Sony Technical Information on PS3


Recommended Posts

http://www.watch.impress.co.jp/game/docs/20060329/3dps3.htm

* PPE has 128 VMX registers

* Memory Latency to cell:

1) Register: 1 Cycle

2) L1 Cache : 8 Cycles

3) L2 Cache : 32 Cycles

4) Main memory: 140 Cycles

* Overall Cycles: 500, same as Xbox 360

* 6 SPEs reserved for games, 1 SPE reserved for Operating System

* RSX is based on N47 Nvidia Architecture = Nvidia 7800 GTX

* RSX is 384 Flops per Clock at @550 Mhz, Nvidia 7800 GTX is 728 Flops per Clock @430 Mhz

* SPEs cannot access main memory directly

2)

Link to comment
https://www.neowin.net/forum/topic/447554-sony-technical-information-on-ps3/
Share on other sites

* Memory Latency to cell:

1) Register: 1 Cycle

2) L1 Cache : 8 Cycles

3) L2 Cache : 32 Cycles

4) Main memory: 140 Cycles

* Overall Cycles: 500, same as Xbox 360

Not overall cycles. Xbox 360 main memory have 500+ cycles latency, with latency lower is better.

Anyway most of this is already know before.

Sony upgraded the VMX registers on the PPE from 64 to 128

I thought RSX was just a G70 @ 128bit, if that is true its a downgrade from what I assumed it to be.

it shouldnt have been a surprise. GTX has more memory bandwidth than RSX. even X1900 has more memory bandwidth than the Xenos in Xbox 360

it shouldnt have been a surprise. GTX has more memory bandwidth than RSX. even X1900 has more memory bandwidth than the Xenos in Xbox 360

No, my point was that I expected RSX to turn out the same number of flops per clock as a regular G70 but it would have less memory bandwidth (128bit).

According to your specs, you have the the RSX at lower flop per clock performance. <- Where did you get that figure btw? :huh:

No, my point was that I expected RSX to turn out the same number of flops per clock as a regular G70 but it would have less memory bandwidth (128bit).

According to your specs, you have the the RSX at lower flop per clock performance. <- Where did you get that figure btw? :huh:

RSX

http://www.watch.impress.co.jp/game/docs/2...329/3dps309.jpg

G70(7800 GTX)

http://www.pcperspective.com/images/review.../g70summary.jpg

personally I dont think it really matters, does it? I mean if the programmers can take full utilization from the cell, couldn't they really just render most of the stuff needed on the processor subtract textures-fillings-etc? If so this would take the load off of the gpu, and allowing the gpu to do all textures-hdr-bumpmapping-etc..?

So if the cell only has 6 usable SPEs and it's supposed to do some graphics, all those kick-ass physics and whatnot -- whats going to run the gamecode? :blink:

People have to realise that the Cell isn't some super processor (outside of the usual marketing speak ofcourse). There has to be balance somewhere. It may be able to do alot of things but surely it can't do them all at once and do them well :rolleyes:

So if the cell only has 6 usable SPEs and it's supposed to do some graphics, all those kick-ass physics and whatnot -- whats going to run the gamecode? :blink:

People have to realise that the Cell isn't some super processor (outside of the usual marketing speak ofcourse). There has to be balance somewhere. It may be able to do alot of things but surely it can't do them all at once and do them well :rolleyes:

The RSX takes care of the graphics the Cell does everything else as far as i know.

This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • No, "a great deal" for 32GB of DDR5 is $50, not $350. I mean I see what you mean, that it's a decent price compared to what's currently available, but you really should put a disclaimer in this articles explaining that it's still multiple times more expensive than it used to be.
    • Linux 7.1 stable launch looms as Linus Torvalds releases the final release candidate by Paul Hill Linus Torvalds has just released what’s expected to be the final release candidate of Linux 7.1, rc7. The Linux founder said that this RC is not small, but smaller than recent releases, which is a good sign because he expects the stable version to drop next week if things continue on this trajectory. Linux kernels see a merge window for the first two weeks of their life, where developers add new features, then there are about seven or eight weeks of release candidates before the stable version. Typically, there are seven release candidates, but if more time is needed, then an eighth release candidate is released too. This week’s RC’s biggest area of fixes was for GPUs, with networking just behind. Torvalds said that the rest of the release was “pretty random and spread out” with some architecture fixes, driver fixes, filesystem improvements, and build fixes for more unusual configs. In terms of specific pieces of hardware receiving improvements in this update, we had more AMD Zen6 models supported and fixes for AMD SDMA 7.1 and GFX11. Hardware that got improvements includes Lenovo laptops, HONOR laptops, and MSI laptops. Here are the changelogs for those: ASoC: amd: acp: Add DMI quirk for Lenovo Yoga Pro 7 15ASH11 Input: atkbd - add DMI quirk for Lenovo Yoga Air 14 (83QK) Input: atkbd - skip deactivate for HONOR BCC-N's internal keyboard ASoC: amd: yc: Add MSI Raider A18 HX A9WJG to quirk table ASoC: amd: yc: Enable internal mic on MSI Bravo 17 C7VF When the stable Linux 7.1 is released, it will be up to distribution maintainers, such as Canonical and Red Hat, to release the update to their users via the update manager. Some versions of Linux will get it before others, and some will never get it at all. Fedora and Arch-based distros will be among the first to get it, though. If you don’t get it, the security fixes will be backported to your system’s kernel, so you won’t be at risk, but you won’t get newer hardware support, which is fine if your computer works now.
    • Ideally, the algorithm is smart enough to see the real sender ID and non-spoofed address to block it. Ideally.
  • Recent Achievements

    • One Month Later
      DJC50PLUS earned a badge
      One Month Later
    • Week One Done
      DJC50PLUS earned a badge
      Week One Done
    • Proficient
      Eric Biran went up a rank
      Proficient
    • Dedicated
      Conjor earned a badge
      Dedicated
    • Week One Done
      Windows Guy earned a badge
      Week One Done
  • Popular Contributors

    1. 1
      +primortal
      493
    2. 2
      PsYcHoKiLLa
      249
    3. 3
      Steven P.
      71
    4. 4
      ATLien_0
      68
    5. 5
      +Edouard
      68
  • Tell a friend

    Love Neowin? Tell a friend!