Is AMD Doomed?

[MikeC]

Is AMD Doomed? It's a good oped piece by Loyd Case at extremetech.com that nicely encapsulates the Intel/AMD rivalry over the past few years, and AMD's current situation.

My answer to his question is no. I think even Intel would move to save AMD if necessary.

[jaganath]

what happened:

1) Intel took their eye off the ball, went to sleep at the wheel, whatever metaphor you want to use.
2) AMD snuck in through the back door that Intel held open for them, and started increasing their market share to a viable level.
3) rot set in with AM2. negligible performance increase vs. 939, why upgrade?
4) now the killer blow, Conroe. Intel is back with a vengeance, and focused on keeping AMD where they belong, on the ropes. Each new die shrink and microarchitecture delivered either on or ahead of time ('tick-tock' strategy) is a body blow to AMD, who are now losers on price, performance and watts.

where did AMD go wrong? by not capitalising on that short period of time when their Athlon 64 architecture blew Netburst out of the water, and also by buying ATI (where are the synergies from this deal? who benefited apart from ATI stockholders?).

I think even Intel would move to save AMD if necessary.

hard to imagine. why help out a competitor? no doubt antitrust regulators would get involved if Intel was left as the sole chip manufacturer, but that doesn't necessarily mean bailing out AMD.

[MikeC]

why help out a competitor?

Lots of reasons, a big one being that an enemy keeps your own people edgier & more alert. Besides, buying a big piece of your competition w/o taking over completely can be very useful and profitable.

[Das_Saunamies]

AMD will always be the cheap champion. Just like ATI gained from being the integrated/preinstalled manufacturer of choice when they came along, so will AMD benefit from having the cheapest price/performance ratio.

Definitely agree Intel would rather intervene and take advantage of AMD rather than let them sink.

[bendit]

i agree that intel needs amd around, even if they don't want them around. but I don't believe they would be allowed to invest in them. Samsung, or a large Taiwanese concern seems a natural to eventually take over AMD. IBM is a long shot.

[aristide1]

Is AMD Doomed? It's a good oped piece by Loyd Case at extremetech.com that nicely encapsulates the Intel/AMD rivalry over the past few years, and AMD's current situation.

My answer to his question is no. I think even Intel would move to save AMD if necessary.

One would ask if there was no AMD then would Intel look like MS to most people; a monopoly, whether or not it actually is one? From what I recall GM and Ford did not want Chrysler to go belly up in the 1980s.

Other - Even if Intel had taken a bigger market share during Netburst, would they be in a better position now?

They had to switch to DDR2-memory, performance aside those super high DDR2-800 numbers look more appealing than DDR-400. Why confuse people with latency issues? And if the numbers didn't do it initially then the price certainly would sooner or later, as demand switched accordingly. But if Intel is about to get a new boost in speed by putting the NB on the die then why hasn't AMD done better than what they have now, since it's already there

Comparing my X2 4000+ to my E6400 I called it 70% of the performance at 30% of the price. If AMD can continue that kind of value they would still have a place, but their new quads are nothing to write home about, they're only now becoming available on AM2+ sockets, and the models priced like a Q6600 sure don't run like a Q6600.

As they said in Full Metal Jacket - They are in a world of s__t.


I must admit though, for folding, a Biostar TForce board with built in video for $70 has a lot of initial appeal. I do wish the 5000+ Black Edition had come with twice the cache. I mean really, they removed the lock and the heat sink and sell it for double the price of a 4000+. What is that? Besides apparently good marketing?

[djkest]

Comparing my X2 4000+ to my E6400 I called it 70% of the performance at 30% of the price. If AMD can continue that kind of value they would still have a place, but their new quads are nothing to write home about, they're only now becoming available on AM2+ sockets, and the models priced like a Q6600 sure don't run like a Q6600.

AMD seems to have a pretty good hold on the budget market. Perhaps not so much in the hearts and minds of enthusiasts, but they are still viable.

I'm pretty sure I heard just a couple months ago AMD shipped more CPUs than Intel did for the first time ever.

AMD has a wide range of CPUs which are good for 90% of the population for under $100. And they save $50 on average compared to their intel counterparts. Also consider most people dont' overclock, and that takes another advantage away from Intel.

