Slashdot Mirror
Coppermine vs. Athlon
SaDan writes "I checked out a comparison of the new Intel Coppermine processors and AMD's Athlon chips at Tom's Hardware Guide last night. It's kind of interesting, and I thought others would be curious about how Athlon stacks up against Intel's latest offering. "
The referenced review basically shows that Intel once dropped to .18 and clocked 5% faster can pretty much keep pace with AMD's shipping product for $150 more.
Add to that Coppermines OLD core, Athlons scalability and Fab30 coming online soon and it is clear that INTC is going to have to do more than in the past to stay relevant.
If it isn't great news for AMD investors, this is at least great news for CPU buyers as INTC will have to WORK for it's money for a change!
It's always good to see the underdog getting ahead. Athlon seems to be the superior game performer, except for quake 3. But still, it really looks like AMD is becoming the best choice for gamers, not only because it's faster, but because it's cheaper.
Maybe they should both seek seperate markets, AMD should go for gaming/low cost 3D workstations, and Intel should stick to servers and the like. That may even out the market, and get rid of some of this "do everything we can to keep the competitors down" attitude. I mean really, when you have that much of the market, can they really be that much of a threat?
Start working towards real technological advances, rather than mediocre enhancements to beat the competition.
it still pushes AMD to improve their design, perhaps to the point that the K7 can also be used in portable devices. i think that the major appeal of intel's newest processors will be in the laptop arena. it would be awesome to have the same power in a laptop as on the desktop with an x86 processor.
i really hate to say it, but it's getting to the point where it can be cheaper to buy a new system than to upgrade an older one (especially with DRAM prices so high.) of course, with the introduction of new processors and technology, older (obsolete?) models become much cheaper, and hey, a better processor is a better processor. competition is good, as long as the competition doesn't force either company out of business. that would be bad.
With the recent attitude of "innovate, retaliate, counterretaliate" in the chip market, both companies are really giving the consumers a lot to look forward to. The only thing I wonder is, how long will it be before Intel and AMD are forced to slow their shrinking price to performance ratio in order to protect their bottom line?
On an unrelated gripe, "Tom's Hardware Guide" isn't helping its credibility much by having a major error in the very first sentance. "pushed from its thrown"? Ugh.
(Of course, anyone that should be considering these high-end processors should be competent enough to know that performance is only positively correlated with the number of MHz. They don't run lock-step...)
(Of course, there aren't many really-high-powered laptops; there was only ever one Alpha-based laptop, few SPARC-based, and such... I agree with others that availability of faster AMD chips in the K6 series is more important at this time...)
(I half-expect to see a report from VA-Linux Systems some time soon...)
If you're not part of the solution, you're part of the precipitate.
I'm hoping someone can clarify something for me. If I understand correctly, the slot that Pentium II/III's go in is called Slot 1. Intel patented Slot 1 so that other companies could no longer make replacement processors.
So AMD had to stick with the Socket370(?) form (which was the non-proprietary Pentium Pro socket.) Until they came out with the Athlon. AMD (or somebody) created the Slot A socket which is suprisingly(not!) similar to Slot 1 but not compatible. Main question: Is Slot A proprietary also? Or can other companies make processors for it other than AMD?
NOTE: The above is based on many guesses, assumptions, things I've read, and things I may have imagined reading. Please correct me or clarify (I'm sure there are errors.)
numb
?syntax error
There's a reference to K6-III-Ps at speeds of 350, 366, and 380 MHz in AMD's retail employee website (registration required), and there's this less-detailed information in their public website. I don't know how any of this translates to notebooks that you can actually buy that have this processor...K6-2-Ps are available at speeds up to 475 MHz, and people seem to be fixating on megahertz alone. (Not that the K6-2's bad...I have one myself and it runs like a champ, but the K6-III, from all I've seen, is substantially faster, especially at high clock speeds where the L2 cache speed difference gets totally out-of-control.)
