• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Intel finally catches up to AMD

a_unique_person

Director of Hatcheries and Conditioning
a_unique_person
Joined
Jul 19, 2002
Messages
49,606
Location
Waiting for the pod bay door to open.
Been reading that the new Core2 Duo is the must have chip. But it appears to me it is really just going down the development path that AMD started, and using it's research power to enhance. That is, wider, not deeper execution paths, more in a clock cycle, not just faster clock speeds. Intel might be doing it better (now) but AMD had already started out that way.
 
a_unique_person said:
Intel might be doing it better (now) but AMD had already started out that way.
Click to expand...
Um... which company really started it?

Does Intel make AMD-clone chips, or does AMD make Intel-clone chips?

It's important to remember which company has been wearing the coattails since the late 60's and which company has merely been riding them.
 
They went their different ways years ago, in terms of engineering and design. IIRC, the Pentium was the time they split. Although they use a common instruction set, how it is executed is entirely up to them.
 
This is worse than the Mac/PC arguments. I remember well when cpus were Intel, but there were some others call AMD and Cyrix (and a couple of others that fell early). They were the ones you went for if you were purely after cheap and nothing else. I seem to recall that Cyrix went for a Burton because of the lack of a maths coprocessor about the same time that AMD became respectable.

My first AMD was an Athlon 700. At the time, it was just cheaper than the Intel that gave equivalent power. I've used AMD ever since, but I have no antipathy for Intel. They just haven't given similar value when I've been buying.

Cheers,
Rat.
 
The admiration for fast clock speeds is deeply rooted in me. If Intel thinks I'm going to change my P3.4 for "wider paths" and "better predictions", then they're wrong. The same for AMDs that are named after their Intel equivalents and not after their clock speed.
 
Well, I upgraded from a 3.2ghz P4 Hyper-threading cpu to an e6400 core 2 duo, and I must say, the difference is night and day. Since I record TV with my computer, I an constantly transcoding video to keep my hard drives from filling up. The old CPU could do a two-pass encode of a 40 minute episode in about 2 hours. The e6400 will do it in 38 minutes at its rated 2.13ghz, and when I overclock it to 2.8ghz, 30 minutes. The difference in video transcoding time alone is worth the upgrade price for me. I haven't yet tried doing two videos simultaneously, one on each core.

I think I'm now a believer in the whole idea that clock speed does not necessarily predict performance.
 
A few comments here and I hope that somebody with more knowledge or drive to look stuff up will comment also:

1. The idea of wider execution paths is not new. There have been architectures around like that for awhile. Intel went down that path big time in a non 386 compatible processor that it designed with HP to be its big high performance chip. Initially it was to go into servers only. I haven't heard about it for awhile, but then I haven't been working as an electrical engineer for three years or so.

2. AMD went down a very different path and put all its next generation chip bets into just making 386 compatible processors go faster. They also avoided the Rambus debacle that must have cost Intel a ton. So AMD seems to have gained a little market share here by avoiding the disastrous strategic decisions that Intel made. If AMD had made the same disastrous strategic decisions there probably wouldn't be an AMD around today.

3. Processor speed is only one aspect of processor designs that relates to performance. My, only partially informed guess, is that if you took an 8086 and revved up it processor speed by building it with newer devices but kept its architecture the same its performance would be less than 10% of a modern processor's.

A little esoteric digression here about item 3. The 8086 architecture doesn't really lend itself to just having its clock speed revved. An 8086 did not have internal cache and couldn't really be run internally at higher clock speeds. The way a modern processor is able to run at much higher clock cpeeds than could ever be used effectively on a PC board is to overclock a much slower clock that is connected to it via the PC board. The instructions codes that the processor actually executes are retrieved from a very fast onboard cache so that the processor can run like mad until every now and then it tries to execute an instruction that isn't in its cache or it tries to process some data that's not in it's data cache and then it has to go to the offboard RAM to get the data. But at some point these occasional retrievals by the processor of offboard data or instructions really starts to slow things down. The 8086 could just retrieve data on a 16 bit path from RAM that was quite a bit slower than today's RAM. So newer processors have much wider paths into the processor that allows data from RAM to be retrieved much faster.

The bottom line here is that clock speed is only one of many fairly complex issues that determine the performance of a computer.
 
davefoc said:
An 8086 did not have internal cache and couldn't really be run internally at higher clock speeds. The way a modern processor is able to run at much higher clock cpeeds than could ever be used effectively on a PC board is to overclock a much slower clock that is connected to it via the PC board. The instructions codes that the processor actually executes are retrieved from a very fast onboard cache so that the processor can run like mad until every now and then it tries to execute an instruction that isn't in its cache or it tries to process some data that's not in it's data cache and then it has to go to the offboard RAM to get the data.
Click to expand...

