Barton launches!

[Dave H]

Reviews of the Athlon XP 3000+ with new Barton (512k L2) core:

• Ace's has perhaps the most interesting review, benching a truly huge variety of recent games and content creation programs, as well as a great examination of the cache characteristics to kick it off. Negative points: game benchmarks only use GF4; not all benchmark results directly comparable due to changing platforms; SPECviewperf benches possibly biased due to bad AMD drivers for Quadro4?
• Anand's review compares just about every CPU imaginable, and even has handy graphs charting AXP vs. P4 performance scaling. Negative points: uses PC800 for the P4 (why??); had worse overclocking luck than others
• Tom's review also compares gazillions of CPUs, but in this case half of them seem to be overclocked. Lots of information, including underclocked numbers to simulate Barton-based 2800+ and 2500+, but graphs difficult to read. Several pictures of a toddler at end, which may be a plus to some
• [H]OCP has relatively dull benchmark choices, but presents numbers vs. P4 seperately from vs. Thoroughbred, which may make the comparisons easier to understand

Overall the extra 256k L2 seems to net an average of around 3-4% over the otherwise identical (2.167 GHz) 2700+ Thoroughbred. It seems difficult to claim this is worth 300 Quantispeed points; 200 would seem to be the best one can do without straining credibility. Nonetheless, the 3000+ is at least able to clear the Thoroughbred based 2800+ in the large majority of benches, if only by around 1% overall.

Comparison to the 3.06 GHz P4 with HT is generally negative, although a lot depends on the benchmarks chosen and P4 platform used. (Everyone uses nForce2 for the AXP, of course.) Tom's seems to be using 850E with PC1066 (I say seems because he also mentions a Granite Bay, but that looks like a typo), and he seems also to have the greatest margin of victory for Intel. [H] uses an 845PE, and is the only one where the 3000+ seems to slightly beat the 3.06 GHz P4, but IMO the benchmark selection plays perhaps a larger role in this than the choice of P4 chipset.

In the end, it's a solid but boring part. It may be slightly disappointing to those that expected a similar performance boost from Barton that Northwood gave over Willy, but various differences in the organizations and performance of the cache hierarchies on the respective parts made this unrealistic. Or to those that assume, as AMD insists, that the extra cache really is "worth" 300 points.

OTOH I was ever so slightly favorably surprised. The cache seems to be "worth" a good 150-200 points, which is towards the high end of what I expected. The availability looks very good (has a spot on pricewatch already), especially in contrast to the last two AMD launches. And both Tom and [H] managed to OC to ~2500 MHz (3400+?) without resorting to anything too extreme, which augurs reasonably well for AMD's ability to at least execute on the 3200+ Barton part on their roadmap.

Of course the big event in the AMD/Intel race is the launch in April of 800 MHz FSB P4s and the adjoining dual-channel PC3200 865P and 875 chipsets. It will be very interesting to see performance and price of those combinations, which should determine if AMD can keep any serious presence at the high-end until Athlon64 launches in September.

 

[BenSkywalker]

Continuing on from the line of discussion in the other thread.

Looking over all of the reviews of the XP 3K all of them seem to have some serious flaws. Anand is using slower RAM then he should be for the P4 while he doesn't specify what he is running the RAM @on the nForce. Ace is using an antiquated graphics card and is running both platforms off of peak, Tom's is just so screwed up it is hard to tell what is going on and Kyle's is a bit light on the benches. Commenting one way or the other on Barton right now isn't something I would want to do in comparison to the broader subject.

The initial link that I believed you supplied comparing the XP 28 to the P4 2.8 over at Anand's had a bit of a problem, namely he crippled the XP running it in single channel memory configuration. I'd like to point to his review of the P4 3.0GHZ-

http://www.anandtech.com/cpu/showdoc.html?i=1746&p=10

It has the 3.0, 2.8 and 2800 on their ideal current platform. The Intel parts running on the 850 w/1066 and the Athlon on the nF2 w/dual channel. Going through the benches and looking at the place each chip finishes in for each benchmark the 3.0GHZ P4 averages out to 1.82 with the XP 2800 @2.36 and the P4 @2.8GHZ coming in at 3.14. There is a larger edge for the XP 2800 over the P4 2.8 then the P4 3.0(with hyper threading enabled) has over the XP 2800. Given a wide range of benches, the XP 2800 comes out ahead of the P4 3GHZ 10 times out of 28, ahead of the 2.8 17 out of 28. I think that the rating on the XP 2800 is still well within the reasonable range, based on the overall bench representation it seems they could have added another 100 to the score.

This ignores the pricing advantage between the XP2800 and P4 2.8 both on their ideal platforms. Running the P4 2.8 using a reasonable RAM configuration for a high end chip, 2x256MB, you are out $236(all prices NewEgg) while you are out $98 or $148 for 166/200MHZ DDR respectively. You save yourself ~$100 for a setup that wins more often then it loses against its direct competitor just on RAM costs(not to mention the ~$50 rift in mobos for like features). Comparing the two mobos that Anand used, which are priced roughly the same, you get Firewire, an extra LAN, six channel audio vs two, serial ATA v ATA133, and extra USB 2.0 ports on top of your savings due to RAM running the XP 28 over the P4 2.8. Better performance, more features and lower price.

