AGEIA comments about the affinity between PhysX and Cell (Goto's article @ PC Watch)

Discussion in 'GPGPU Technology & Programming' started by one, Apr 7, 2006.

Thread Status:
Not open for further replies.
  1. dukmahsik

    Banned

    Joined:
    May 19, 2005
    Messages:
    994
    Likes Received:
    9
    there's nothing new in the article imo

    architecturally 7900 is same to the 7800 to the 7600
     
  2. SPM

    SPM
    Regular

    Joined:
    Dec 18, 2005
    Messages:
    639
    Likes Received:
    16
    http://www-128.ibm.com/developerworks/power/library/pa-fpfeib/

    The Cell Processor's Element Interconnect Bus won't saturate even if RSX access to the main 256MB XDR memory has to go through the Cell. Also if inter SPE or PPE to SPE communication has to jump over another SPE to SPE communication it still doesn't block communication fully because there are two paths in each direction. Also transfers on the EIB are apparently managed to minimise bus contention.

    According to the above the data rings can sustain 204.8GB/s for some workloads, with transient rates as high as 307.2GB/s. This means that even with both XDR running at 25.6GB/sec outbound
    and Flex i/o transfers running at 30GB/s, the EIB will not be close to saturated.

    Also Cell's SPE to SPE and PPE to SPE bandwidth will be much larger compared with the Aegis PPU, if stream processing rather than random communication between SPEs is considered.

    I presume the Xbox's unified memory architecture means there will be bus contention between the Xenon CPU and Xenos GPU so the 22.4 GB/s memory interface bus bandwidth to the 512MB of RAM will be shared between Xenon and Xenos.

    On the PS3, the 256MB XDR RAM and the 256MB flex i/o ram can be accessed simultaneously without contention between the two memory buses.
     
  3. dskneo

    Regular

    Joined:
    Jul 25, 2005
    Messages:
    531
    Likes Received:
    45
    Its all good for EIB supporting rsx at 100% without a hit, but pratically cell as its own work to do with its end of the xdr B/W
     
  4. DarkRage

    Newcomer

    Joined:
    Jul 25, 2005
    Messages:
    70
    Likes Received:
    1
    Location:
    Spain
    I believe Ageia is not specially interested in bringing physics calculations to the GPUs. They are investing a lot of money on their chip to also support GPUs calculating physics, even if they would be great for it (which I think they are only in some cases).
    Let's see what Havok can do with GPUs calculating physics.

    And for similarities, it's OK. However Ageia PhysicX chip is dedicated hardware. With only 6 SPEs available in Cell to developers, and a full game with rendering, IA, audio, game mechanics and so on, I don't expect more than a couple of SPEs dedicated to physics, which would underperform dedicated hardware, for both optimization and available resources.

    Then we can discuss about money, and about if it is worthy.
     
  5. ADEX

    Newcomer

    Joined:
    Sep 11, 2005
    Messages:
    231
    Likes Received:
    10
    Location:
    Here
    I wrote an article on PhysX a while back based on the patent, Goto's piece seems to confirm a lot of what I thought.

    My article is here (I compared it with Cell in the 3rd part):
    http://www.blachford.info/computer/articles/PhysX.html
     
  6. predicate

    Newcomer

    Joined:
    Jan 6, 2006
    Messages:
    128
    Likes Received:
    2
    I was under the impression that there are some fairly significant architectural changes between the 7800 and 7900?
     
  7. SugarCoat

    Veteran

    Joined:
    Jul 17, 2005
    Messages:
    2,091
    Likes Received:
    52
    Location:
    State of Illusionism

    nope. this is not ment to be rude, but read a review.
     
  8. DeanoC

    DeanoC Trust me, I'm a renderer person!
    Veteran Subscriber

    Joined:
    Feb 6, 2003
    Messages:
    1,469
    Likes Received:
    185
    Location:
    Viking lands
    Cell EIB has a 'feature' that it always takes the shortest path between any two points on the ring regardless of anything else.
     
  9. predicate

    Newcomer

    Joined:
    Jan 6, 2006
    Messages:
    128
    Likes Received:
    2
    I know they are the same architecture, but I thought nVidia had messed around with it some for the 7900? i.e. more power with fewer transistors.
     
    #49 predicate, Apr 9, 2006
    Last edited by a moderator: Apr 9, 2006
  10. Alpha_Spartan

    Regular

    Joined:
    May 26, 2005
    Messages:
    559
    Likes Received:
    10
    QFT. Notice how the same people that are quick to conclude that the 256 GB thing is bunk (which it is) ran with Sony's big E3 PS3 marketing-fest, faked videos and all. It's all marketing BS.

