AMD doing things right.

[DAPUNISHER]

"One For You, One For Me"

For the first time, the AMD financials this quarter showed four units. The first three are the Computation Products; the Memory Products; and the Personal Connectivity segments. The first made $209 million, the other two lost $50 and $14 million units.

Then we have "All Other." All Other had a terrible quarter. They lost $66 million on just $2 million in revenue.

Needless to say, somebody asked about this new unit.

Well, it turns out that this appears to be the Santa Claus segment, aka "pay bonuses to AMD employees" unit. It's not yet clear if this is essentially all this "segment" does, but if it is, this could be highly, highly disturbing.

If this is the case, and you take this unit. AMD wouldn't have made $76 million, they would have made $142 million, or almost double the profit.

Put another way, this quarter, the AMD execs essentially put the money they made into two big piles, one for them, one for the shareholders, and counted it out saying, "One for me, one for you."

If that's accurate, that seems a bit greedy. It's about four times the amount shown for (adjusted) previous quarters.

It may well not be (and we're going to look deeper into this, and compare whatever we find to what Intel does, and yes, Intel puts its bonuses into an "All Other" category, too, though it also contains working units). It might be a one quarter fluke representing a payoff after a few years of slim pickings, but the conference call seemed to suggest that this was a figure that would remain substantial so long as the good times kept rolling.

For the moment, though, it looks pretty bad. It was noticed by the media, and it probably didn't help the 10%+ drop in AMD's stock price yesterday (though it probably didn't cause it).

More on this probably next week

I'll wait till all the facts are in before calling the above the truth. Possibly AMD's all others contains working units as well?

BTW, I am interested in wether they will indeed become an outsouce for other companies chipsets, ATI and/or Nvidia perhaps?

 

[dmens]

AMD's financial health *should* be a lot better given the mistakes their competition made over the past couple years. By the way, adding stuff onto a CPU isn't that hard. Making it go fast enough within the thermal specification is hard. Both Intel and AMD is capable of designing absurdly fast processors without a TDP. I recall seeing a 2001 proposal after WMT... 10-issue 16-way SMT core at 250W. Yikes!

 

[nealh]

Intel's track record of perfromance as a company is very impressive...do not underestimate their R&D...they are behemoth that is still in bad cpu cycle making a tidy profit

I am not sure why they are not more proactive in releasing a more competitive processor to the A64/Opteron single/dual cores

I agree given Intel's missteps/mishaps...I would hope AMD would grew and take advantage of this opportunity and I only hope the bonus were wisely paid...because AMD's financials are not as strong as Intel...most businesses(sp?) will cycle to some degree..this looks to be the first big Intel mishap I can recall(but I do not know their enite history)

AMD does not have money and R&D to fall back on as Intel...in the big scheme

Competition is good but if Intel does release anything close or faster than AMD this will be a step back for AMD as Intel is a household brand for the novice lay public...AMD is very slowly getting branding.....

 

[Furen]

There's a difference between technology and products. AMD has shared interest in x86 (as decided by the courts) so, thus, it can use x86 as its own. This means that anything that affects x86 DIRECTLY is supposed to be shared by both companies (this is why AMD can use SSE and Intel can use AMD64). This does not mean that the microarchitectures are shared, since neither Intel's nor AMD's microarchitectures actually execute x86 code directly. AMD actually used to clone Intel designs directly but now HAS to create its own CPUs, as per court order (the Athlon was the first microarchitecture that AMD built from scratch). Also, the exact structure of most (if not all) Intel CPUs is trade secret and if AMD happened to get a copy of the relevant documents this would be misappropiation of trade secrets, which is a pretty harsh offense. So no, Intel cannot make A64s nor can AMD make Pentium Ms.

Another thing: adding a memory controller is not "hard" at all. All you need to do is throw the northbridge into the CPU die and modify it so it can run at CPU clock properly. It IS a major redesign, however, and it requires a significant investment. Also, a new CPU interface to the motherboard has to be designed, since the "front-side" bus is now integrated into the CPU core and, thus, runs at CPU clock. It also requires a major investment in designing a new CPU package and, of course, more pins (even if LGA775 has enough unused pins for this you wouldnt want CPUs with an integrated mem controller to use the same socket, or something too similar).

EDIT:
Zebo: you claim that AMD is still on the K7? Then Intel is still on the P6 and will be for the forseable future.