I don't really understand how it can be said with a straight face that AMD can't compete on the high end right now. Does Intel have an edge at the highest end? I would say overall yes at the moment, although why anyone would spend double the money for ~5% rift is beyond me(looking at the 3.0 or 3000, not just Intel's parts). Even comparing the 2800/2.8 their price premium over the 2400/2,53 certainly isn't in line with the performance edge they offer. Ignoring that though, AMD is either neck and neck or edging out Intel in most benches at the same clock/PR, and costing less doing it.

Are they the insanely better deal they were a year or two ago? I don't think so, although I'm about to overhaul my rig and stacking the current chips up against each other there is no way that Intel's price premium is close to worth the price difference, particularly when outside of the 3.0/3000 range they are actually a bit behind overall in performance and still more expensive. I'm looking at a Asus A7N8X Deluxe with 512MB and a XP~2400 at the moment, if someone can point me to an Intel alternative with feature and performance parity within $100 I'd take a look at it. Make it even money and I'd be very interested. If it's cheaper, then I'd almost certainly put it together over the XP setup but my browser shows it costing either ~$150 more for like performance and feature incomplete or ~$300 more with parity.

Of course the big event in the AMD/Intel race is the launch in April of 800 MHz FSB P4s and the adjoining dual-channel PC3200 865P and 875 chipsets. It will be very interesting to see performance and price of those combinations, which should determine if AMD can keep any serious presence at the high-end until Athlon64 launches in September.

Given Intel's pricing at the high end, why isn't Opteron a viable counterpart? I know what AMD's public stance is, they also said the Athlon was their pro part when it launched and the K6 core was the consumer part.

 

[Johnny Rotten]

Of course the big event in the AMD/Intel race is the launch in April of 800 MHz FSB P4s and the adjoining dual-channel PC3200 865P and 875 chipsets. It will be very interesting to see performance and price of those combinations, which should determine if AMD can keep any serious presence at the high-end until Athlon64 launches in September.

Given Intel's pricing at the high end, why isn't Opteron a viable counterpart? I know what AMD's public stance is, they also said the Athlon was their pro part when it launched and the K6 core was the consumer part.

Two reasons in my mind:

1. The opteron chip is going to be extremely expensive. Its an enterprise/server chip that is going to be competing with the xeons and itaniums for sales. The chip will likely be at least twice the price of any athlon or P4.

2. I doubt initial opterons are going to be as fast as a 3.2ghz+ P4.

 

[overclocked]

quick Q!

I can´t find my pcb number on my mainboard where should it stand?

 

[Ty]

Didn't AMD state that there will be a Desktop version of the Opteron?

 

[BenSkywalker]

The opteron chip is going to be extremely expensive. Its an enterprise/server chip that is going to be competing with the xeons and itaniums for sales. The chip will likely be at least twice the price of any athlon or P4.

I wouldn't be putting Opteron in the same class as Itanium, an entire Opteron system is likely to cost less then an Itanium CPU. Current best price I can find for an Itanium processor is $2,758, quite a bit more then a decent dual Athlon MP system would cost. The highest end Xeon on the other hand can be had for under $500(less then the top tier P4 or Athlon XP although it is a 2.8 part instead of a 3GHZ).

Looking at it vs the Xeon, the Athlon MP is currently in that same market and there really isn't that much of a difference in price between either of them and the P4/XP desktop processors(although the premium obviously isn't worth it to typical end users). The Opteron should be priced against the Xeon which is certainly within the same general range as the desktop counterparts.

I doubt initial opterons are going to be as fast as a 3.2ghz+ P4.

In terms of clock speed I find it extremely unlikely, in terms of performance from what I have read it seems likely it will be well within the range of the highest end P4's-

With yields high, production ramping up and strong support from their partners – it seems as if Hammer is finally ready for its debut. The motherboards and chipsets are quite stable and with final CPU silicon being produced the only question that remains is how fast the first CPUs will be. At Comdex last year AMD had a handful of Athlon 64s running at 1.8GHz and they’ve been targeting a ~2GHz launch all along, it will be interesting to see how much farther above 2GHz AMD is able to hit by the time the CPU launches. From what we’ve seen, a 2GHz Athlon 64 will definitely be competitive with Intel’s 3.06GHz Pentium 4 but at this point the market is expecting as much as possible from AMD.

http://www.anandtech.com/cpu/showdoc.html?i=1773&p=6

They had 1.8GHZ parts up and running a year ago and people who have seen the chips in action say that they will be able to compete with P4s clocked 50% higher. Considering how much of a premium Intel places on new platforms I would expect pricing between the Opteron and highest end P4 solution to be quite comparable. Time will tell if that is correct, but already the x86 based server chips are certainly within the same range.

Overclocked-

Which mobo do you have? The all aren't always in the same spot

Ty-

That's the Athlon64 and it doesn't hit until September while the Opteron will be out soon.