20 January 2017: the End of an Error.
The main thing that AMD has gained out of this whole Athlon thing is respect. Real respect. Not just, "Gee, since I'm a very poor geek, I'm going to put an AMD chip in my box."
To illustrate: my college's career fair was just a few days ago. AMD was there. They have been there in the past. In the past, only die-hard computer engineering hardware geeks talked to them. This year, however, as they had Athlon processor periphenalia and even a couple actual processors (none of which they were giving away) there was a line. I'm talking about a long line. Everybody wanted to talk to teh AMD guys. I waited in it for over twenty minutes then decided that since hardware doesn't make me feel an extreme amount of joy inside I would go talk to someone else.
Bright Young Minds (at least, I think that's what we are) are taking notice of AMD and are intensely interested in being hired by them. This seems like a Good Crowd to have on your side when speculating on the future.
The argument can backfire: it looks like they had to go .15 and .13 microns to reach 500 and 550 Mhz. If so, how do they expect to reach 1 Ghz ? FYI, the Alpha 21264 at 600 Mhz is 0.35, at 800 Mhz ~0.25 and at >= 1 Ghz is expected to be 0.18 micron.
First off, check your core voltage. Going from 0.25 um to 0.18 um requires you to lower your voltage, possibly to a point your MB doesn't offer. Best of luck to you!
They didn't HAVE to go smaller to get more speed. They (Mot) did so they could reduce costs. The more chips you squeeze onto a wafer, the more money you generate from said wafer. That's all there is to that argument.
Intel drags it's heels in order to maximize it's profit from each successive generation of tech, only switching when they absolutely cannot do anymore with what they've got (witness the 7th generation x86).
Don't think that all the other semi-conductor co.'s follow their footsteps in that regard.
I don't know about the rest of you, but I am VERY happy that AMD is giving Intel a run for our money.
Personally, I was getting REALLY tired of the "new" and exciting "innovative" processors a whole 33Mhz / 50Mhz faster than last quarters every quarter.
Since the Athlon has been out, P3 prices have been plummeting; I've been keeping track
of the fall in prices; there is NO WAY prices would have fallen so low without the Athlon goosing Intel!
I am very much looking forward to testing some Coppermine's, as soon as I can get my hands on them.
--------- Webmaster, http://www.cpureview.com and
Why are CPU's considered more "cool" than other components? Why do people spend $250 more to gain a few MHz when the same money will get them more memory and a faster drive? I think it's because it's easy to quantify. People like that MHz is one simple number.
So what if the Athlon is a tiny bit more value for the money - I'll get a lot more value if the price of a really good 19" monitor drops a couple of hundred bucks.
Every article I have read on that site has at least 1 point that gets on my nerves due to a low tech understanding of the issues. He claims that its a compiler difference that causes the big jump in performance on the Coppermine with the Q3 benchmark.
Lets think about this for a minute! If Intel didn't change the SSE core then why would a compiler with SSE changes produce a binary that ran better on the same SSE core? The answer: Because its not the compiler! A better answer would be that the changes intel made to the cache result in significant performance gains in some situations. Tom dumps their tech doc's on what they did:
1 they increased the associatively of the cache
2 they widened the L2 data path to the CPU
3 they decreased the latency
4 Lastly they decreased the size by 1/2.
In general 1 and 4 tend to cancel themselves to give similar performance (pick up an architecture book and read about caches if you don't know what I'm talking about here) So we are left with 2 and 3. Now 2 and 3 tend to allow you to get to the cache faster and get more data per cycle. Now quake is really an tiny engine (significant amount of the time it supposedly fits in L1) accessing a massive amount of data. Now lets assume that quake is so tight that it manages to fetch its data out of L2 cache a very large percentage of its time (as opposed to windows just randomly switching tasks, and using main memory like a big disk cache) now if suddenly your data loads which were always in the cache get to the processor faster keeping it from stalling a pipeline for 5 or 6 cycles what happens?
