Help me navigate the mess of computer processor names.

February 28, 2006 11:16 AM

I'm somewhat of a professional computer guy but even I am totally mystified by the current state of Intel and AMD processors. Can someone hip to the latest explain why chips are suddenly only going 1.8Ghz again, which chips from Intel are powerful enough for HD playback, and what, if any, good a 64 bit processor does for today?

I was helping a friend go computer shopping the other day and I was blown away by the mess of chip names, speeds, and prices.

Pentium chips were easy -- you just picked the fastest Ghz and highest numbered chip (like a P4, 3.5Ghz). Now I see "core duos" and all sorts of weird number schemes that serve as names and Intel's entire chip line is a mystery. AMD's is pretty much the same.

I thought processor speeds were breaking 3Ghz well over a year ago? Why when I shop for a ~$1k PC, the chip is only a 1.8Ghz? Some of the AMD stuff is 64bit, but there's not 64bit WinXP right? Is there any benefit today?

Are the core duos really like having a virtual second processor of equal speed? If I see a core duo 2Ghz chip, is that supposed to act like 2x2Ghz chips to the OS?

Finally, what if I want to decode 1080i HD video content on a PC. What's the minimum Intel chip I'd need? Do any of the new intel-based Macs make the grade?

I ask that last one because my old 3200XP AMD chip can only do 480p, and chokes on 720p or 1080i. My 2Ghz G5 iMac can do 720p, but not 1080i. I was hoping my next mac or PC was capable of 1080i playback.

posted by mathowie to technology (23 comments total)

there's not 64bit WinXP right

Wrong.
posted by holgate at 11:34 AM on February 28, 2006

Can someone hip to the latest explain why chips are suddenly only going 1.8Ghz again

Because Intel totally blew it with the Pentium 4, building a chip that sucked but that could achieve blistering clock speed, which they new they could market (the result is that a 3.5 GHz Pentium is not generally "faster" than some other chips at 1.8 or 2 GHz).

The Intel Core Duo and Solos in the 1.6 to 2 GHz range are evolved Pentium IIIs, in the direction Intel should have taken a few years ago (the Core Duos are two processor chips on one chip, if you will. They are nice, and a good redirection for Intel).

AMD didn't really make this mistake with their chips. They are generally superior, especially in bang-for-buck comparisons.

To really know which chips are best, you have to know something about chip architecture. That's hard. It's why Intel sunk their money into a dog P4 that was "fast," because they know how hard it is to explain the entire architecture of a chip to a general audience.
posted by teece at 11:42 AM on February 28, 2006

The Pentium 4's architecture was designed for sky-high GHz numbers, for purely marketing-related reasons. Pentium 4s broke 3GHz in 2002, and they haven't gone up much since then. This is probably because with the current way chips are made, they can't get many more clock cycles out of them. So, what they're doing is trying to improve your instructions per cycle by doing things adding HyperThreading, and now, dual cores. Dual cores are mostly like having dual CPUs, in that you'll get higher performance in general, but usually never twice as good, especially within a single application.

AMD's Athlon 64 CPUs were designed from the beginning to execute more instructions per clock, which is why they're much faster than Pentium 4s while having a lower clock speed and generating much less heat.

The Core Duos you're seeing now are basically dual-core versions of Pentium-Ms, which are improved versions of the old Pentium III. This is because the old PIII is a much more efficient design (when the first P4s came out at 1.4GHz, they were slower than 1.0GHz PIIIs). Intel has finally realized they can't pump up the GHz like they have in the past, and their upcoming desktop CPU designs will have a PIII-Pentium M lineage.
posted by zsazsa at 11:46 AM on February 28, 2006

1. 64-bit windows *does* exist. It's a bit harder to find and only avaliable OEM (newegg sells it for around $150). But if you use 64-bit windows, you should be sure that you're prepared for some of the fundamental differences. For example, you *must* have 64-bit drivers for each component in your PC. It may be impossible to find 64-bit drivers for old or specialized hardware. You can load 32-bit programs (under a pseudo-emulation layer called WOW), but the 32-bit processes *cannot* load 64-bit DLLs and vice-versa. That means that if you, for example, download some future version of Photoshop that takes advantage of 64-bit code, then your 3rd party plugins won't work - they must be recompiled. And the 64-bit OS can be faster or slower for some tasks - it just depends on the task. At this point, 64-bit is used mainly for servers and computers with specialized tasks. Perhaps it will be more broadly supported by the time Vista is released.