 

[overclocked]

I have the A7V8X, if uppgrading to a Barton(mmm ) you should have pcbversion 1.04, i think i have the latest revision cause in the bios there already a option for 333mhz fsb(not overclocking mode), just want to be sure, ok i´m 99% sure it´s support it but just for the case of it!

One Q more, you that have a ASUS mainboard, have you hade any problems with the EZ-flash Bios?

 

[Dave H]

The initial link that I believed you supplied comparing the XP 28 to the P4 2.8 over at Anand's had a bit of a problem, namely he crippled the XP running it in single channel memory configuration.

Aha! I was begining to wonder about that first review, because the same chips are much more competitive in Anand's new Barton review, using ostensibly the same configuration. I'll consider those results thrown out and perhaps if I want to take the time do a similar analysis using some of the slew of benchmarks released last night.

I'll step back, then, and say that, with platform costs taken into consideration, AXP is probably price/performance competitive everywhere except perhaps the top two speed bins. The fact still remains that the performance value of a Quantispeed point has crumbled since the scheme was introduced. Giving the extra 256k of L2 a value of 300 points appears to be a serious further devaluing. OTOH, assigning a clock speed increase of 83 MHz a value of 100 points would seem to increase the performance value of a point, so to the degree that they stick with that scheme, it's a good thing. (Of course the roadmap only goes to Barton 3200+ before Athlon64 introduces an entire new discontinuity, and with it the opportunity to either do the Right or the Wrong Thing.)

In general, though, the whole system is a hopeless muddle, with scores clearly dictated more by marketing concerns than actual performance. Probably the main reason I harp on the problems with the ratings system so much is because I actively supported it when it was introduced, believing that AMD would stay true to their word and ground it in an objective performance-based metric. Maybe they really intended to. But as manufacturing problems shredded their roadmap time and again, this became less and less of an option.

My original point (which started all this fuss) is that AMD's widespread reputation as the obvious price/performance winner stems from the time 2-3 years ago when buying AMD really was an amazing deal. This enhanced reputation seems to have allowed them to get away with increasingly worse price/performance on their newer chips, to the point where the discount that any generic brand should carry has completely disappeared at the high end. And to get the sort of "AMD discount" that existed across the entire line 1.5 years ago, now you need to go way down to the 1800 (+/MHz) bracket or so.

The apparently misleading Anand scores I posted kind of deflected the discussion away from that, but I certainly stand by my original point 100%.

Given Intel's pricing at the high end, why isn't Opteron a viable counterpart? I know what AMD's public stance is, they also said the Athlon was their pro part when it launched and the K6 core was the consumer part.

They said the Athlon was their performance part; that's quite a bit different from being their server part, which is what Opteron is. And while Opteron's 128-bit wide memory bus could certainly help in desktop performance, the main difference between Opteron and A64--the 3 HT links instead of 1 for up to 8-way multiprocessing--is 100% an server/HPC feature. Going by the prices that've been bandied about, there's no way Opteron will be a desktop part; and at the clock speeds they'll be launching at, it won't be performance competitive in a single CPU situation either.

And there's this: Hammer is...fabulously late (we'll just leave it at that), and wildly anticipated by many computer enthusiast consumers; AMD is bleeding cash and generally in not so good financial shape; Hammer taped out over a year ago already with at least dozens of perfectly working samples circulating since that time; chipset infrastructure finished and on display at industry trade shows for over half a year now; AMD has absolutely no presence whatsoever in the enterprise market and no server OEM contracts even with its proven architecture, Athlon MP, not to mention a new one. And yet AMD postpones Athlon64, the consumer part, until September. And releases Opteron, the server part, in April.

How does this make any sense?

One answer only: yields. AMD is obviously still having trouble getting significant volumes at competitive clock rates out of their SOI process. Now, if AMD launches a consumer chip in April and they only have a few tens of thousands of them to sell before September, you'd get huge waits and everyone screaming paper launch and so forth. Whereas, if AMD launches a server chip in April and manages to sell a few tens of thousands of them by September, that's a surprising success.

The clincher is that they added a 1 MB L2 Athlon64 part, which is a ridiculous waste of die space for the consumer market, unless you're trying to make up for slow clock speeds, and even better if you have a rating system that lets you trade off L2 size for clock speed at whatever exchange rate you come up with. (Wait a second: doesn't including more cache increase the number of defects per die and thus decrease yields? No. Chips are designed with a few redundant cache sub-arrays to practically eliminate the chance of losing a die because of fab defects in the cache.)

Meanwhile, AMD will want to maximize the ASPs on whatever Hammers they can put out, and despite the fact that it'd be very surprising for Opteron to net a major OEM win at launch, I think there's plenty of demand to unload tens of thousands of them at a very high price. Consider that the large cache Xeons (which you need to go >2 CPUs) run at least $1500 with 1 MB L2, and something like $2500 for a 2 GHz with 1 MB L2. And that's as high as they go: 2.0 GHz! And they all share a single 3.2 GB/s DRAM bus! Opteron is going to cream them performance wise, even at the relatively low launch speeds we've been hearing. And there's enough people in that market running Linux to make the 64-bitness worth something even in the absence of a gold Windows64.

