Wild share speculation and CPU wars...
Finally I see the X2 processors for sale at Dabs. Quite pricey units - cheapest at £330. I was considering upgrading my desktop to one of those about 2 months ago, but stayed my hand, and bought the epia instead. My Athlon XP gets little enough serious use as it is, so as much as I would like it I cannot justify it.
What was also interesting is the Pentium D's were no where to be seen. Come on Intel - where are all those cheap dual processors that you have been bragging about?AMD announced theirs after yours! More vapourware. Well if the price reports of what the Intel beasties should cost are right a 2.8 will cost about 150 quid. Putting aside the extra 50 quid you'll need a year to keep it running, thats a cheap chip.
Must state at this point I have been an AMD shareholder for a while, and have been disapointed more times than I care to remember about their performance. I watched them soar to 45 dollars, then crash to 8 or so. I have watched them bounce up and down more times than I care to remember since then. They simply ran out of products after the first K7 and took an age to get the A64s out. This time round things are definitely different. Here's why: they are not chasing the volume market any more.
Previous attempts K6, and K7 were targetted at the volume. No server offering - or when there was it was a secondary concern. This time the Opterons are the key - what has to work in a server can easily be down tuned to work in a desktop. Turn off the more paranoid stuff like ECC memory, and you get a quick boost in performance making the desktop chips a bit quicker, for little work.
Also by targetting the server, you have to solve reliability issues out of the box. This can be a double edged sword if you have a problem, so if they stumble in the next few months that will probably be the cause.
The Importance of ASP
Server chips have another key advantage: high price. For a company competing against a mamoth like Intel, whom you can never out produce, you need to make quality product. ASPs (average selling prices) are the key metric in how much a chip manufacturer can make. At the end of the day, the chip is the product, so getting the most for each should number one in every chip manufacturers interest. Intel understood this way before AMD, and helped by having a better product could keep prices higher. Intel never decimated its margins against AMD even at its lowest ebb. I expect Intel realized as they do now, the number of chips the little upstart can physically make is much smaller than the market demands. They can, from their monopolistic position, control how many chips AMD sells by minor tinkering with their ASPs.
So whats different this time? Well by having the best in the high end, AMD have high ASPs, rather than the best in the mid range which puts them much more at the mercy of Intel. If their chips are percieved as the best, they win those with big cash to spend on machines. Add this to the fact they need to make less of such high end chips for the same revenue, and you can see a rosey picture emerging. Given time the highend moves down. Look at nVidia. Originally few mid range offerings, now they're their biggest selling item.
AMD production volume may not be anywhere near Intel's, but it is still capable of churning out a lot of chips. If that number increases, along with a large increase in ASPs, they could pull some rabits out of the bag.
The Fight
AMD have been hit hard in the gut by flash sales of late. Intel obviously saw this coming attack, and want to starve AMD of cash whilst they make a counter attack. It was a clever move. Those precious ASPs have been maintained for both parties, and bought them time. 6 months is a long time in chip design.
However hard they hit though, they didn't solve the problem: they are loosing the hearts and minds of the top spenders. The best server solutions now have AMD. Power is a big factor in racks, and rack space is precious. A wise IT manager will look long and hard at AMD before Intel if they want most bang for their buck. The dual processor offerings make AMD even more compelling - there are no Dual Xeons about. You see Intel have always built from the bottom up - first you design a desktop chip, then you shape it into a server chip. The time saving they make from this decision over the years has paid dividends (literally). They beat up Alpha, and even their own IA64 by having the benefits of huge sales and money. Fixing the design to do the best it could at the low end proved to product chips that were better than the high end.
So why is it different this time? AMD expect to make desktop chips. Their server chips are not the old skool "buy 16 for one box per company" type of chips, they are product tested high end desktop chips masquerading as server chips. This is the same shift that nVidia get with their high-end ultra cards. They are the same chip design sold at a massive premium. Its noticable they haven't done any super-expensive chip offerings. Most server chips have always been available with enormous caches (I think 32 Meg on an Itanium). They would dilute the effort from AMD's goal of producing a fast desktop. Testing and making server chips in odd configurations is expensive, as each derivative needs lots of tuning. AMD merely take it to the 8xx series - which may be a model for future desktops anyway, with lots of HT channels. Caches and unusual server stuff stays off the cards.
The Wildcards
The wildcard for me is the mobile market. Intel are driving the market that way, and laptops are very desirable items. A 760 Pentium M is an expensive chip, and AMDs offerings in this area are weak. A decent laptop is something people are prepared to pay for, and the rapid increase of horsepower means they can be used as desktop replacements for most people in business. Added to the fact Intel's platform strategy has led to a much better experience on Intel lappies than AMD, Intel are the dominant force here. AMD are niche in this area, and the new Turions are not good enough.
Intel would love all those desktops to be replaced by laptops, however I cannot see that happening. Laptops still have key weaknesses:
* Screen size and more imporant positioning. Attaching the screen to the keyboard makes them uncomformtable to use for long periods. Yes you can buy a big brick, but a docked laptop is more comfortable.
