AMD + ATi Doomed to fail ; no chance

[LordPope]

http://www.theinquirer.net/default.aspx?article=33524

ATI is a duff purchase for AMD

Comment The opposite of synergy


By Arron Rouse: Monday 07 August 2006, 16:05

ANALYSTS AT INTEL must be laughing their socks off. Once the smoke clears and the mirrors have been taken down, AMD will discover that it has purchased rather less than it might have hoped in acquiring ATI. In fact, it could turn out to be a bad move for the industry as a whole.
The problem is not so much what ATI has but what it doesn't have. The graphics industry is driven largely by the fanboys and they are driven largely by getting extra frames per second. So far ATI seems to be onto a winner: put two X1900XTXs into a Crossfire capable machine and you'll get frame rates to make you proud. But there's a word in that last sentence that should make you stop and think. It should have made AMD stop and think. It's Crossfire. And nobody seems to have noticed that Crossfire is dead.

The Way of the Dodo
Intel has been plugging Crossfire for use with the new Core 2 Duo but that's ending quickly now that AMD owns the technology. It was bound to go that way and AMD must have known it. So scratch Intel as a long-term platform for Crossfire.

And AMD can't use Crossfire either. For a start, despite lots of promises, plenty of column inches and many product announcements, try to buy a Crossfire motherboard for an AM2 socket processor and see how far you get. They're incredibly rare. There's no sign of products from Asus, Gigabyte, MSI, DFI or any of the other big names apart from a few press releases. In fact, the only one we could find for sale was an ECS that was out of stock. This is two and half months after the launch of AM2.

We tried asking ATI about this lack of Crossfire for AM2 some time ago. Its PR people said they'd find some on sale for us to see and were never heard from again.

Even if motherboard manufacturers were willing to suddenly switch on large streams of ATI chipset-based products, AMD won't let them now. After all, that would be too much like treading on the toes of its existing chipset partners, Nvidia, Via and SIS. AMD really wouldn't want Nvidia or Via pulling out as it just couldn't replace them quickly enough.

That leaves Crossfire gone. That means no more ultra-high frame rates with X1900XTXs in Crossfire. Which means the fanboys will start defecting to Nvidia's SLI. Fewer fanboys means fewer column inches means fewer sales. And that means ATI is worth less.

Frisson Chips
AMD is in the awkward position of having to dump ATI's chipset business. Not that there was much to it apart from a nice deal with Intel which is probably dead now. And AMD won't want to compete with Nvidia, Via and SIS so it can't even keep producing chipsets for its own processors.

That leaves laptop graphics, mobile phone graphics and set-top box chips as the only other areas worth something. That's a reasonable area of the business but it's hardly going to keep the stockholders' dividends pouring in.

One Way
So far, the purchase looks like a one way spiral downwards to loss of profitability. The chipset business has to go. That means Crossfire is history. That means a trailing off of sales on graphics processors as Nvidia permanently secures the performance crown. In short, the whole thing looks like a very expensive way of downsizing the entire ATI workforce.

AMD has said that it wants to start building graphics technology into its processors but there's a fair chance of that being a dead end. Sony already tried to do that with Cell and discovered just how fast the graphics market moves in comparison with processors. It ended up having to put a separate graphics chip into the PS3.

There's only one way that springs to mind as a potential escape route towards profit: releasing graphics chips that fit into the same sockets as AMD's processors and use HyperTransport instead of PCIe. It would mean convincing memory manufacturers to release GDDR4 memory modules, would lead to huge motherboards (barring some radical form factor changes), incredible cooling requirements, etc. But it would be enormously risky as, once again, it could incur the wrath of Nvidia.

Where does this leave AMD? There's a fair chance that, once the smoke clears and the mirrors have been taken down, it has spent $5.4 on a company that would have carried on its profitable dual for the top of the graphics market for years to come but will now gently evaporate into nothing. µ

 

[Gary_Busey]

Does The Inquirer just pay random homeless people off the street to come in and write articles for them?

 

[LordPope]

i don't understand anything he wrote.....

 

[Viperabyss]

i highly doubt it.

first of all, well, its theinquirer, do i need to say more?

secondly, i'm actually seeing a market for AMD's torrenza project. the feature of simply dropping a co-processor in the system to suit your need is just brilliant.

where does ATi fit into this? i believe it can fit both graphical solutions and physics solutions onto AMD's boards. thus eliminating the need for another boards and slots.

if ATi can design both graphic cards and physics card to share system's memory, IMO, that would be grand.