[fri2219]

One issue that wasn't discussed in that op-ed was the incredible bad timing of the ATi purchase. Just as they needed to focus on getting the new chips out the door, they shot themselves in the foot by adding all the distractions and finance issues that come from a consolidating purchase.

[jaganath]

I'm pretty sure I heard just a couple months ago AMD shipped more CPUs than Intel did for the first time ever.

unlikely, given that Intel has a market share of over 70% by CPU shipments.

AMD has a wide range of CPUs which are good for 90% of the population for under $100.

as does intel.

And they save $50 on average compared to their intel counterparts.

verrry debatable.

[walle]

Fanboys in the room

[mr. poopyhead]

AM2 was one of the biggest mistakes, imo...

with core2 looming on the horizon at the time, why would they leave all of their 939 customers out in the dark? the whole 939/K8 era was what put AMD into the public eye... before that, only forum nerds like us knew AMD even existed...

new socket, same architecture... what's the point?

[Das_Saunamies]

I know I wanted to stab the idiot who decided they wouldn't be making processors for my then brand-spanking-new S939 board.

Regarding budget processors, I'd rather buy one from AMD's 'proper' line than Intel's 'budget' line. Never will I touch a Celeron ever again!

[gentonix]

and also by buying ATI (where are the synergies from this deal? who benefited apart from ATI stockholders?).

AMD wanted to build a platform consisting of a chipset, processor and graphics card, and now they've got one (altough not a very convincing one). In the future they're probably going to release another platform for laptops to compete with Intel's hugely succesful Centrino platforms, since the current AMD laptop platform isn't much of a platform, but more of a statement that says "anything goes". That and the future CPU-GPU integration* were the main concerns I think. The move was made way too late, though.

*) http://en.wikipedia.org/wiki/AMD_Fusion

[Tzupy]

I read that Extremetech article when it was published, and it's not worth mentioning.

The main reason for the performance of Conroe is not the wide execution power, but the low latency access to memory, due to the large cache and smart hardware prefetching.
A Prescott with the same amount of cache and smart prefetchers could have matched the C2D performance, of course at twice the power (due to higher frequency).
The (now) multi-threaded software I'm writing still suffers from memory accesses on my C2D. In code areas with low memory accesses, performance doubles from single-threaded, but the gain drops to 0 in code areas with high memory acceses. So IMO when it comes, Nehalem will give a big boost for multithreaded applications.

AMD didn't make a mistake going AM2, they had to go DDR2, for the sake of their future quad-core - that got delayed.
So the move to AM2 was just hasty and they sure made a mistake dropping 939 so quickly.

Another mistake was to stick to the 'purist' quad-core. If they would have put toghether two dual-cores in one package, they could have had a quad-core a year before Intel.
For very short distances (same package) the HT connections could run extremely fast and almost negate the advantage of single-die quad-core.

Now for AMD it's only a matter of speeding up execution: new revision of their 65 nm Phenom, allowing for higher fequencies, then the move to 45 nm and then... well, Bulldozer!

[~El~Jefe~]

yeah. i think AMD is toast.

back to being toast. sux.

Phenom 2.4 should be 199 dollars and then it would be king though.

All amd boards for the past 5 years have been 1/2 the price with 3x the features and connections on them. On die mem controllers make for no difference in performance across chipsets and brands (1-3% at most). I am free to purchase ANY motherboard with whatever connections I want and not worry about retarded p35 or 965 differences. 65 dollar motherboard = 250 dollar motherboard in stock performance numbers for amd.

ball is dropped and 45nm Penryn just pounds so far in efficiency and low idle wattage.



Really though, at this point, there isnt much of a need for quad core. the area currently lacking is Hardrive speed, Vista 32 only using 2.5-3.0 gigs, and video cards not being able to play anything on max for dx10 under 500 dollars at full HD resolutions. Processor is least of worries, but amd is lagging bad still.

[CA_Steve]

Stepping away from the gory cpu details, Intel can outspend AMD in R&D whenever and where ever they want. Plus, AMD just doesn't generate enough cash to cover new process development and the capital cost of new fabs.

