HSR efficiency on different cards in real life applications

N

Nappe1

lp0 On Fire!
Veteran
so, I made this thread for testing how much HSR really affects in nowadays applications in different cards.

I quickly run 3DMark 2001 default bench with and without Hyper Z. (only Fast Z-Clear was enabled in both runs.)

Card is AIW Radeon 32MB DDR with default clocks (core 166 Mhz / memory 166Mhz DDR).

I found the following:
Game test 1, Low Detail:
without: 43.3 fps
with: 49.9 fps
15.2 % increase in speed.

Game test 1, High Detail:
without: 20.0 fps
with: 20.1 fps
0.5 % increase in speed.

Game test 2, Low Detail:
without: 41.6 fps
with: 51.6 fps
25.5 % increase in speed.

Game test 2, High Detail:
without: 24.4 fps
with: 25.3 fps
3.6 % increase in speed.

Game test 3, Low Detail:
without: 47.9 fps
with: 54.9 fps
14.6 % increase in speed.

Game test 3, High Detail:
without: 25.5 fps
with: 26.8 fps
5.0 % increase in speed.

Average increase in low detail tests: 18.4 %
Average increase in High detail tests: 3.3 %

hmmh... I was expecting more increase in High than in low detail tests... Maybe AIW is already bandwidth limited so much in 1024x768 that it causes that.

I'll do some tests more.
I have image about compare of results and I'll post link as soon as I get it uploaded...

And, oh yeah... I'd be happy to se what happens to GF3, Radeon 8500 and GF4Ti, when HSR functions (excluding Fast Z-Clear) are turned off. ;)

Edit:
fixed Game Test 2, low Detail. Thanks to Pascal. :)

I also find one reason for such lower High detail scores. AIW Radeon is Geometry Limited on High detail. even in 640x480 mode, it gives hardly 1 or 2 fps more, without depending HyperZ.

And the Compare picture: http://rp-design.totalnfs.net/pics/HyperZEfficiency1.gif
 
I would like to see what a Geforce 4 Ti 4600 with HSR functions off could do VS a Geforce 4 Ti 4600 with 256 bit DDR.

I guess that if Matrox chosed to go with a 128 bit and very efficent HSR method and Nvidia chosed to do things as Matrox did, everyone would just say: WHY ONLY 128 bit? 256 WITHOUT HSR FUNCTIONS IS JUST THE WAY!
 
First post updated.

Pascal: Thanks for the noticing about that mistake. :)

Nocturn Dragon: Good HSR is always positive addon and it is really shame that Parhelia don't have it. Conversation in this thread is meant for trying to figure out, how much Parhelia will suffer on those missing features. In the Parhelia's case it's performance is really depending about amount of overdraw. (Simply because it is drawing pixels that no one can see anyway.)

So benchmarking exiting cards with HSR capabilities, we should be able to terminate how much those cards get benefit Occlusion Culling features.

and Nocturn Dragon, please take that Caps Lock off... ;)
 
Nappe, I don't think the card is geometry-limited, I think your system is CPU-limited. Game1 test-score at low detail isn't very high for example.

Try INCREASING the resolution instead.

*G*
 
It is mainly CPU limited, but the GPU (Radeon) is starting to limit too.

I know that because the Game test 1 high details always interrested me and I have a GF3Ti200 now but I upgraded from a Radeon 32MB DDR and the performance gain was almost 40% with this specific test (from 26 to 34fps).
edited: mistake changed test 2 for 1
 
Well, Kyro II has the relative speccage of a TNT2 or Voodoo4.

