSE3 instructions that are not on and additional pipelining logic for branch prediction, and to allow more parallelism in between the stages, especially the fetching and decode stages. Will all the new trace and victim caches added, it will need more logic to check all the different caches.
You mean branch prediction? They did some serious improvements to Prescott's branch prediction in order to compensate for the performance losses that come from having such a long pipeline (or, rather, the performance losses that come from not keeping that pipeline full.).
tasks(723) drafts(105) languages(484) examples(29106)