To recap: Opteron is not designed for the desktop (no, not even the performance desktop), and the fact that AMD is releasing Opteron first indicates that volumes will be very low. Enough demand probably exists for AMD to sell out those very low volumes at >$1000 ASPs, and that is exactly what they want to do.

So no, it shouldn't be viewed as a competitor to the 800 MHz FSB P4 with 875 chipset. Although hopefully that won't stop AMD from handing out review samples to the benchmark sites so we can see the comparison anyways!

 

[BenSkywalker]

I'll step back, then, and say that, with platform costs taken into consideration, AXP is probably price/performance competitive everywhere except perhaps the top two speed bins. The fact still remains that the performance value of a Quantispeed point has crumbled since the scheme was introduced.

I'd say that even the 2800XP still has a decent edge in performance alone, it bested the P4 2.8 in 17 out of 28 benches that Anand ran as a general example. When you add in the pricing edge, it makes if fairly clear cut from my point of view. Barton is certainly confusing things, in some instances its additional cache seems to more then cover the rating advancement it has, while in others it fails to compete with the 2800. I'm hoping that we will see some better benching results in the next few days. Anand does have a chart tracing AMD's PR rating v Intel's clock rating and their resultant impact on performance for each bench, Barton bounces back and forth betwee accelerating and decelerating the effectiveess of the rating scheme, rarely falling in to the linear category. With the current benches we have seen I wouldn't be putting money on the Barton retaining the overall performance edge over the 3GHZ P7, but when price is factored in I still see a ~$150 rift not being nearly worth a few percentage points overall.

As far as them staying in line with their PR lines, I'd say that the Barton may be the first chip to step out of line, although I thought the same about the 2800XP until some decent nF2 w/dual mem benches started showing up. If a chip wins a decent amount more often then it loses in a performance comparison, I'd say that is the better overall performing chip. In the case of the P4 v XP outisde a few tasks where the P4 tends to be significantly faster and vice versa they are within spitting distance of each other comparing their PR v MHZ numbers. I'd say so far AMD has done a very good job at maintaing their system.

My original point (which started all this fuss) is that AMD's widespread reputation as the obvious price/performance winner stems from the time 2-3 years ago when buying AMD really was an amazing deal.

But why has the market changed? Was it AMD?

http://www.anandtech.com/showdoc.html?i=1351&p=2

That is from October of 2K. The highest end Intel part was ~$1200 while the highest end AMD part was ~$500. Over the last few years AMD's highest tier chip price has raised slightly due in part to their far greater mindshare(although that doesn't always translate to marketshare), while Intel's prices have dropped by ~50% due to pressure from AMD. You are correct in your assertion that AMD is not the screaming bargain compared to Intel that it was two or three years ago, however their value relative to the market changed due to Intel taking a massive hit in their pricing structure, not due to anything AMD has done.

This enhanced reputation seems to have allowed them to get away with increasingly worse price/performance on their newer chips, to the point where the discount that any generic brand should carry has completely disappeared at the high end.

AMD has been around as long as Intel, I'm not sure what you mean by generic. Generic tends to be a perception that the mass market has about certain products. Oreo cookies are generics, and they cost the same as the 'name brand' that they outsell 20 to 1 or so. Coke outsells Pepsi on a global basis by a very wide margin, yet people don't consider Pepsi to be generic. The situation we are looking at currently is more akin to Coke lowering their price to $.99 from $1.99 because Pepsi was too much of a threat at $.89. Pepsi's value is only changed in relative terms due to Coke dropping the price. Same thing here with AMD.

And to get the sort of "AMD discount" that existed across the entire line 1.5 years ago, now you need to go way down to the 1800 (+/MHz) bracket or so.

This is entirely a function of supply v demand. The 2800XP was constantly sold out for months at its launch price, AMD likely could have and should have marked them higher then they did. When you get to where the yields are high enough, such as in the 1800XP range, they bottom out the price on the chips as they don't have the kind of OEM presence to move as many chips as they can make on virtue alone and the enthusiast market won't swallow all of the chips they are producing.

They said the Athlon was their performance part; that's quite a bit different from being their server part, which is what Opteron is.

AMD has absolutely no presence whatsoever in the enterprise market and no server OEM contracts even with its proven architecture

These two relate directly to each other particularly combined with the pricing structure for the Athlon MP. AMD has no OEM contracts for high end server parts, and really the only big inroads they stand to gain with Opteron is Linux64 server based solution which is a very, very small niche. AMD's MP line was supposed to be very high end in the x86 market, instead of commanding the four figure price tag it dragged down the price of Xeons an enormous amount.

As far as AMD's financials go, I've been tracking the company for years. They are at their normal non dot com boom state. For the months following the launch of the Athlon they had a like market valuation, like losses and slightly lower market share(they had yet to peak). Their market cap is currently less then the value of the real estate they own by a signficant margin. If they were in dire trouble I would expect Big Blue or Mot to chew them up, Fab30 offers an incredible value right now v AMD's market cap ignoring the rest of the company.

And yet AMD postpones Athlon64, the consumer part, until September. And releases Opteron, the server part, in April.