Both companies plan to stick to a solid microarchitecture and just tweak the hell out of it, which is how CPUs are normally built. You dont redesign your whole microarchitecture just so people see that you're doing something. You tweak and tweak and only redesign once tweaks stop giving any returns. Meron's Pipeline will be 4 stages longer the the Pentium M's (if the PentiumM's is 10 stages) I believe, which should allow Intel to hit similar clocks to current A64s (maybe a bit higher since the A64 has 2 stages less, but SOI fabrication helps with yields so I'd guess the same). So a longer pipeline will make it a bit less efficient (that's what pipeline stages do), a higher FSB should make it do a bit better (if the P6 actually benefits from this, which is not a given), and there should be some microarchitectural tweaks to increase performance per clock as well. Overall, I'd expect it to perform pretty much on par with current Pentium Ms (clock for clock) but at a higher clock speed.

Now let's look at why Intel DOES NOT integrate the memory controller:

1) The huge caches that Intel has its sights on actually lessen the impact of an external memory controller.

2) You lose a LOT of flexibility by integrating it, as any AMD user knows.

3) Questionable performance benefits because of the already big caches.

4) Quad-pumped FSB gives more than enough bandwidth to the CPU.

5) Prefetching technologies also lessen the impact of an external memory controller.

 

[dmens]

As far as tweaks go, merom is a pretty significant architectural tweak. The current P-M's are crippled for power reasons... I guess merom removes those limiters, so to speak. Also, circuit methodology changes can yield higher frequencies on the same uarch.

 

[Furen]

We're kind of hijacking the thread, but whatever. The biggest tweak on Meron are the 4-issue execution width, but it's hard to assess the performance benefit that will arise from this. The pipeline is also longer, which could be good or bad, depending on how you look at it. The rest of the tweaks are just what you'd expect, bigger caches (4MB shared), higher FSB speed, Macro-op fusion (finally)...

I wouldn't call the Pentium M crippled, but rather very optimized. There are only two things in which its significantly inferior to a Pentium 4: SSE performance and FSB bandwidth (which, in turn, bottlenecks the memory bandwidth). SSE performance should increase significantly, which WILL increase performance. Bandwidth, however, doesnt look like it will help much. Overclocking a Pentium M right now makes it a bit better than an A64 clock for clock (which is kind of expected considering that it has a shorter pipeline) and also removes the bandwidth penalty (I'm guessing you've all seen Pentium Ms 1.6s @ 2.6GHz with a 1GHz+ FSB ) and gives it access to low-latency DDR1.

 

[dmens]

4-issue is the biggest change on the frontend and rename. The execution stacks on merom have been significantly altered, and there were a lot of changes to the memory interface besides the bigger caches... imo the SSE performance on the current P-M is not memory limited but design limited on purpose.

Macro-op fusion is not worth the ROI in my opinion... the complexity added to much of the out-of-order logic does not justify the performance gain. Then again, I might say the same thing for SMT, hehe.

 

[Fox5]

Originally posted by: Zebo

Originally posted by: Aenslead
Ooooh!! yeah... WHEN Conroe comes.

And Conroe will perfrom like this, and it will do that, too!
But... strangely... people seem to think AMD is doing nothing? Do you think AMD will have no tangible response to Intel's next gen processors?

That sounds like:
"Oh.... you naughty AMD Athlon! just you wait until Pentium 4 comes! you'll see!"

I agree. Silly as can be.

Share price has risen because profits and increased market shared are already predicted. Speculation on futures? Maybe... but based on facts and hard, cold numbers.

They arnt doing anything. For all practical purposes they are on K7 still and I havre heard nothing about a new CPU design. And if intel gets serious, slaps a mem controller on thier p6 (aka conroe/PentiumM) it's all over! Did everyone miss the crippled P-M benches against a full fledged FX? Pretty close of you ask me. Add in real memory, and not worried about power envelope so much, and whatever other tricks they have from now till then Intel can really work this chip. Then like Dems just said, even is results are similar, Intels household trusted name will carry the day at all levels. Not to mention they offer a complete platform CPU + chipset which is very important to many..rather than AMD relying on any hack who comes along.

Even as big of a mistake as p4 was , I mean reviewers literally laughing at it, Intel still gained market share.. what you think will happen when they release a chip that runs faster, cooler, less power?

AMD's response to Conroe....
Slight increase in mhz per mhz performance...
The move to quad core...
Integration of a full northbridge onto the cpu, not just the memory controller. AMD's plans for 2006/2007 call for a cpu with integrated graphics, pci express bus, imc, and possibly more.

BTW, if you say AMD is still on the original Athlon, then Intel is still on the Pentium Pro. AMD has made no major changes to their core, but niether has Intel, even the P4 and P-M still retain the same P-P core.
And an IMC is not an easy add on, Intel has yet to do it, and AMD was a year or two late with it. Barton should have never existed, Athlon 64 was supposed to be out by then.