* Disk speed. Users of intensive apps will really notice the difference between 5400 and 7200rpm desktop drives - especially since many laptops have less memory.
* Cost. They have got cheaper, but a desktop buys you more. If an employee never leaves the office, the expense of buying a laptop is not worth it.
* Security. Laptops open up companies to loosing more than just a machine if its left on a train. It also makes important data move around, something IT departments are already worried about. Virus attacks that do not come from emails are most likely brought in by a laptop.
So I think there will always be a balance. I also don't think the recent Laptop craze is AMDs problem. Its a diversionary tactic from Intel. Intel is disappearing from desktops, so as the biggest player they shout about their new expanding market. Best not remind the analysts of their shrinking market share.
Early Intel Comeback
What is apparent is Intel know all this, and are not exactly floudering around. They have made some extremely smart decisions, such as the flash attack, and Centrino. In their labs they are beavering away at chips that will decimate AMD's current line up, so for AMD to secure their position, they need to have a compelling next generation. Dual cores are now yesterdays news as far at the battle goes. AMD have won the first round, but Intel could easily pull it back. The centrino platform is coming to the desktop. This will make the commodity Intel machine a much more consumable item with decent drivers and the "it simply works"-ness AMD are still lacking. Few people notice machine speed, but all notice machines crashing. AMD need to pick a partner (or a small number of partners), and stick to them like glue to ensure they have the same rock solid platform. I have had experience of 3 nForce chipsets, and I can say though good, they have a long way to go before getting to Intel chipsets level of robustness (see recent the Tom's Hardware soak test of the dual cores. The nForce 4 for Intel collapsed.)
Conclusions
For me AMD are in the driving seat for the next 6 months. A large increase in ASPs in in the offing, and their highend line up looks good. The A64 has good power usage, so they are heading in the right direction to meet the Centrino on the desktop. They also have much better FPUs - something that Intel is no doubt labouring to rectify. Intel has shot themselves in the foot with the P4, as that made all code much more heavily reliant on FPU than it would have been. No doubt new compilers will redress this balance. Luckily the A64 is pretty good at both.
Things to look out for:
* Intel pushing out a dual core desktop Centrino before the end of the year. This could be a major headache for AMD, looking at current roadmaps.
* Production problems. The 90nm process seems to be still a little slow getting chips out. Availability of chips in shops is a good way to gauge this. Friends who wanted to buy Winchester cores found few available a few months ago.
* Laptops really replace desktops despite my earlier arguments.
Good luck AMD, you'll need it!
What was also interesting is the Pentium D's were no where to be seen. Come on Intel - where are all those cheap dual processors that you have been bragging about?AMD announced theirs after yours! More vapourware. Well if the price reports of what the Intel beasties should cost are right a 2.8 will cost about 150 quid. Putting aside the extra 50 quid you'll need a year to keep it running, thats a cheap chip.
Must state at this point I have been an AMD shareholder for a while, and have been disapointed more times than I care to remember about their performance. I watched them soar to 45 dollars, then crash to 8 or so. I have watched them bounce up and down more times than I care to remember since then. They simply ran out of products after the first K7 and took an age to get the A64s out. This time round things are definitely different. Here's why: they are not chasing the volume market any more.
Previous attempts K6, and K7 were targetted at the volume. No server offering - or when there was it was a secondary concern. This time the Opterons are the key - what has to work in a server can easily be down tuned to work in a desktop. Turn off the more paranoid stuff like ECC memory, and you get a quick boost in performance making the desktop chips a bit quicker, for little work.
Also by targetting the server, you have to solve reliability issues out of the box. This can be a double edged sword if you have a problem, so if they stumble in the next few months that will probably be the cause.
The Importance of ASP
Server chips have another key advantage: high price. For a company competing against a mamoth like Intel, whom you can never out produce, you need to make quality product. ASPs (average selling prices) are the key metric in how much a chip manufacturer can make. At the end of the day, the chip is the product, so getting the most for each should number one in every chip manufacturers interest. Intel understood this way before AMD, and helped by having a better product could keep prices higher. Intel never decimated its margins against AMD even at its lowest ebb. I expect Intel realized as they do now, the number of chips the little upstart can physically make is much smaller than the market demands. They can, from their monopolistic position, control how many chips AMD sells by minor tinkering with their ASPs.
So whats different this time? Well by having the best in the high end, AMD have high ASPs, rather than the best in the mid range which puts them much more at the mercy of Intel. If their chips are percieved as the best, they win those with big cash to spend on machines. Add this to the fact they need to make less of such high end chips for the same revenue, and you can see a rosey picture emerging. Given time the highend moves down. Look at nVidia. Originally few mid range offerings, now they're their biggest selling item.
AMD production volume may not be anywhere near Intel's, but it is still capable of churning out a lot of chips. If that number increases, along with a large increase in ASPs, they could pull some rabits out of the bag.