How does this make any sense?

We also tried to get some more clarification involving the delay of Hammer, as it did not seem to be related to manufacturing issues. If you take into account the die size differences between Hammer and Athlon XP and compare the yields, currently Hammer is yielding just as high as the 0.13-micron Athlon XP CPUs and apparently yields have not really been a problem during the Hammer production process. Instead, AMD reiterated that the reason for the delay in the CPU’s launch was the performance and design of the part. Once all of the performance issues were hammered out (no pun intended) and the design was finalized then AMD was able to crank it up a notch and work towards the final stretch before the CPU’s release.

http://www.anandtech.com/cpu/showdoc.html?i=1773&p=6

Yields aren't the problem.

Meanwhile, AMD will want to maximize the ASPs on whatever Hammers they can put out

I disagree. If the yields are what are being reported their primary concern will be marketshare. Of course they will try and land the highest price they can while maintaining nigh sell out status, but marketshare should and almost certainly will be their primary concern. They are going to play it out like the Opteron is direct competition to Itanium on a PR basis while making every attempt to chew up the Xeon's market. Prior to the launch of a Win64 based OS the enthusiast market is too big to ignore.

To recap: Opteron is not designed for the desktop (no, not even the performance desktop), and the fact that AMD is releasing Opteron first indicates that volumes will be very low.

Or it could be a situation where AMD is using the lead time to have yield levels rise to the point of having a major mismatch in price/performance v the P4 before they make a major play at the consumer market. Based on all the comments on yields I have seen they are simply far too high to have this chip placed in the enterprise server niche, AMD doesn't have a chance of moving that many parts in that market.

I guess to summarize my end of the discussion, I see AMD having the same value they always have, Intel has simply greatly improved on that front. AMD still dominates the enthusiast market, and there is still validity as to why that is the case from a price performance perspective. No, it isn't the raging bargain it was two or three years back, but it's still better then Intel overall.

Overclocked-

Not entirely sure where it is, let me see if I can find it, looking at the images I can find of it it appears that it should be near your floppy connector, but that is based of relatively low jpegs so take it for what it's worth

 

[Mulciber]

I'd say that even the 2800XP still has a decent edge in performance alone, it bested the P4 2.8 in 17 out of 28 benches that Anand ran as a general example. When you add in the pricing edge, it makes if fairly clear cut from my point of view. Barton is certainly confusing things, in some instances its additional cache seems to more then cover the rating advancement it has, while in others it fails to compete with the 2800. I'm hoping that we will see some better benching results in the next few days. Anand does have a chart tracing AMD's PR rating v Intel's clock rating and their resultant impact on performance for each bench, Barton bounces back and forth betwee accelerating and decelerating the effectiveess of the rating scheme, rarely falling in to the linear category.

I dont think that will change much due to any benchmarking methidology change. Certain applications do not excede the need for 256k cache, and the 2800 T-bred wins by virtue of its higher clock speed.

 

[Mulciber]

OTOH I was ever so slightly favorably surprised. The cache seems to be "worth" a good 150-200 points, which is towards the high end of what I expected. The availability looks very good (has a spot on pricewatch already), especially in contrast to the last two AMD launches. And both Tom and [H] managed to OC to ~2500 MHz (3400+?) without resorting to anything too extreme, which augurs reasonably well for AMD's ability to at least execute on the 3200+ Barton part on their roadmap.

Of course the big event in the AMD/Intel race is the launch in April of 800 MHz FSB P4s and the adjoining dual-channel PC3200 865P and 875 chipsets. It will be very interesting to see performance and price of those combinations, which should determine if AMD can keep any serious presence at the high-end until Athlon64 launches in September.

I really dont think AMD will be able to keep up with Intel on the high end unless they can manage to bring the Barton up to around 2.3ghz on a 400mhz FSB. The 800mhz FSB and clock speed will just be to much.

As far as them being able to compete again with the Athlon64 is anyone guess. My impression though will be that with the introduction of the Athlon64, it will be tremendously less expensive to produce motherboards for the system. This thanks to the serial interconnects via hypertransport and also to single chip motherboard designs such as nVidias (where the AGP controller is located on the south bridge). Things like this could very well bring high-end system boards for AMD into the $50 range easily. Whereas I predict Intels 800mhz dual channel ddr motherboards will still hover in the $150 to $250 price range, like they have been for the past several years.

 

[Dave H]

As far as them staying in line with their PR lines, I'd say that the Barton may be the first chip to step out of line, although I thought the same about the 2800XP until some decent nF2 w/dual mem benches started showing up. If a chip wins a decent amount more often then it loses in a performance comparison, I'd say that is the better overall performing chip. In the case of the P4 v XP outisde a few tasks where the P4 tends to be significantly faster and vice versa they are within spitting distance of each other comparing their PR v MHZ numbers. I'd say so far AMD has done a very good job at maintaing their system.

You seem to be forgetting that the system began with "1 point = 1 Thunderbird MHz" in performance. Our disagreement seems to be whether today at the high end "1 point = 1 P4 MHz" or "1 point < 1 P4 MHz". Thunderbird level IPC is nowhere in sight.