2. Chips are going back to 1.8 ghz because
A) Intel made a lot of tradeoffs when pushing the P4 to extremely high clock rates (notably huge pipelines). and B) Both AMD and Intel are focusing more on reducing power consumption instead of pushing clock speeds, because they have somewhat ignored power consumption in the "race" to win benchmarks, and it's becoming more important in a market where laptops finally account for more than 50% of all PCs sold.

Core-DUOs are, for the most part, the same to the OS as dual-CPUs. The differences aren't really worth talking about for a consumer. But any communication and/or sharing that must occur between the processors will be faster in a dual-core because everything is on-die.

There were lots of choices to make in the P4 days too - some P4s had 512k of cache, some 256k, some were 0k (celerons), AMD had their megahertz-equivalent ratings, etc. The best thing to do is usually to go to anandtech or a similar site and read benchmarks for the types of applications you care about.

I'm not familiar with video decoding, so I can't answer that.
posted by helios at 11:52 AM on February 28, 2006

I'll just attack some of the low-hanging fruit in your question, as there are a few technical ones hiding in there for the video-decoders, though I have overheard that an FPGA extension card might be the solution for straight decoding.


I thought processor speeds were breaking 3Ghz well over a year ago? Why when I shop for a ~$1k PC, the chip is only a 1.8Ghz? Some of the AMD stuff is 64bit, but there's not 64bit WinXP right? Is there any benefit today?

Just a result of the processor speed not being a good measure of performance. A 64-bit processor should get better performance for operations its well-suited for, mathematical operations in 64 bits. I'm not a decoding whiz but unless there's some native 64-bit data that's getting mucked around with, I don't think you're going to see much benefit over 32-bit processors in the same price range.

Having a dual-core is roughly equivalent to slightly better than having two processors separately.

According to the Linux HTPC HowTo, the decoding alone requires a 2.8 gHZ plus a powerful graphics card to run stably. The required processor speed for the more efficient Core Duos will be significantly less.

On Preview: I concure with teece and zsazsa
posted by onalark at 11:54 AM on February 28, 2006

Here's what Apple says you need to play HD video.
posted by zsazsa at 12:02 PM on February 28, 2006

As far as 1080i decoding goes, your 3200 should be fine, provided you have a video card with good hardware acceleration. There are plenty of reports of 1080i on slower CPUs than yours with newer Nvidia cards.

On the Linux side, CPU info from Myth TV with XvMC (the interface for hardware-assist decode).

From Microsoft's info about Windows Media + HD:

Optimum Configuration
(to play 1080p video with 5.1 surround sound)
# Windows XP
# Windows Media Player 9 Series
# DirectX 9.0
# 3.0 GHz processor or equivalent
# 512 MB of RAM
# 128 MB video card
# DVD drive
# 1920 x 1440 screen resolution
# 24-bit 96 kHz multichannel sound card
# 5.1 surround sound speaker system
posted by SpookyFish at 12:09 PM on February 28, 2006

Are the core duos really like having a virtual second processor of equal speed?

Nom you have two physical chips. It's the same as having two separate chips, except they happen to be on one piece of silicon. There's nothing virtual about it.

If I see a core duo 2Ghz chip, is that supposed to act like 2x2Ghz chips to the OS?

There's no "acting" involved. There are two complete functional CPUs separately running code.
posted by cillit bang at 12:50 PM on February 28, 2006

As I understand it,

Intel thought that as customers believed The Mhz Myth* they may as well design the Pentium 4 to crank up Mhz at the expense of good design**. Well that's not entirely fair -- theoretically compilers could use those Mhz but it turns out that writing those compilers was so difficult that no one could. In every thousand CPU cycles a good third of them were firing blanks.

And then Intel's Itanium was shit too.

So they went back to the Pentium 3 and made a lower power version, Pentium M. Then they made a dual core version of a derivation of that.

They're not expected to catch up to AMD for a few years.