The Fight
AMD have been hit hard in the gut by flash sales of late. Intel obviously saw this coming attack, and want to starve AMD of cash whilst they make a counter attack. It was a clever move. Those precious ASPs have been maintained for both parties, and bought them time. 6 months is a long time in chip design.
However hard they hit though, they didn't solve the problem: they are loosing the hearts and minds of the top spenders. The best server solutions now have AMD. Power is a big factor in racks, and rack space is precious. A wise IT manager will look long and hard at AMD before Intel if they want most bang for their buck. The dual processor offerings make AMD even more compelling - there are no Dual Xeons about. You see Intel have always built from the bottom up - first you design a desktop chip, then you shape it into a server chip. The time saving they make from this decision over the years has paid dividends (literally). They beat up Alpha, and even their own IA64 by having the benefits of huge sales and money. Fixing the design to do the best it could at the low end proved to product chips that were better than the high end.
So why is it different this time? AMD expect to make desktop chips. Their server chips are not the old skool "buy 16 for one box per company" type of chips, they are product tested high end desktop chips masquerading as server chips. This is the same shift that nVidia get with their high-end ultra cards. They are the same chip design sold at a massive premium. Its noticable they haven't done any super-expensive chip offerings. Most server chips have always been available with enormous caches (I think 32 Meg on an Itanium). They would dilute the effort from AMD's goal of producing a fast desktop. Testing and making server chips in odd configurations is expensive, as each derivative needs lots of tuning. AMD merely take it to the 8xx series - which may be a model for future desktops anyway, with lots of HT channels. Caches and unusual server stuff stays off the cards.
The Wildcards
The wildcard for me is the mobile market. Intel are driving the market that way, and laptops are very desirable items. A 760 Pentium M is an expensive chip, and AMDs offerings in this area are weak. A decent laptop is something people are prepared to pay for, and the rapid increase of horsepower means they can be used as desktop replacements for most people in business. Added to the fact Intel's platform strategy has led to a much better experience on Intel lappies than AMD, Intel are the dominant force here. AMD are niche in this area, and the new Turions are not good enough.
Intel would love all those desktops to be replaced by laptops, however I cannot see that happening. Laptops still have key weaknesses:
* Screen size and more imporant positioning. Attaching the screen to the keyboard makes them uncomformtable to use for long periods. Yes you can buy a big brick, but a docked laptop is more comfortable.
* Disk speed. Users of intensive apps will really notice the difference between 5400 and 7200rpm desktop drives - especially since many laptops have less memory.
* Cost. They have got cheaper, but a desktop buys you more. If an employee never leaves the office, the expense of buying a laptop is not worth it.
* Security. Laptops open up companies to loosing more than just a machine if its left on a train. It also makes important data move around, something IT departments are already worried about. Virus attacks that do not come from emails are most likely brought in by a laptop.
So I think there will always be a balance. I also don't think the recent Laptop craze is AMDs problem. Its a diversionary tactic from Intel. Intel is disappearing from desktops, so as the biggest player they shout about their new expanding market. Best not remind the analysts of their shrinking market share.
Early Intel Comeback
What is apparent is Intel know all this, and are not exactly floudering around. They have made some extremely smart decisions, such as the flash attack, and Centrino. In their labs they are beavering away at chips that will decimate AMD's current line up, so for AMD to secure their position, they need to have a compelling next generation. Dual cores are now yesterdays news as far at the battle goes. AMD have won the first round, but Intel could easily pull it back. The centrino platform is coming to the desktop. This will make the commodity Intel machine a much more consumable item with decent drivers and the "it simply works"-ness AMD are still lacking. Few people notice machine speed, but all notice machines crashing. AMD need to pick a partner (or a small number of partners), and stick to them like glue to ensure they have the same rock solid platform. I have had experience of 3 nForce chipsets, and I can say though good, they have a long way to go before getting to Intel chipsets level of robustness (see recent the Tom's Hardware soak test of the dual cores. The nForce 4 for Intel collapsed.)
Conclusions
For me AMD are in the driving seat for the next 6 months. A large increase in ASPs in in the offing, and their highend line up looks good. The A64 has good power usage, so they are heading in the right direction to meet the Centrino on the desktop. They also have much better FPUs - something that Intel is no doubt labouring to rectify. Intel has shot themselves in the foot with the P4, as that made all code much more heavily reliant on FPU than it would have been. No doubt new compilers will redress this balance. Luckily the A64 is pretty good at both.
Things to look out for:
* Intel pushing out a dual core desktop Centrino before the end of the year. This could be a major headache for AMD, looking at current roadmaps.
* Production problems. The 90nm process seems to be still a little slow getting chips out. Availability of chips in shops is a good way to gauge this. Friends who wanted to buy Winchester cores found few available a few months ago.
* Laptops really replace desktops despite my earlier arguments.
Good luck AMD, you'll need it!
Comments