3Q 07
Intel Revenue $10B (USD), R&D ~$1.4B, Gross Margin 52%
AMD Revenue $1.6B, R&D $467M, GM 41%

It's a bumpy road at best for AMD. If ATI were in a good place with their GPU development, that could help out with cash flow...but, the ATI arm is behind as well.

[hvengel]

One issue that wasn't discussed in that op-ed was the incredible bad timing of the ATi purchase. Just as they needed to focus on getting the new chips out the door, they shot themselves in the foot by adding all the distractions and finance issues that come from a consolidating purchase.

But long term AMD needed something like an ATI. High end graphics chip vendors like ATI and Nvidia have gotten very good at designing highly parallel processors and in the long run both AMD and Intel will be producing processors that need this type of technology. So AMDs intent with buying ATI was focused 5 years out and because ATI was MUCH more advanced in this area than Intel it means that we will see AMD becoming more competitive as we move forward and this technology becomes part of their CPU line.

Also there have already been benefits for people other than ATI investors as there have been tangible improvements for Linux and X11 users of ATI graphics hardware. This includes the public release of detailed hardware specs. for current ATI graphics chips and AMD actively supporting the production of open source 3D drivers for ATI GPUs. These would have never happened if AMD had not purchased ATI.

[smilingcrow]

In future years I’m sure students of various disciplines will study the AMD v Intel battle of the first decade of the [21st] century with some interest.

How did little AMD manage to get so far ahead of Intel in design terms and hold the advantage for so long only to not capitalise when they had the lead? It’s one thing getting in the lead but another thing to win the race. AMD have either lost their nerve or their Mojo.

All amd boards for the past 5 years have been 1/2 the price with 3x the features and connections on them. On die mem controllers make for no difference in performance across chipsets and brands (1-3% at most). I am free to purchase ANY motherboard with whatever connections I want and not worry about retarded p35 or 965 differences.

Some might say that you are free to buy any AM2 motherboard and get equally poor performance from any of them.
But seriously, the comparative reviews that I’ve seen show that the 945G holds its own pretty well against the G965 and I don’t remember the P35 being a big step forward in memory performance. So I think you have blown this out of all proportion especially since Core 2 Duo is far enough ahead of K8 as to make a 5% difference in memory performance insignificant; if it is that much.
I agree that the boards are cheaper and typically seem to offer better configurations than Intel boards. Ironically the cheapness of AMD boards doesn’t help AMD though as it means that the chipsets are being sold for peanuts.

I just hope AMD’s 45nm process turns out as good as their 90nm one. Good luck AMD, you seem to need a bit of that at the moment.

[aristide1]

AMD had has the $65 dollar dual core market. Problem with that is where are the big profits in a $65 dual core? I don't think Intel would even offer a CPU under a $100 if it didn't have to.

[aristide1]

Phenom 2.4 should be 199 dollars and then it would be king though.

The Intel Q6600 would be $199 the very next day.

All amd boards for the past 5 years have been 1/2 the price with 3x the features and connections on them. On die mem controllers make for no difference in performance across chipsets and brands (1-3% at most). I am free to purchase ANY motherboard with whatever connections I want and not worry about retarded p35 or 965 differences. 65 dollar motherboard = 250 dollar motherboard in stock performance numbers for amd.

That's an extreme, but your generally correct, it's more like 65 versus 150.

Really though, at this point, there isnt much of a need for quad core.

Actually quad cores which generally speaking would be of a smaller manufacturing process than whatever they would replace, would be a great way to decreases costs of running servers. On another thread someone commented on running 49 blades. Can you imagine how much less electricity would result from switching 90nm to 45nm? Or perhaps that server is still running one of those 130nm Intel Pentium Ds that have a TDP of 130 watts. Yuck.
But again you're right for the most part, the average guy doesn't need it, and the folding crowd is not average.

[smilingcrow]

Anandtech have a further review of Barcelona versus Clovertown and Harpertown using mainly Server benchmarks and for the first time you can clearly see that AMD’s architecture scales as well as hoped. They just need to iron out the manufacturing problems which seem huge at this point. The Achilles heel for AMD is that power consumption of their 65nm process seems very poor. Good to see some positive AMD data for once.