* that Mhz = speed, when it's merely influential.
** and processing speed... and power consumption
posted by holloway at 1:09 PM on February 28, 2006

btw, it's GHz and MHz, not Ghz and Mhz. /pedantic
posted by todbot at 1:31 PM on February 28, 2006

3.0 GHz processor or equivalent

Note the word equivalent. I would bet they mean 3GHz Pentium 4 or equivalent. There's the rub — which AMD chips and Intel Core Solo/Duo chips are equivalent to a 3GHz Pentium 4?

They won't need to have a 3 GHz clock speed to be equivalent, but I don't know the answer other than that. Sites like Tom's Hardware and AnandTech and whatnot can give you the answer, if you are willing to wade through the thicket of technical jargon.

(The same thing probably applies to the Linux HD requirement for 2.8 GHz: it is, again, meaning a 2.8 GHz Pentium 4 or equivalent).

A final note: 64 bit makes no difference with respect to this encoding problem. Floating point is 64 bit and has been 64 bit for awhile. It's integer floating point that is meant by 64 bit, which mostly won't apply to something like video encoding. Indeed, it might even be done with 32 bit floating point vector operations (ie, MMX, SSE, SSE2, SSE3, 3D Now!, Altivec, or whatever brand name is given to such).

Indeed, 64 bit is largely useless to apps unless they are scientific/mathematic integer crunching apps, or math/sci/server apps that need huge address spaces.

But, a whole bunch of the work has to be offloaded to your video card, so there's a need for real umph there, too.
posted by teece at 1:41 PM on February 28, 2006

Note the word equivalent. I would bet they mean 3GHz Pentium 4 or equivalent. There's the rub — which AMD chips and Intel Core Solo/Duo chips are equivalent to a 3GHz Pentium 4?

That's exactly what prompted the question. I see things like that for HD playback requirements, but how am I supposed to figure out if a random Intel or AMD 1.67 GHz chip counts?

Thanks for all the answers everyone, looks like it's still one big confusing mess.
posted by mathowie at 1:43 PM on February 28, 2006

Compare benchmarks between processors that test the applications you use. It's the only way to have any idea for sure.
posted by cellphone at 1:55 PM on February 28, 2006

Secretly, not so much a big confusing mess.
posted by Jairus at 2:00 PM on February 28, 2006

AMD's "PR" (Performance Rating) system was intended to equate to a Pentium's speed. Of course, with everything much above your 3200, it goes to hell. In your case it is pretty accurate, so despite running at only 2.2 Ghz it will be generally comparable to a 3.2 Ghz Pentium 4.
posted by SpookyFish at 2:12 PM on February 28, 2006

The Core Duos you're seeing now are basically dual-core versions of Pentium-Ms, which are improved versions of the old Pentium III.

Indeed. When I bring up the Processor pane on my new Intel iMac Core Duo, it actually says they're Pentium Ms.
posted by WCityMike at 2:50 PM on February 28, 2006

64 bit chips, coupled with Windows XP x64 or Windows 2003 64 bit can have a compelling I/0 advantage, in that these O/S are built on the new Storport storage architecture, as opposed to the old Scisport garbage. If you are going to be doing RAID at any serious level, this could be something to keep in mind. But be aware that "x86 64 bit" isn't "real" 64 bit throughout the chip; as others have pointed out upthread, only selected registers and operands work with the 64 bit extensions to the traditional 8086 instructions set. So, even though a great deal more memory space is theoretically addressable by these "64 bit" chips, in practicality, the memory controllers and supporting chipsets with which they will work are only going to show you some limited support for configurations above 4GB of physical memory. Whether even 4 GB configurations are going to be practically meaningful in the future for Windows applications depends on whether Microsoft does something about the 2 GB memory barrier, and that depends on Intel doing something about PAE for desktop chip sets, and AMD doing something similar in the builtin memory controllers for its dual core chips. I wouldn't hold my breath. But beyond even this pretty basic limitation, desktop chips like AMD and Intel dual cores are not NUMA aware, so that their ability to do anything truly useful for accelerating multi-threaded applications is going to be limited to things where the threads of any application are never data interdependent. Mainly, they're being marketed as boosts to productivity for people who multi-task several applications at once.