 

[phreejak]

too funny..... made me spit my coke through my nose

 

[shinigamiX]

I thought ATI and Nvidia were both upgrading to 80nm soon...

ATI and AMD will work out IMO, The Inquirer goes a bit overboard with its predictions. Crossfire is fine, and no doubt before the merger is finalised more hardware will be brought to the market.

 

[Viperabyss]

Well first of all with quad core single slot PCI-e cards already on the market, SLI could be dead too.
Then again, PCI-e could be dead too. AMD may use their current 90nm fabs to produce 9nm gpu's that fit into processor sockets. Who knows, your next video card may look like a cpu with a fan heatsink on it just like your processor.
Do you really think AMD would put everything it had into a company if it didnt have a plan?
How bout a real 4x4, 2 gpu sockets and 2 cpu sockets. You can use 1 or 2 of each??!! 8O :lol:

i believe 4x4 is just a temporary solution, to counter Core 2s, and its not succeeding.

well of course 4x4 gets an 80% boost in performance, but considering the cost, its not entirely cost effective.

that aside, it was reported that 4x4 was benchmarked with non-gaming apps. while getting an 80% boost in performance does sound nice, but 4x4 was originally designed to be a gaming platform, instead of a workstation platform.

however, if you're talking about torrenza, its a whole different story. you can say that AMD releases 4x4 to gain experience in designing multi-socket computing solutions.

 

[phreejak]

Is the Inquirer even aware of the reputation it has amongst enthusiasts?

 

[Mind_Rebuilding]

Is the Inquirer even aware of the reputation it has amongst enthusiasts?

The Inquirer is now just for gossips only. :lol:

 

[phreejak]

It is just incredible to me how far off they are sometimes and even how wrong they've been and yet I keep seing people quote them when it suits their "fanboyism". Surely someone there must know that their material is basically only good for something that a cat can read when it's taking a dump.

 

[chunkymonster]

That article is a load of crap...

The graphics industry is driven largely by the fanboys...

That's an instant classic!

 

[russki]

That's about the dumbest article on the subject. And that goes even with consideration of the stuff on these forums, not even press. The Inquirer is useless, indeed.

Chipset business was the primary driver behind the merger, and they casually dismiss it in two sentenses...

 

[LordPope]

yeah the chipset part....with Amd TRYING TO GO ONE STOP SHOP platform..... why would just get rid of Ati's chipset division


like i said this article is one for the ages..... but i am sure INTEL fanboys wont argue with it too much....

 

[VillageIdiot]

I thought that the Inquierer was pro-AMD...

 

[Caboose-1]

That article is a load of crap...

The graphics industry is driven largely by the fanboys...

That's an instant classic!Could have swore Intel was the graphics industry.

 

[Grimmy]

Inquirer+TomsForumz=headache.

I need more coffee.

<(so.. how many lumps you want with that coffee doc?)

I'll make it myself... you, you.. screwy rabbit.

 

[Anoobis]

Well first of all with quad core single slot PCI-e cards already on the market, SLI could be dead too.
Then again, PCI-e could be dead too. AMD may use their current 90nm fabs to produce 9nm gpu's that fit into processor sockets. Who knows, your next video card may look like a cpu with a fan heatsink on it just like your processor.
Do you really think AMD would put everything it had into a company if it didnt have a plan?
How bout a real 4x4, 2 gpu sockets and 2 cpu sockets. You can use 1 or 2 of each??!! 8O :lol:

i believe 4x4 is just a temporary solution, to counter Core 2s, and its not succeeding.

well of course 4x4 gets an 80% boost in performance, but considering the cost, its not entirely cost effective.

that aside, it was reported that 4x4 was benchmarked with non-gaming apps. while getting an 80% boost in performance does sound nice, but 4x4 was originally designed to be a gaming platform, instead of a workstation platform.

however, if you're talking about torrenza, its a whole different story. you can say that AMD releases 4x4 to gain experience in designing multi-socket computing solutions.
IIRC 4x4 is supposed to only work with the FX series of Athlon processors and I could be wrong on that. If I'm right then 4x4 isn't worth it. However 4x4 would be viable if they allow you to use CPUs other than the FX series of chips. At the current prices, you can pick up 2 3800 X2s on Newegg for a bit over $300.00. It would be interesting to see how well they'd perform. Of course so much for power efficiency.

 

[darkstar782]

Personally I think the additional HT links for 'coprocessors' etc will not be an AM2 or similar socket, but a HTX slot.

AMD will just use HTX to compete with PCI-e, which is after all an intel standard.