[adam_mccullough]

Don't know about the blades, but I imagine they're pre "Core" Xeons... yuck indeed! I'd guess the cluster probably runs at about 10 kW at the very least, not including the air conditioning.

My new toy at work is an 8-core 65nm (Clovertown) machine. It lives in a surprisingly cool and small case, considering the processing grunt that's inside... I'm seriously impressed with these new chips; the 45nm ones look very nice.

[djkest]

Intel vs. Amd.... hmmm interesting
2120 is $75, it's 1.6Ghz per core
3800+ x2 is $60, it's 2.0 Ghz per core

For $100:
E2140 1.8GHz per core
4800+ 2.5GHz per core

5000+ @ 2.6 GHz, $109
E6750 @2.66 GHz, $189

I see value. As you get higher up though there isn't really anything from AMD to compete with the fastest Intel CPUs. And while the intel CPUs may be faster even at the same clock speed, the average productivity user wont' notice a whole lot of difference.

Interesting that although people think AMD might be dead, they actually did well for themselves in the third quarter.
http://www.theinquirer.net/gb/inquirer/ ... -price-war

[Plissken]

while the intel CPUs may be faster even at the same clock speed, the average productivity user wont' notice a whole lot of difference.

True, but the average enthusiast knows (and sees) the difference, and enthusiasts drive the leading edge of the market. They did it for AMD, and now they do it for Intel. AMD has a tough road to hoe. If AMD continues this pattern they will establish themselves as the "budget CPU choice"... not a good situation for them in the long term.

[smilingcrow]

while the intel CPUs may be faster even at the same clock speed, the average productivity user wont' notice a whole lot of difference.

True, but the average enthusiast knows (and sees) the difference, and enthusiasts drive the leading edge of the market. They did it for AMD, and now they do it for Intel. AMD has a tough road to hoe. If AMD continues this pattern they will establish themselves as the "budget CPU choice"... not a good situation for them in the long term.

One of the problems that I see is not so much that AMD are now the budget CPU choice but that Intel is in a position to dictate pricing for the whole market and they decided to push pricing to a low level by historical standards. If Intel hadn’t done this AMD would have had much better margins. There’s nothing wrong with being the leader in the budget end of the market provided the competition doesn’t squeeze you too much.
If Intel becomes even more aggressive with Penryn pricing then that would be a big problem. Fortunately it doesn’t seem to be going to do so.

The performance of K10 doesn’t look so bad in the Server market except that the power consumption of their 65nm process looks poor. If their 45nm process is also poor and/or late and Nehalem arrives on time and with good performance then AMD may suddenly find themselves in 2nd place by a large margin in the Server market which would just put more pressure on them. Working at AMD must be quite scary at the moment as they can’t afford to make many more slip ups if Intel continue to hit the ball out of the park.

[seraphyn]

http://www.techreport.com/discussions.x/13721

Seems that AMD's problems with the TLD bug is keeping them from entering the server market with their new high end quad core Barcelona.

A few quotes for those that don't want to read the whole article:

the TLB erratum has led to a "stop ship" order on all Barcelona Opterons.

News of this problem is notable because it confirms that the TLB erratum affects Barcelona server processors as well as Phenom desktop CPUs, and that the problem impacts AMD's quad-core processors at lower clock speeds.

AMD has said a revision B3 is in the works and expected in Q1. One source told TR that large quantities of B3 chips might not be available until the end of Q1.

So presumably, the new B3 stepping fixes the bugs, but again sets back the big release onto the market. Even though the problems occur very rarely, for the server market this is of course unacceptable.
I would assume that for us normal consumers the stock of the B2 revision Phenoms also won't be so plentifull, if you can get one now, I'd suggest waiting for the B3.

A lot of doomsaying on AMD resulted from my favorite Dutch tech site, but i doubt this will really end AMD. They still have funds from selling a productionplant and an infusion from an investor, on top of that this is only the high end market we're talking about. AMD still performs fine in the low/mid segment as far as i'm aware.

[QuietOC]

The main reason for the performance of Conroe is not the wide execution power, but the low latency access to memory, due to the large cache and smart hardware prefetching.

Large fast L2 caches have been an Intel strength. The 1MB cache e21x0 Pentiums Dual Cores are surprising slow.