So, if you've got a database running on a machine, and some front end application, you can get decent performance for large or complicated queries, without killing your screen responsiveness. Or, if you're going to rip DVD's in the background while you surf the Web, you'll be happy (or at least happier than you would have been with a single core chip of equivalent clock and cache spec). If you're counting on getting the cores to cooperatively accelerate some big game application, well, to put it simply, they won't, and so if you're working with something that is best done as NUMA aware code, stick to Xeons.

Coming back to your question about 1080i HD playback requirements, according to this thread, your hardware requirements need not be astronomical, if you are not wasting cycles with poor codecs, and if your basic I/O and video hardware can operate together smoothly. High end video cards don't provide much help, because their graphics engines 3D power isn't called upon for HDTV, but PCI-E might be a benefit in reducing contention for memory access when playing back from disk, depending on the chipset for whatever motherboard you choose (thinking AMD Nvidia based 939 socket boards here).
posted by paulsc at 3:21 PM on February 28, 2006

It's integer floating point that is meant by 64 bit

Integer floating point??? Is that like an opaque transparency?
posted by grouse at 3:48 PM on February 28, 2006

Integer floating point??? Is that like an opaque transparency?

Yes, exactly like that.

I didn't notice the extraneous floating point there: It's integer floating point instructions that is are meant by 64 bit, generally.
posted by teece at 4:00 PM on February 28, 2006

That's exactly what prompted the question. I see things like that for HD playback requirements, but how am I supposed to figure out if a random Intel or AMD 1.67 GHz chip counts?

The first thing you need to do is discard your notion that clock speed is a reliable way to compare chips. The actual speed that those AMD chips are clocked is irrelevant, at least when comparing them to Intel products. This is exactly why AMD uses their PR rating instead of advertising the clock speed.
posted by Rhomboid at 5:29 PM on February 28, 2006

Yeah, go on the AMD speed rating. Add another half on AMD's PR rating if you want a general comparison.

shsc.info's PC Parts Picking Guide sez,

AMD's dual-core chips (Athlon 64 X2) are faster than Intel's (Pentium D) in pretty much everything, even A/V encoding. Intel does have a $250 Pentium D chip, however, whereas the A64 X2s start at $300 (this $250 chip isn't necessary a good buy, though, see section 3-III). That being said, A64 X2s generally have the performance advantage over Pentium Ds at the same price points.

Athlon 64 X2 chips work in most existing Socket 939 boards (except some VIA K8T890 ones) with just a BIOS update. On the Intel side, the Pentium D requires a new motherboard with an Intel 945 or better chipset, or an nForce 4 Intel Edition chipset. This means that if you already have an LGA775 Pentium 4 system, you'll likely need a new motherboard to be able to use a Pentium D. The same goes for AMD systems with some VIA K8T890 chipsets.

Regarding power consumption, the picture is the same as with single-core chips: the Pentium D 840, for instance, draws over 100W more at full load than its A64 X2 equivalent (the 4200+). You would therefore be ill advised to go with a Pentium D if you intend on building a quiet system.

Lastly, both the Pentium D and the Athlon 64 X2 have 64-bit support, although again 64-bit performance is slightly better with the AMD chips.

//

The Intel Core Duo is really two CPUs sitting beside each other on the same silicon, They still talk to each other via the FSB.

A more ideal situation however is to share some parts and allow more direct communication, hence AMDs direct connect.

(A separate issue is that Intel's chips do "Hyper Threading" and present spare CPU time as a second CPU. So a duo core appears as 4 CPUs despite only being 2... just remember they're a good idea but fake)

AMD still have the lead in cost, speed, and nerdishly-getting-to-feel-your-cpu-is-smarterness but not so much that you'd change your mind and get a PC when you wanted a Mac.

When getting a PC though, AMD is what you choose.
posted by holloway at 5:49 PM on February 28, 2006

When getting a PC though, AMD is what you choose

Absolutely. I don't think most non-geeks comprehend how noticeably behind the market "leader" Intel is compared to AMD.

I wish Apple would have gone with AMD, but it makes perfect sense that they did not (for reasons entirely unrelated to performance).
posted by teece at 8:22 PM on February 28, 2006

It's my understanding that when running in 64-bit mode you get a bunch of extra registers, which can be nice.
posted by rbs at 7:05 PM on November 14, 2006