Memory architecture is a __BIG__ deal with modern CPU's. A very large percentage of time on modern CPU design is spent trying to optimize data accesses. The intel engineers have done their homework. The PC market now considers games the standard benchmark (Quake being the main one, Celery-vs-K6! When was the last time you out typed Word? On the other hand when was the last time your Celery helped you kick that poor K6 owners ass because you were getting an extra 20fps?) so they discovered a way to help quake out while maintaining decent performance with data sets that were more sensitive to cache size rather than access times.
they didn't HAVE to go smaller to get more speed.
Just a couple of niggles here. I'm not a hardware engineer, but I do believe that's wrong. At smaller feature sizes capacitors get more efficient and switching gets faster. Because your capacitor is more efficient you can use lower voltage. As voltage decreases so does power consumption, as the square. Less power consumption = less heat, so higher clock rates.
They (Mot) did so they could reduce costs. The more chips you squeeze onto a wafer, the more money you generate from said wafer.
Errr, somewhat correct. Yield plays a big part in the equation - as feature size goes down, so does yield, especially since new untried manufacturing processes have to be brought on line each time feature size ratchets down.
The bottom line is that smaller features size is good - very good.
Life's a bitch but somebody's gotta do it.
...has to do with how many memory different places in the cache each memory address can reside. This is not a measure of size, rather a measure of versatility. That is, in a direct mapped cache each memory address can be cached in one and only one cache line. This obviously leads to an overlap since main memory is alwasys larger than the cache. So, if a particular memory location is cached in a particular line, and another memory location that must be mapped to the same line in the cache is accessed, then the first must be flushed no matter how "fresh" it is.
A two way set associative cache allows any memory address to be placed in one of two locations in the cache. A four way cache has four spots where each memory address can be cached, etc. Again this is not a matter of size, rather a measure of how flexible the cache is.
The more "ways" the cache has the more flexible it is and this results in fewer flushes and overall more "fresh" data in the cache. This is what a cache is all about. That is, keeping the data that is needed right now, right here close to the CPU.
These "X ways set associations" are expensive in terms of logic and chip space. Ideally, a cache would be fully associative and allow any memory address to be cached anywhere in the cache memory. Because this stuff is expensive, it is usually reserved only for the highest performance parts, that is, the level 1 cache which is the closest to the processor core and usually the smaller one. AFAIK, all mainboard caches are direct mapped. They compensate by usually being bigger and even thoough they are slower, they are still a good bit faster than main memory, but nowhere near as fast as a level 1 cache or register access.
According to the previous poster, Tom got the two caches backwards. I don't have a data book on the new chip, but I'd really be surprised if they actually made an 8 way set associative cache that is 256K in size. No biggie, but it's an obvious error to those of us that know something of what the h*ll he's talking about.
As far as his comment on the benchmarks goes, I have no idea where he's coming from on that one.
Good judgement comes from experience, and experience comes from bad judgement.
- W. Wriston, former Citibank CEO
Disadvantage of FCPGA: the name is extremely silly.
--
Win dain a lotica, en vai tu ri silota
1) room for much more of the expensive L2 cache for full speed operation and modules up to 2MB.
2) redesign to the AGTL+ bus, which allows four processors to run, even using modules (versus the normal P6 bus, delivering two way SMP with the module, four way with the older socket).
3) more money. Intel knows that people who need 4-way or better SMP systems will pay for this, often to a foolish extent. Intel loves to milk some sector for high margins, and you can guess this won't happen when there's a direct (or near so) replacement like K6 or Athlon.
Ok, so now consider PIII Xeon with only 256KB of L2 cache. Certainly, this is the same chip as you get when you buy a PIII-regular. But of course, it's on the Slot 2 module, and if you want a four processor (or better) SMP system, you have no other choice. Intel basically has you, and they like it that way.
In fact, I'm surprised it only around $50.
-Dave Haynie
Yes. They are. Though to be fair you have to also consider that they typically pick (if I remember correctly) names of streams and rivers to name the experiments by- apparently there was a Coppermine River or something, so naturally this name they're keeping.