AMD didn't make a mistake going AM2, they had to go DDR2, for the sake of their future quad-core - that got delayed.
So the move to AM2 was just hasty and they sure made a mistake dropping 939 so quickly.

Hindsight shows that dual-channel memory was needless for AMD and Intel. In that sense S939 was really a mistake. It seemed like AMD rushed it to market because of mistaken demand for a dual-channel DDR desktop platform.

Intel's current FSB was really designed for RAMBUS, and of course required dual-channel DDR (Granite Bay/E7205) to really perform when PC-3200 was the fastest memory around. Dual-channel DDR2 doesn't really make sense for either platform, and now they will try to push dual-channel DDR3.

As far as AMD S754 was fine, and S939 was pretty pointless, as is AM2 and the now dropped 4X4. What AMD could have done instead of S939/AM2 is make simple single-channel K8 cores with multiple sockets in mind (multiple HT links). That could have been done cheaply. The 4X4 platform was way too complicated. But who would need quad-cores chips if dual or quad socket motherboards were simple and relatively cheap?

A single socket, single-channel DDR2 AMD platform would be very nice for an inexpensive, low-power desktop, but AMD made it mobile only (S1).

[Techno Pride]

the mere fact that there're very few Phenom threads on SPCR after the official launch of Phenom is disturbing.

Never mind the fact that Phenom doesn't work on 99% of AM2 motherboards out there, there hasnt been any Phenom undervolting posts yet!

No one seems to have any idea whether the Phenom works in casings with "crappy" airflow, which is like 99% of the casings here.

[smilingcrow]

Large fast L2 caches have been an Intel strength. The 1MB cache e21x0 Pentiums Dual Cores are surprising slow.

The review that I’ve seen that compared the three cache sizes didn’t generally tally with your observation. I was surprised how well the 1MB cache chips did except in certain games where the performance drop off was significant. If you don’t play games or run Scientific Applications I personally wouldn’t worry about the cache size. The 4MB cache versions very rarely offer a gain over the 2MB cache versions and just aren’t worth it IMO. If you aren’t over-clocking and require high clock speeds you’ll have to go with 4MB anyway.

Your comment regarding single v dual channel memory on AMD systems surprises me as my understanding is that AMD systems benefits from the higher bandwidth due to their caches being smaller and the architecture being less refined than Intels.
Benchmarks typically show AMD systems performing better with faster RAM which suggests to me that if you halved the memory bandwidth performance would suffer.

the mere fact that there're very few Phenom threads on SPCR after the official launch of Phenom is disturbing.
Never mind the fact that Phenom doesn't work on 99% of AM2 motherboards out there, there hasnt been any Phenom undervolting posts yet!
No one seems to have any idea whether the Phenom works in casings with "crappy" airflow, which is like 99% of the casings here.

It’s not really surprising that there is little info on Phenom at SPCR with availability being low and BIOS support being slow to materialise. Their high power consumption also makes them seem about as useful as a P4 Prescott for the SPCR crowd.
Until they release the B3 stepping of the chip I can’t say I have much interest anyway in what the broken ones can manage.

[QuietOC]

Your comment regarding single v dual channel memory on AMD systems surprises me as my understanding is that AMD systems benefits from the higher bandwidth due to their caches being smaller and the architecture being less refined than Intels.
Benchmarks typically show AMD systems performing better with faster RAM which suggests to me that if you halved the memory bandwidth performance would suffer.

I would say that system memory on AMD seems to act more like higher level cache on the Intel bus. Memory performance of Intel's chipsets hasn't been the best since the SDRAM 440BX. I have to think its the heritage of the P4 bus which was originally targeted (with the i840) at RAMBUS.

I've been playing with the CPU test from the Crysis demo and have been surprised at how dependant it seems on memory bandwidth (and suppossedly doesn't care about memory size).

I have a nVidia GeForce 7100 motherboard coming for my E2140, so I should get to try out how well a high-clockspeed (>3.2GHz) , small cache Core 2 does with single channel DDR2. According to this the MCP73 might be the best economy chipset for low multiplier Core 2s. My experience with the ASUS G33 motherboard wasn't so positive.