HTX has some advatages with 'bursty' data, smaller packet header sizes etc, but for log sustained transfers on a point-to-point bus I think PCI-E is a better choice tbh.

The fact that HT links are supported by the CPU is not really an advantage of HTX itself, but rather AMDs CPU design, they could after all have integrated a PCI-E controller instead if they had liked!

 

[exit2dos]

Does The Inquirer just pay random homeless people off the street to come in and write articles for them?

Hey, they are hiring!!!
Wanted: Ace reviewer for the INQ

I guess the random homeless have too much dignity, so they're looking elsewhere.

 

[MarkyParky]

OK, first of all, I'm new to the forums, so hi...

Next, this intrigued me...

How bout a real 4x4, 2 quad gpu sockets and 2 quad cpu sockets. You can use 1 or 2 cpu's or gpu's??!! 8O :lol:

I have been thinking about this for a while, and while I recognise that I am by no means close to being as educated as some of you guys out there I've read (I think) articles based on AMD leasing out their HT technology, so that other devices can communicate on the bus directly. Surely if this is the case it opens up massive possibilities (especially now AMD own ATI - in theory, if not yet in practice) for the GPU to be connected to the system via the hypertransport bus.

Is this possible? am I talking rubbish? I'm not sure, it's just the ramblings of a wannabe expert.

Basically my thinking is that if AMD can do this, then they can use their multisocket idea, and do what "MrsBytch" suggested and have a combination of GPUs and CPUs to give incredible performance, considering the latencies and bandwidth of the HT bus compared with PCIE (again, correct me if I'm wrong)

**waiting to be educated/corrected/all of the above**

Cheers

Mark

 

[cyborg_ninja-117]

the inquirer fails

 

[TheMaster]

Well the inquirer says if that ever happened ATi and AMD would incur the "wrath of Nvidia" :roll: whatever that means.

 

[exit2dos]

OK, first of all, I'm new to the forums, so hi...

Next, this intrigued me...

How bout a real 4x4, 2 quad gpu sockets and 2 quad cpu sockets. You can use 1 or 2 cpu's or gpu's??!! 8O :lol:

I have been thinking about this for a while, and while I recognise that I am by no means close to being as educated as some of you guys out there I've read (I think) articles based on AMD leasing out their HT technology, so that other devices can communicate on the bus directly. Surely if this is the case it opens up massive possibilities (especially now AMD own ATI - in theory, if not yet in practice) for the GPU to be connected to the system via the hypertransport bus.

Is this possible? am I talking rubbish? I'm not sure, it's just the ramblings of a wannabe expert.

Basically my thinking is that if AMD can do this, then they can use their multisocket idea, and do what "MrsBytch" suggested and have a combination of GPUs and CPUs to give incredible performance, considering the latencies and bandwidth of the HT bus compared with PCIE (again, correct me if I'm wrong)

**waiting to be educated/corrected/all of the above**

Cheers

Mark

In effect this is true, but several other factors come into play:
1) GPUs use much faster and more expensive RAM. A GPU socket would need its own RAM and the GPU would need its own Integrated Memory Controller (IMC). While a low/mid range GPU could get by, performance GPUs would still be card based.

2) As far as other companies developing a chip for a HT-linked socket, AMD would need a much higher penetration rate. A developer isn't going to put resources into providing a chip for a motherboard very few own.

Consider: If AMD had 50% of the destop market, how many HT socket motherboards would be out there? Cheaper/Entry level computers (which make up the bulk of destop sales) wouldn't, as the manufacturers wouldn't want the socket or the mobo trace expense. Office workstations wouldn't. And alot of the mid-range computer owners don't upgrade.
This just leaves the Enthusiast and some of the mid-range, which I would estimate being about 20% (at the most) of AMD's desktop sales.

If your market is only 20% of a company that does 50% of the business, then you are only targeting 10% of the market as a whole. Now, factor in that AMD only has 20% (not even this in reality) of the desktop market - then your total market is 20% of 20% which is only 4% of the market.

 

[The_Abyss]

The article also conveniently forgets that the full support for unified shaders on the R600 will pretty much negate the need for Crossfire and offer higher performance in existing DX9 games than Crossfire ever could. DX10 is still to be tested of course, but the principle is even more sound there to increase performance further.

 

[exit2dos]

The article also conveniently forgets that the full support for unified shaders on the R600 will pretty much negate the need for Crossfire and offer higher performance in existing DX9 games than Crossfire ever could. DX10 is still to be tested of course, but the principle is even more sound there to increase performance further.

Don't forget the PCI-e 2.0 spec.
PCIe gets a version 2.0

5Gbps!!!