Just to demonstrate the point, between the original Quantispeed rated chip, the 1.33 GHz 1500+, and the 2.13 GHz 2600+, clock speed increased 60% but the rating increased 73.3%, which is to say the rating increased 22.2% faster than the clock speed. None of the changes to the core made over that stretch (die shrink and relayout) modified same-clock performance one iota. The PR ratings "stayed in line" alright; unfortunately it's a line that doesn't cross the origin of the Cartesian plane. (That is, according to the scaling factor used from 1500+ to 2200+, a theoretical Athlon XP 0+ would actually be clocked at 333 MHz!)

Meanwhile, if we were talking clock speed, one would expect a larger FSB boost over that time period than just 266->333 MHz. At least AMD only "charged" us around 50 points for the FSB boost.

But why has the market changed? Was it AMD?

http://www.anandtech.com/showdoc.html?i=1351&p=2

That is from October of 2K. The highest end Intel part was ~$1200 while the highest end AMD part was ~$500.

Only one problem with that comparison (which is actually from Nov. 10, 2000, BTW.) The $1200 Intel part was the 1.5 GHz P4. Which launched on Nov. 19, 2000, 9 days later Meanwhile, the highest priced Intel part that was officially on sale at the time was the 1 GHz PIII, priced at...$463.

So, why has the market changed? Why, even after a huge tech recession and with significantly less reason for end users to upgrade their systems than back in late 2000, is Intel's consumer top-of-the-line priced 30% higher now? Was it AMD?? Of course it was. In late 2000 AMD's yield sweet spot was in the 1000 MHz range, and they were pricing their high-end CPUs to move, because their main goal was to establish marketshare in the performance sector. Back then, Intel was lowering their prices to match AMD's. (Albeit on a "ranked" basis--fastest = fastest, 2nd fastest = 2nd fastest, etc. Nevermind that they were 2 speed grades behind.)

Nowadays, AMD's yield sweet spot is in the 1.7 GHz range, which is the low-midrange. They can't affect Intel's pricing one bit. (After all, nearly all their chips are 2400+ and slower, which is exactly where Intel's P4 line fails to match AMD's prices!) Instead, what's happening nowadays is that AMD is raising the prices on their high-end chips to match Intel! Why? Because they want to generate enough cash to keep in business until Hammer is ramped.

For the months following the launch of the Athlon they had a like market valuation, like losses and slightly lower market share(they had yet to peak).

They had also yet to start selling Athlons in volume. It is an interesting comparison you make, though, because then as now their share price had been pummelled after they completely failed to execute on their public roadmap for over a year and were left making huge quarterly losses and without the confidence of the investing community. Then as now AMD was widely regarded as only a few quarters away from bankruptcy. Last time the Athlon swooped in and saved the day, primarily due to its extremely rampable (in .25 and .18 processes) design. Will Hammer do the same?

Big maybe.

Their market cap is currently less then the value of the real estate they own by a signficant margin.

What about their debt and other liabilities?

If they were in dire trouble I would expect Big Blue or Mot to chew them up, Fab30 offers an incredible value right now v AMD's market cap

You think anyone wants more fab capacity in this environment?? Mot's semiconductor division is barely hanging on itself. As for IBM...it's a possibility, especially due to the similarities (and now cooperation) between IBM and AMD's fab technology, but they're certainly in no hurry. If Hammer doesn't save the day IBM will be able to get a much better price by waiting.

http://www.anandtech.com/cpu/showdoc.html?i=1773&p=6

Yields aren't the problem.

I should have made myself more clear: yields at salable clock speeds. Perhaps I should have pointed out that most of those dozens of samples running around last year were clocked at 800 MHz, and now seem to be averaging maybe 1.4 GHz. No, some handpicked cores at 1.8 GHz does not impress.

It's not that the quote at Anand is wrong, you just need to know how to parse it. AMD is indeed making circuit level changes to Hammer in an attempt to get higher performance, but they're not changing the high-level design in ways that will change performance at a particular clock. Rather, they're stamping out problems resulting when circuit designs from their traditional process fail at high clock speeds when ported to SOI. Thing is, the sorts of circuit techniques that can squeeze extra clockability out of a normal process just don't work on SOI; instead you have to figure out new ones. This learning curve is the reason why the only major ICs to use SOI up 'til now have been from design houses like IBM and Motorola that rely heavily on automated functional cells in their designs rather than low-level custom tweaking as Intel, AMD and the Alpha (in particular) do.

I guess to summarize my end of the discussion, I see AMD having the same value they always have, Intel has simply greatly improved on that front.

I see it differently: I think Intel's value proposition (to the consumer) has been increasing significantly since the onset of serious competition from the K6 in 1998. Even though Intel eventually couldn't keep up, they nonetheless ramped their performance dramatically in the 1999/2000 "MHz race" timeframe, and lowered prices to boot. The transition to P4 was an awkward time, but despite a few bumps along the way I think the progress has been overall pretty steady.

Whereas AMD still retains most of its value on the low end, but has certainly fallen behind its previous great value on the high end.