It's my understanding AMD and Intel have Tech share agreements so if AMD wanted to make netburst they could and if Intel wanted to make A64's they could.

They don't share quite that much, they share things like instruction sets (x86-64 and SSE), protocols (hypertransport), and maybe ideas (out of order execution), but they can't just go and take each other's cores and use them. Their tech share agreement basically just protects them from being sued for infringing on ideas, the implementation has to be different though.

I am not sure why they are not more proactive in releasing a more competitive processor to the A64/Opteron single/dual cores

Because P4 is still their most competitive performance processor in high end markets (minus gaming), and it's not an overnight switch to change all their fabs to producing p-m motherboards and cpus.

as per court order (the Athlon was the first microarchitecture that AMD built from scratch).

True and false, AMD's first non-clone architecture was the K6, but that design was acquired. I wonder how much of the Athlon is based on the K6 though.

 

[Viditor]

Originally posted by: dmens
AMD's financial health *should* be a lot better given the mistakes their competition made over the past couple years.

AMD's financial health WOULD be better if they had been allowed free access to the market...Your statement is actually the same one AMD is making to the courts right now.

 

[Viditor]

Originally posted by: Furen
There's a difference between technology and products. AMD has shared interest in x86 (as decided by the courts) so, thus, it can use x86 as its own. This means that anything that affects x86 DIRECTLY is supposed to be shared by both companies (this is why AMD can use SSE and Intel can use AMD64). This does not mean that the microarchitectures are shared, since neither Intel's nor AMD's microarchitectures actually execute x86 code directly. AMD actually used to clone Intel designs directly but now HAS to create its own CPUs, as per court order (the Athlon was the first microarchitecture that AMD built from scratch). Also, the exact structure of most (if not all) Intel CPUs is trade secret and if AMD happened to get a copy of the relevant documents this would be misappropiation of trade secrets, which is a pretty harsh offense. So no, Intel cannot make A64s nor can AMD make Pentium Ms.

Another thing: adding a memory controller is not "hard" at all. All you need to do is throw the northbridge into the CPU die and modify it so it can run at CPU clock properly. It IS a major redesign, however, and it requires a significant investment. Also, a new CPU interface to the motherboard has to be designed, since the "front-side" bus is now integrated into the CPU core and, thus, runs at CPU clock. It also requires a major investment in designing a new CPU package and, of course, more pins (even if LGA775 has enough unused pins for this you wouldnt want CPUs with an integrated mem controller to use the same socket, or something too similar).

EDIT:
Zebo: you claim that AMD is still on the K7? Then Intel is still on the P6 and will be for the forseable future.

Both companies plan to stick to a solid microarchitecture and just tweak the hell out of it, which is how CPUs are normally built. You dont redesign your whole microarchitecture just so people see that you're doing something. You tweak and tweak and only redesign once tweaks stop giving any returns. Meron's Pipeline will be 4 stages longer the the Pentium M's (if the PentiumM's is 10 stages) I believe, which should allow Intel to hit similar clocks to current A64s (maybe a bit higher since the A64 has 2 stages less, but SOI fabrication helps with yields so I'd guess the same). So a longer pipeline will make it a bit less efficient (that's what pipeline stages do), a higher FSB should make it do a bit better (if the P6 actually benefits from this, which is not a given), and there should be some microarchitectural tweaks to increase performance per clock as well. Overall, I'd expect it to perform pretty much on par with current Pentium Ms (clock for clock) but at a higher clock speed.

Now let's look at why Intel DOES NOT integrate the memory controller:

1) The huge caches that Intel has its sights on actually lessen the impact of an external memory controller.

2) You lose a LOT of flexibility by integrating it, as any AMD user knows.

3) Questionable performance benefits because of the already big caches.

4) Quad-pumped FSB gives more than enough bandwidth to the CPU.

5) Prefetching technologies also lessen the impact of an external memory controller.

Some good points Furen...but by "hard", my meaning was "takes a great deal of time".
A major redesign isn't something you can whip up in a year or two...

 

[dmens]

Originally posted by: Viditor
AMD's financial health WOULD be better if they had been allowed free access to the market...Your statement is actually the same one AMD is making to the courts right now.

LOL, here we go again. I don't feel like beating this dead horse any more, but the whole assumption that AMD can do no wrong and all their woes must be caused by intel is laughable.

 

[Markfw]

Originally posted by: dmens

Originally posted by: Viditor
AMD's financial health WOULD be better if they had been allowed free access to the market...Your statement is actually the same one AMD is making to the courts right now.

LOL, here we go again. I don't feel like beating this dead horse any more, but the whole assumption that AMD can do no wrong and all their woes must be caused by intel is laughable.

We will see who is laughing when the courts get done.