#1 Multiprocessor discussion
Posted: Thu Jun 09, 2005 5:28 pm
Intel is pursuing higher clockspeed, while AMD goes for more instruction per cycle. In consumer-level processing, it seems there are only those two paths.
However, outside the PC world, Sun has been continuing to refine the art of Symmetric Multiprocessing. However, such approach tends to go unnoticed in the consumer market.
I wonder:
(1) Is PC going to benefit from multiprocessing? IMO, such approaches has been less than successful in PC market, particularly games. How many games actually take advantage of Hyperthreading, for instance? (yes I know HT is not really multiprocessor approach but it was close).
(2) How would graphic-intensive applications like games benefit from multiprocessing? I remember a friend at 3dfxZone claimed his Falcon 4.0 showed some noticeable improvements when running on dual AMD processor (the OS is Windows 2000 professional). Do games have to be specifically written to take advantage of multiprocessing, or it's just the matter of the OS?
(3) How difficult it is to make a scalable processor? I'm not sure whether this is relevant to the discussion, but if we take a look on video cards, I get the impression that it's fairly difficult to create a scalable GPU. One of the reasons while VSA-100 (the chip of Voodoo5) was admired is because its flawless scalable implementation, while other attempts like ATI Rage Fury MAXX and Volari Duo have not been very successful. So it is that difficult to design a scalable processor?
And how about CPU? From my own experience, even multiprocessor Pentium IV servers can never beat Sun servers in terms of speed and stability (note that my implementations are Oracle databases, which involve a lot of parallel processing). Sun always keeps it up with increasing amount of transactions, while the Pentium starts to be brought down to its knees.
(4) How likely would AMD or Intel go for processor scalability?
However, outside the PC world, Sun has been continuing to refine the art of Symmetric Multiprocessing. However, such approach tends to go unnoticed in the consumer market.
I wonder:
(1) Is PC going to benefit from multiprocessing? IMO, such approaches has been less than successful in PC market, particularly games. How many games actually take advantage of Hyperthreading, for instance? (yes I know HT is not really multiprocessor approach but it was close).
(2) How would graphic-intensive applications like games benefit from multiprocessing? I remember a friend at 3dfxZone claimed his Falcon 4.0 showed some noticeable improvements when running on dual AMD processor (the OS is Windows 2000 professional). Do games have to be specifically written to take advantage of multiprocessing, or it's just the matter of the OS?
(3) How difficult it is to make a scalable processor? I'm not sure whether this is relevant to the discussion, but if we take a look on video cards, I get the impression that it's fairly difficult to create a scalable GPU. One of the reasons while VSA-100 (the chip of Voodoo5) was admired is because its flawless scalable implementation, while other attempts like ATI Rage Fury MAXX and Volari Duo have not been very successful. So it is that difficult to design a scalable processor?
And how about CPU? From my own experience, even multiprocessor Pentium IV servers can never beat Sun servers in terms of speed and stability (note that my implementations are Oracle databases, which involve a lot of parallel processing). Sun always keeps it up with increasing amount of transactions, while the Pentium starts to be brought down to its knees.
(4) How likely would AMD or Intel go for processor scalability?