 

[Mulciber]

Just to demonstrate the point, between the original Quantispeed rated chip, the 1.33 GHz 1500+, and the 2.13 GHz 2600+, clock speed increased 60% but the rating increased 73.3%, which is to say the rating increased 22.2% faster than the clock speed. None of the changes to the core made over that stretch (die shrink and relayout) modified same-clock performance one iota. The PR ratings "stayed in line" alright; unfortunately it's a line that doesn't cross the origin of the Cartesian plane. (That is, according to the scaling factor used from 1500+ to 2200+, a theoretical Athlon XP 0+ would actually be clocked at 333 MHz!)

This isn't making any sense dave. You state that they don't change their core any in those two periods, which is true, but you're leaving out the fact that they changed over to 333mhz bus in that time, and THAT is why they had another PR number change. And its obvious that switching to a faster bus warrented a different PR number.

You then go on to mention the 333mhz bus, but dont factor that at all into your graph.

 

[Mulciber]

http://www.anandtech.com/showdoc.html?i=1351&p=2

That is from October of 2K. The highest end Intel part was ~$1200 while the highest end AMD part was ~$500.

Only one problem with that comparison (which is actually from Nov. 10, 2000, BTW.) The $1200 Intel part was the 1.5 GHz P4. Which launched on Nov. 19, 2000, 9 days later Meanwhile, the highest priced Intel part that was officially on sale at the time was the 1 GHz PIII, priced at...$463.

This also fails logic. Right now you yourself are compairing the prices of the Barton 3000+ with the PIV 3ghz. The later has been out for a couple months now while the former came out YESTERDAY.

 

[Althornin]

DaveH, i think you make unwarranted assumptions.

 

[Dave H]

This isn't making any sense dave. You state that they don't change their core any in those two periods, which is true, but you're leaving out the fact that they changed over to 333mhz bus in that time, and THAT is why they had another PR number change. And its obvious that switching to a faster bus warrented a different PR number.

Err, no. I'm talking about the 266 FSB 2600+, which runs at 2133 MHz. There's also a 333 FSB 2600+, clocked at 2083 MHz. Sorry for the confusion, but my comparison is correct.

This also fails logic. Right now you yourself are compairing the prices of the Barton 3000+ with the PIV 3ghz. The later has been out for a couple months now while the former came out YESTERDAY.

And how much did it cost 10 days ago? In any case, the street price of the 3000+ is now $598 according to pricewatch, which is only $10 more than the list price of $588. Even if we go according to the list price it is essentially priced even with the 3.06 GHz P4; there is a $30 difference or so, but then again FWIW it's not called a 3066+.

The 2800+ is currently holding to its list price at pricewatch (which incidentally is dead even with the 2.8 GHz P4's list), although that may be due to very low volumes. The 2600+, for example, is running about $50 below list.

But all of this seems irrelevant, as I don't believe I've made any comments about Athlon XP pricing that would be significantly undermined when and if the 3000+ drops $60 in price. Meanwhile, saying that Intel's top CPU price in late 2000 was $1200 is completely misleading. It may seem silly now, but the launch of an entirely new Intel core was a big deal and for some odd reason a very few people were willing to pay absurd amounts of money in order to get one two weeks early. (Or at least a very few retailers thought they could try to rip off a very few such people; who knows if they succeeded?)

The release of the 3000+, or of the 3.06 GHz P4, while each does introduce minor architecture changes, are not big events and certainly don't attract that kind of speculation. Prices may drop from a bit over list (at launch when allocation is tight) to a bit under in the first month or so, but there's no comparison to that so-called $1200 P4.

 

[BenSkywalker]

Mulciber-

I really dont think AMD will be able to keep up with Intel on the high end unless they can manage to bring the Barton up to around 2.3ghz on a 400mhz FSB. The 800mhz FSB and clock speed will just be to much.

Given the actual yields of the TBred Bs that shouldn't be too much of a problem barring an accelerated ramping from Intel.

DaveH-

You seem to be forgetting that the system began with "1 point = 1 Thunderbird MHz" in performance. Our disagreement seems to be whether today at the high end "1 point = 1 P4 MHz" or "1 point < 1 P4 MHz". Thunderbird level IPC is nowhere in sight.

Not so much forgetting what it started out as, more like factoring in the hefty improvement the P4 is showing clock for clock vs its debut. Take a 3GHZ P4 on its ideal platform and down clock it to 1.5GHZ and compare it against a Tbird.

Just to demonstrate the point, between the original Quantispeed rated chip, the 1.33 GHz 1500+, and the 2.13 GHz 2600+, clock speed increased 60% but the rating increased 73.3%, which is to say the rating increased 22.2% faster than the clock speed. None of the changes to the core made over that stretch (die shrink and relayout) modified same-clock performance one iota.

But platform changes did make quite the difference in performance. Compare a 2600+ on its ideal platform compared to a TBird on its ideal platform when the ratings were introduced.

Only one problem with that comparison (which is actually from Nov. 10, 2000, BTW.) The $1200 Intel part was the 1.5 GHz P4. Which launched on Nov. 19, 2000, 9 days later Meanwhile, the highest priced Intel part that was officially on sale at the time was the 1 GHz PIII, priced at...$463.

