Deep Inside AMD’s Master Plan to Topple Intel

(Last Updated On: July 29, 2011)



Deep Inside AMD’s Master Plan to Topple Intel


AMD’s new graphics architecture isn’t merely about painting prettier pictures. It’s about changing the way computers compute.


Mmmm…. Real Gaming Notebooks….


There have been real gaming notebooks, like Alienware’s, for a while. They just weigh 8 lbs (or so). So I’m also looking forward to seeing acrual lap-able units.


As it is very hard to make a core go faster, adding more cores is natural, as i adding different kind of cores that have different characteristics (such as GP-GPU’s, maybe a low power core for powersave when idle, networking, crypto, etc (as in Niagara).
Does anyone know if there is also any work to speed things up for non-parallel, old code?
In theory, if threads/cores are sharing caches, one could run a “scout thread” (as on Rock) on one core to make sure all data is in cache when the “real” thread need it. It would not be the same thing as having two HW threads on the same core, but might at least gain a little..


you said it yourself — it is very hard to make a core go faster. Many, many techniques have been tried and are being used. Not just silicon technology to increase clock frequency; no more than about 50% of performance improvements came from that in the recent past. Things are now used like branch prediction, issuing multiple instructions simultaneously (when they don’t conflict), simultaneously following both sides of a branch and throwing away the “wrong” one when you finally figure out what the right one was, and so on. I know one guy at the U or Wisconson who investigated predicting what value will come back from memory when a read finishes, and scooting ahead on that assumption rather than waiting for the result. And yes, I’ve heard some folks suggest scout threads.

There’s general agreement, though, that such things like that have reached and passed the point of diminishing returns. Hence we have multicore and multi-system computing facilities.


IDK It’s hard to see a big market for this card, Especially in the HPC world. Basically you probably want to Min/Max which is exactly what this card isn’t.

If your running an embarrassingly parallel problem, this isn’t going to replace discrete cards. 400 shaders just doesn’t compete with the up to the 4100 you get with a 6990. And the fact that you can put 4 in a box.
If your memory limited to say 1 or 2 cards. Your not really going to see a boost either. The lack of L3 cache. And the striping of the gddr5 means your probably be better off with a $75 5670.

So the 2 areas where this will shine is the budget market where making a $75 card available for pretty much nothing is a boost. And the portables market where having a discrete solution and fast and hot GPU memory isn’t a good option anyway.


I still don’t understand why AMD has not gone into a hybrid solution, such as a fusion + FPGA pci coprocessing card… I feel like they forgot about the hpc market. Further, do they have any competition lined for Intel’s “Knights Ferry” cloud compute x86 coprocessor??


AMD isn’t a systems company, nor are they an FPGA company. Solutions which are compatible with being embedded in the same chip as the CPU are clearly more in line with their corporate interest.


ANSYS (Finite Element Analysis) starting mentioning the use of GPUs at the release 13 kick off earlier this year.


I think I am going to buy one system with ADM APU.
Let’s see how to write code for it …


Here is my 6-core 4.0 GHz “James Bond Laptop”


It outperforms a Gulftown @ 3.33 GHz and the i7-990X by a significant margin.

Not available in stores.


And my fastest systems now run at 5.27 GHz so it’s not THAT hard to make the CPU run faster.




NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!
Don't miss out!

You will received instantly the download links.

Invalid email address
Give it a try. You can unsubscribe at any time.


Check NEW site on stock forex and ETF analysis and automation

Scroll to Top