    Secondly, let's get this straight. 128-bit XDR + 128-bit GDDR3 != single 256-bit memory bus performance-wise. I don't want to insult anyone's intelligence by stating the obvious, but someone has been dragging this idea to other boards with the "B3D boards says..." line (as if the B3D forums is a de facto source). Just because something CAN be done doesn't mean it WILL be done or it's PRACTICAL in real-world scenarios. You can always have a point when you argue in theory. But just remember that IN THEORY the PS3 is 2TFLOP and the Xbox 360 has 240 GB of bandwidth.
     
  11. mckmas8808

    Legend

    Joined:
    Mar 8, 2005
    Messages:
    6,744
    Likes Received:
    28
    If you read Mr. Wibble's quote even he says that the 128-bit XDR + 128-bit GDDR3 is not as good as a single 256-bit bus. That quote was posted on that "other" site. So no, nobody is saying that it is equal.
     
  12. Mmmkay

    Regular

    Joined:
    Jul 3, 2005
    Messages:
    627
    Likes Received:
    31
    And rei-rom is the unimpeachable source of integrity with you and Deadmeat leading the charge right? The problem there is you've got two incredibly disingenuous techno 'literates' leading a forum of people who don't care for details and will accept blindly whatever is presented. Those kids are bound to take external information in just as blindly. You may be seeing a lot of 'B3D says', but in fairness you and DM have said things just as preposterous which is causing that kind of desperation.
     
  13. j^aws

    Veteran

    Joined:
    Jun 1, 2004
    Messages:
    1,939
    Likes Received:
    42
    FYI, XDR has an OCTAL data rate, it's 'effective' clock is 3.2 GHz and to give 25.6 GB/sec b/w (the important number),

    3.2 GHz x 8 bytes (64bit bus) ~ 25.6 GB/sec

    The XDR bus would be 64 bit, NOT 128bit.
     
  14. Titanio

    Legend

    Joined:
    Dec 1, 2004
    Messages:
    5,670
    Likes Received:
    51
    Yeah, good point.

    I'd also say that while 2 busses aren't the same as one with a higher bandwidth than either individually, the differences aren't all bad. The XDR and GDDR3 busses allow for more bandwidth than a single 256-bit GDDR3 bus clocked at the same rate as RSX's, would, and presumably there may be less contention with two busses versus one.

    I guess it's also good if it allows them to use different memory types with behaviours tuned to each chip. I guess there's a reason why STI chose XDR for Cell versus something like GDDR3, for example.

    Of course, there are downsides also.
     
  15. rounin

    Veteran

    Joined:
    Sep 21, 2005
    Messages:
    1,251
    Likes Received:
    20
    So...maybe a kind of (Crippled) async dual channel?
     
  16. j^aws

    Veteran

    Joined:
    Jun 1, 2004
    Messages:
    1,939
    Likes Received:
    42
    From the POV of the b/w consumer, i.e. RSX, it has GDDR3 + FlexIO b/w available to it , peak. And yes, it would be greater than GDDR3 + PCI-e...

    Not sure what you mean by crippled... but I expect 4 XDR chips (512Mbit each) in total, making 256 MB. With 2 chips assigned to each channel of the XDR controller. IIRC, the XDR controller in CELL was always dual channel....
     
  17. randycat99

    Veteran

    Joined:
    Jul 24, 2002
    Messages:
    1,772
    Likes Received:
    12
    Location:
    turn around...
    Damn, that was brutal (but I totally agree). ;) I've noticed this phenomenon at a variety of forums in recent times, and it is chilling. Popularity somehow becomes credibility and anyone questioning that person (who is most obviously spewing some amazing BS at one moment) gets shouted down by the forum shills, who would just as soon follow that person through hell and back, rather than attempt to have an independent thought/reasoning in their head. It's most dissapointing to see this when that person happens to be a mod and acts the way they do, completely w/o impunity. You know right away that truth, honesty, and reason have taken a back seat to popularity when they start banning anyone who attempts to disagree with them (places of particular note- AppleInsider Political forum, GranTurismo Open Road forum, HDTVoice forums, SiriusBackstage forums).
     
    #57 randycat99, Apr 9, 2006
    Last edited by a moderator: Apr 9, 2006
  18. Alpha_Spartan

    Regular

    Joined:
    May 26, 2005
    Messages:
    559
    Likes Received:
    10
    Where the [Moderated] did you get Rei-rom and Deadmeat from what I [Moderated] said???
     
  19. Tahir2

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    2,978
    Likes Received:
    86
    Location:
    Earth
    Is there really a need to use the "f" word?
     
  20. Mmmkay

    Regular

    Joined:
    Jul 3, 2005
    Messages:
    627
    Likes Received:
    31
    Well jeez, Kr0tchr0t I thought you were talking about this thread here being the 'other boards' where B3D information gets posted. The one where a lay posted some information he read on NeoGAF which mostly originated from B3D posters and got somewhat lost in the translation along the way. The information itself was to address the overwhelming FUD on rei-rom.

    However, the intention of what I said was merely to suggest people in glass houses probably shouldn't be throwing stones.
     
Loading...
Thread Status:
Not open for further replies.

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...