November 24th, $929 for same chip, AMD $492

http://www.anandtech.com/showdoc.html?i=1365&p=2

December 15th, $859 for same chip from Intel, AMD $293

http://www.anandtech.com/showdoc.html?i=1383&p=2

December 29th $849 same from Intel, $274 from AMD

http://www.anandtech.com/showdoc.html?i=1391&p=2

............

March 9th, $577 for Intel, new chip from AMD $309

http://www.anandtech.com/showdoc.html?i=1432&p=2

So, why has the market changed? Why, even after a huge tech recession and with significantly less reason for end users to upgrade their systems than back in late 2000, is Intel's consumer top-of-the-line priced 30% higher now?

Four months with the same chip in excess of $575 from Intel back in 2K in to 2K1.

Nowadays, AMD's yield sweet spot is in the 1.7 GHz range, which is the low-midrange. They can't affect Intel's pricing one bit.

I think the actuality is that AMD has certainly placed Intel under pricing pressure. Thousand dollar chip launches for consumer parts are something that haven't been too common in a couple of years. As far as yields, where are you getting your information from? I can't find a single source to verify it and in fact am seeing widespread reports of how incredibly overclockable the Tbred-Bs are(been researching for my current purchase). Given the hundreds if not thousands of posts on the subject I decided to pick up an 2100XP as they seem to hit clock rates in the 2.2GHZ-2.4GHZ range with air cooling on a not infrequent basis(clearing 2.1GHZ seems to be pretty much a given, even with OEM cooling). A lot of people reccomend the XP1700 as even it seems to constantly break 2GHZ with air cooling. When I see chips with that kind of headroom being the norm I tend not to believe vague reports of yield issues. If the GFFXs that made the review rounds could all hit 600MHZ+ I doubt anyone would be talking about it being cancelled.

It is an interesting comparison you make, though, because then as now their share price had been pummelled after they completely failed to execute on their public roadmap for over a year and were left making huge quarterly losses and without the confidence of the investing community. Then as now AMD was widely regarded as only a few quarters away from bankruptcy. Last time the Athlon swooped in and saved the day, primarily due to its extremely rampable (in .25 and .18 processes) design. Will Hammer do the same?

The comparison being similar to today was exactly my point

What about their debt and other liabilities?

Fab30 is worth more then twice their current market cap, and that's if we forget about Austin. I understand your question relates directly to what their exact debt level is right now, but no way could they get a credit line that deep with their mark cap. They are paying off the loan for Dresden at the moment, but they have been working on that for some time already now.

If Hammer doesn't save the day IBM will be able to get a much better price by waiting.

I think that is a given. For IBM they will give themselves a viable x86 presence right off if they so desire, and with clearly superior management and the ability to give AMD exclusive status on IBM PCs and lower end servers it could work out quite well for them. IBM has shown in the past that they would like to enter the x86 CPU market, they just haven't had the kind of oppurtunity a crumbling AMD would represent.

I should have made myself more clear: yields at salable clock speeds.

Given that that is the definition that Anand always uses, along with pretty much everyone that I am aware of, I thought that was a given.

Perhaps I should have pointed out that most of those dozens of samples running around last year were clocked at 800 MHz, and now seem to be averaging maybe 1.4 GHz. No, some handpicked cores at 1.8 GHz does not impress.

Anand's tour of Fab30 was quite recent, some time after the public display of Hammer @1.8GHZ. Do you have anything to back your assertions of what is going on in terms of Hammer? Obviously the move to SOI brings with it a learning curve much as any reasonably drastic manufacturing change will. The point stands that it was reported that yields are quite good which according to every standard I've ever heard applied(and certainly Anand's standard) is that the chips are of a saleable clock rate.

I see it differently: I think Intel's value proposition (to the consumer) has been increasing significantly since the onset of serious competition from the K6 in 1998. Even though Intel eventually couldn't keep up, they nonetheless ramped their performance dramatically in the 1999/2000 "MHz race" timeframe, and lowered prices to boot.

This portion states that Intel has improved their value to consumers which is what I said

Whereas AMD still retains most of its value on the low end, but has certainly fallen behind its previous great value on the high end.

Due to Barton? It is the first chip I recall being priced much outside of their typical range when it is available in quantity.

 

[overclocked]

tHE headroom for the Barton seem´s now to be like 2500Mhz+ to when hammer arrives.

AMD had to overcome ONE big thing and that is the intel inside logo,tralalaa.. You know.

I have seen AMD been able to push itself up to the market of Intel and beat them on their own plane..

AMD has had a bad year with inventories has been upp and cpu market has asked for more.

AMD is one of the few companies that i se can beat Intel and live up to the hopes we have from AMD.

AMD vill make a strong year 2003 and continue to provide it´s knowledges to future hired at AMD.

 

[epicstruggle]

Although Im a bit dissapointed with what AMD came out with(PR rating might be a tad bit too high) Ill more than likely purchase it when I can get some money together. Dont want them to fold under.