Welcome, Guest. Please login or register.

Author Topic: AmigaONE benchmark - Important  (Read 19380 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline Kronos

  • Resident blue troll
  • Hero Member
  • *****
  • Join Date: Feb 2002
  • Posts: 4017
    • Show all replies
    • http://www.SteamDraw.de
Re: AmigaONE benchmark - Important
« on: February 18, 2003, 09:20:54 PM »
Well if your read the actual post you would see that the client
DOES have altivec-support, and such benches are really good for
altivec.

There is no way a G4/800 could beat a P4/2600 in integer or single
data float.

And the 3rd-level-cache on the G4 is only needed because it
doesn't have any other posibility to use modern DDR-RAM, which
comes naturally for todays x86s.
1. Make an announcment.
2. Wait a while.
3. Check if it can actually be done.
4. Wait for someone else to do it.
5. Start working on it while giving out hillarious progress-reports.
6. Deny that you have ever announced it
7. Blame someone else
 

Offline Kronos

  • Resident blue troll
  • Hero Member
  • *****
  • Join Date: Feb 2002
  • Posts: 4017
    • Show all replies
    • http://www.SteamDraw.de
Re: AmigaONE benchmark - Important
« Reply #1 on: February 19, 2003, 03:17:03 PM »
@KennyR
No I am correct !!

The perfect CPU-RAM setup was what we had in an A2000 with Fast-RAM.

The CPU had access to the full 8mb every cycle.

Problem is that while CPUs got faster and faster the RAM wasn't keeping
up which made caches necessary. A cache is always a bottleneck,
and the more stages you got the worse it gets. Having a huge gap
between to stages is also a problem.

So starting with a slow main-RAM is not a good idea.

The use of a fast cache is much bigger in popular benchmarks
than it is in real-world SW, and that is why PPC is hanging behind
x86 when both run the same SW on the same OS.
1. Make an announcment.
2. Wait a while.
3. Check if it can actually be done.
4. Wait for someone else to do it.
5. Start working on it while giving out hillarious progress-reports.
6. Deny that you have ever announced it
7. Blame someone else