AMD-ATI Conference Call Transcript [LIVE]

Its certainly coming, it doesnt matter if people want to think its not. It'll be in the low end first though and probably for a long time. That's where its advantages will come into play.

 

This is a sad day for 3d graphics. AMD bought ATI for many reasons...top performance 3d graphics cards was probably less than 20% of the reason they bought them for. So research into cutting edge 3d technology is going to get de-emphasized by AMD execs. This will slow down competition. If ATI eases up Nvidia will ease up. Progress will slow.

 

It may also move into the mid range on multi-socket systems if we get enough on-die or on-package memory to make the performance worthwhile. That'll be some years yet, though.

 

Thread-level-parallelism? You may want to have a look at some of the more exotic CPUs out there now, and where things may be going more generally glancing at Intel's roadmap..

Not necessarily. Couldn't you have two 'CPUs' where you now have one CPU and one GPU? Each with their own pool of memory and a high speed interconnect between them?

 

I already mentioned chips like Niagara (or Vega), but Intel ain't going there for a long time, because most workloads aren't TLP friendly today with the exception of some server stuff.

And how you are going to connect these two separate memory pools to the die? 256-bit memory bus is already straining chip packaging to the limit, and you want to add another 64-256bit CPU memory pool as well?

As for SIMT, I haven't heard this terminology before. Presumably, it's just a specialization of TLP wherein each thread has its own context of registers, et al, but share a single program counter?

 

Sure ... if you listened to Intel prior to voodoo the cpu would do everything and cut your lunch. However the exact opposite is true now though more and more functions have gone from discrete units too Integrated on mobo to integrated in the South Bridge. There's certainly a trend towards integration or accumulation of functions though I find it hard to believe I will be buying a cpu that does all the gpu work as well and this merger was the cause of that and Ati's future options in this area suck the integrated arse. It's easier to believe I will be able to buy a relatively high end processor and select the GPU option that goes into it - wether it be a descrete gfx card or another socketed type of gfx product. Maybe Joe Sixpack and business will buy it but they basically are buying the forerunner of that system right now. I can afford the extra cooling, power case space etc for a gaming rig. I am not uber highend - far from it ... further from the best integrated pc now however. I doubt that is going to change.

Unless there is a massive upheaval in the graphics world coming as indicated by other prescient threads here and this merger is the strategic maneuvering prior to the new age dawning in which case it was inevitable anyway... my point being I doubt this merger just pulled the trigger on discrete gfx.

 

Well, no, but GPU's integrated into CPU's seem a near certainty on the AMD side now, for their low-end products.

 

I like this. Just read it.
CPU integrated GPU - south bridge = Low end + business machines.
CPU - cHT socket style Gfx - South Bridge = sounds like the mid-range.

High-end gaming = similar configuartaion to what we have today as it most likely will be in the workstation market.
Nothing new there just a summation.

 

Yup, that's about what I'm expecting, and I'd be willing to bet that nVidia is going to be fighting very hard for marketshare in the mid-range socket market, if that market comes to fruition.

 

Previously I thought the socket concept would not fly or be niche at best. Now, if the acquisition goes through, it's a certainty. I am not clear as too whether or not the cHT socket describes a form factor for the physical socket implementation?? If it is standard, or a standard number of "lanes" becomes the norm then there's is plenty of room for all current chipset players to supply chipset/gfx combo's, mobo makers to add variety + spice and for end users to have the choice of mixing and matching components.

 

DailyTech, specifically Kristopher Kubicki, was kind enough to clean up their piece when the misattribution was brought to their attention. So good on them.

http://www.dailytech.com/article.aspx?newsid=3471

 

Here's an interesting article with comments from Patrick Moorehead of AMD:

The AMD-ATI Acquisition: Integration and Freedom for Customers, IHVs

Some things that stuck out at me:

Huh?

Further..

Also, just so you know who's steering this ship:

 

Are there any other equals when Jen-Hsun is in the room?

Interesting stuff, tho, particularly the "information firewall".

It's nice they are making the effort, and probably it will perserve them some relationships. Doesn't change the fact that some of those other players will be reducing their overall footprint in the AMD world. I wonder if this might be enough in the not-so-long-run to knock Via out of chipsets completely.

 

Wow, I'm infected with the Nvidia conspiracy bug. The first thing that popped into my mind after Titanio's post was that this was a massive Nvidia architected conspiracy to remove their high-end competition in the GPU space while AMD gains ATi's expertise in chipsets/integrated. Though I find it very hard to believe that AMD considers Nvidia to be their equals. But it is interesting that they are openly expressing their loyalty to ATi's biggest competitor - their dance partner so to speak.

 

Well, if I really wanted to be cynical I could note that whether they need NV all that much or not down the road, they certainly do still need them for the next several months to maybe a year! And this deal hasn't closed yet, either. How do you suppose that would help the relationship if AMD was doing the "In yo' face, NV!" dance all over the place and then the damn thing fell thru?

Edit: Said another way, AMD has absolutely no self-interest right now in annoying anyone (other than Intel, of course!) more than the inherent logic and implications of the combination already does. They don't have to close interfaces or whisper in ears to get good things happening for them in the short to mid-term from this.

 

The full article has an explanation. Using the same valuation as AMD did with their ATi (1.7x market cap), an offer of $12.5bn would have been required to acquire nVidia. AMD's market cap is about the same. So a simple acquisition would not have been on the cards - more like a merger - and AMD would not necessarily have been in the driving seat as a consequence.

 

Another point about this deal that I think is worth considering:

AMD as a CPU company has the majority of their experiance invested in full custom designs. On the other hand, GPUs are a strong blend of premade library units and custom units. With CPUs become much larger and more modular (look at multi-core and K8L), ATI's skills may be very useful to AMD so that they can develop CPUs efficently and keep pace with Intel.

 

Heh, good point.

 

That could work the other way around as well. Over time as the feature set of GPUs starts to solidify more, it's possible we'll see the GPU market start to act like CPU markets -- stay with an architecture for years vs changing it rather drastically every 9-12 months. It seems to me they can't continue with the drastic architecture changes forever... (on the software - DX/OGL - and hardware fronts)