It will be nice if in the future all of a PC's RAM will be of the very fast internal variety and RAM modules become obsolete.
 
davefoc said:
A few comments here and I hope that somebody with more knowledge or drive to look stuff up will comment also:

1. The idea of wider execution paths is not new. There have been architectures around like that for awhile. Intel went down that path big time in a non 386 compatible processor that it designed with HP to be its big high performance chip. Initially it was to go into servers only. I haven't heard about it for awhile, but then I haven't been working as an electrical engineer for three years or so.


The bottom line here is that clock speed is only one of many fairly complex issues that determine the performance of a computer.
Click to expand...

According to Zep, the current AMD is really a legacy of the DEC 64bit RISC processor, which has been dropped. That is why AMD was able to get a 64 bit Intel processor out so quickly, they already ran the Intel architecture on what was internally a 64bit RISC machine.
 
stormer said:
So how would you upgrade or add to it?
Click to expand...

You wouldn't. Processors would come with an adequate RAM size, eg 2 or 4 GB for the previous generation of P4s. Although external RAM modules could still be available for a few special purpose PCs.
 
Long, long ago in a job far, far away, I did some consulting for both AMD and Intel. The Intel folks used to joke that they could give a few days' production to AMD just so AMD could finally claim to break the 1 (hundred thousand/million??? - wish I could remember the real number) chip per year production level. Obviously things have changed greatly.
[/derail]

CT
 
El Greco said:
You wouldn't. Processors would come with an adequate RAM size, eg 2 or 4 GB for the previous generation of P4s. Although external RAM modules could still be available for a few special purpose PCs.
Click to expand...

Yup, I think this would work well. After all, a 64 bit OS could optimally address 2 GB or 4 GB of RAM.
 
a_unique_person said:
According to Zep, the current AMD is really a legacy of the DEC 64bit RISC processor, which has been dropped. That is why AMD was able to get a 64 bit Intel processor out so quickly, they already ran the Intel architecture on what was internally a 64bit RISC machine.
Click to expand...

Both Intel (since the Pentium Pro) and AMD (since the K5) have been decoding the x86 instructions into easier to execute ones. Those could be described as RISC instructions, though that's a huge simplification. The important part here is that first decoding x86, then executing them is way faster than directly executing them.

The reason why I would not call them RISC is because that's a design philosophy aimed at using simple instructions, to simplify processor design and speed up performance. At the cost of higher memory usage. Modern x86 processors are actually more complicated because of the decoders and fairly memory-efficient.

The comparisson with DEC 64bit RISC is interesting. What happened is that Dirk Meyer used to work for DEC when they produced some high performance processors, then moved to AMD and headed K7 development. So basically the K7 development team used much the same ideas that were being applied at DEC, courtesy of Meyer. But the legacy is mostly spiritual, apart from the EV6 bus AMD didn't copy any of DEC's implementations.

AMD was able to launch a 64 bit x86 processor before Intel, because the latter didn't want to launch one as it would compete with their Itanium line. Dirk Meyer didn't like AMD's K8 design (headed by the same guy who developed K6), so they threw it out. Instead, AMD ripped the guts out of their K7, redesigned all components with improvements where needed and bolted on a memory controller. Hence, K8. With Prescott being a dud that was more than good enough.

I think Intel got overconfident at some point during the nineties. They were trying to stear the market towards Rambus and Itanium. But with AMD offering a credible alternative the market decided to not follow Intel, and that cost them dearly.

While Intels Conroe/Merom is definately the best CPU at the moment, I think its relative excellence is mostly due to P4 being a dud and K8 having had only minor improvements. AMD has been complacent lately.
 
AMD still has a better memory controller (its simply 'faster'), while Intel has been surviving on its better cache architecture (its simply 'smarter')

So AMD performs better during a cache miss but Intel processors typically have fewer cache misses overall.

In the end it doesnt matter much.. if you want price vs performance than you still gotta go AMD - if you want the best raw performance possible (to date) then you still gotta go with Intels top ($$$) end.

Nothing much has changed.
 
rockoon said:
Nothing much has changed.
Click to expand...
I wouldn't say that - three months ago for the best raw performance you needed to get an FX series processor from AMD. AMD has been raw performance leader all the time between the introduction of the Athlon64 until Intel launched Core2. (Which sounds like an 80's action movie to me, btw)
 
You must log in or register to reply here.

Back
Top Bottom