It's probably about on par with a GeForce2 PRO. Now that's what I call a performance boost :) ~98% efficient HSR can do that you know :LOL:
 
Tagrineth: yes, but Kyro is in different league in HSR. ;)

I am looking for IMR HSR efficiency values....

and as Grall suggested I took AIW to the limits. (1600x1200x32bpp with 24 bit Z and double buffer doesn't leave much room for textures when the card is 32 Megger.) ;)

Anyways, I am kinda in horry right now, so could someone calculate Increase presentages from the picture, please? :)

http://rp-design.totalnfs.net/pics/HyperZEfficiency2.gif

I just quickly looked them and now those make much more sence.

and Pascal, you are right. R100 Radeon HW T&L unit is much slower than GF2GTS'es. In Fact, on that High Polygon Test, My Celeron500A & GF 256 DDR combo is bit faster than this computer. (AIW Radeon 32MB DDR & AMD Athlon TB 1Ghz on 133 Mhz FSB. both have 384 MB memory.)
 
Here are some for Radeon 8500LE and P4 1.9GHz, too lazy to put my Ti4600 back

Driver Omega 6071 Win2k

Hierarchical Z and Hyper Z II with and without

Game test 1, Low Detail:
without: 102.3 fps
with: 119.3 fps
16.6 % increase in speed.

Game test 1, High Detail:
without: 39.1 fps
with: 39.5 fps
1 % increase in speed.

Game test 2, Low Detail:
without: 107.5 fps
with: 131.2 fps
22 % increase in speed.

Game test 2, High Detail:
without: 65.8 fps
with: 73.3 fps
11.4 % increase in speed.

Game test 3, Low Detail:
without: 101.8 fps
with: 113.4 fps
11.4 % increase in speed.

Game test 3, High Detail:
without: 50.5 fps
with: 52.1 fps
3.2 % increase in speed.
 
Whoa! suprising low Increases in high detail... and your system is definately not CPU limited.

Now I really would like to see some GF4Ti benches too...
 
His system is cpu limited when you see the Game test 1 high detail.

For this specific test the fps grows linearly with the cpu mhz. Aproximatelly 1 fps for each 40Mhz (Athlon or P3).

His CPU is a P4 1.9GHz, more or less an Athlon 1.6GHz (XP1900) or 40 fps :)

For me a good system need at least 30 fps with this specific high detail test because it has twice the expected detail of Unreal 2 outdoor and Doom3.
 
My P4 is not Northwood.

No tweaks or whatsoever is applied as I am more with DVD playback on ATI than games.

BTW, I am using 2 x 256M PC800 RDRAM on ASUS P4T-EA.
 
Pascal: oh... yeah, you are right... but practically, you are CPU limited in games that has such a high polygon counts...
 
I need to run few more tests in actual games... but that is not going happen until next sunday.
 
Ok, this time Ti4600 (ELSA GLADIAC 925 VIVO)

Driver 29.40 Win2k

Z compression and Z Occlusion with and without

PS. RivaTuner mentioned that Z Occlusion cannot be disabled [edited]for 27.xx and after[edited], I use GeForce Tweaker to set it but don't know if it was really disabled

Game test 1, Low Detail:
without: 134.6 fps
with: 140.1 fps
4.1 % increase in speed.

Game test 1, High Detail:
without: 40.5 fps
with: 41.0 fps
1.2 % increase in speed.

Game test 2, Low Detail:
without: 170.4 fps
with: 179.5 fps
5.3 % increase in speed.

Game test 2, High Detail:
without: 100.6 fps
with: 102.5 fps
1.9 % increase in speed.

Game test 3, Low Detail:
without: 124.9 fps
with: 124.3 fps
-0.5 % increase in speed.

Game test 3, High Detail:
without: 55.6 fps
with: 55.7 fps
0.1 % increase in speed.
 
You must log in or register to reply here.

Similar threads

J
Graphic card/CPU configuration for best memory graphics in Flight Simulator 98?
2
Replies
37
Views
5K
homerdog
homerdog
iroboto
B3D IQ Spectrum Analysis Thread [2021-07]
2 3 4
Replies
63
Views
15K
OlegSH
O
D
Geforce vs Radeon features
2
Replies
31
Views
10K
Davros
D
G
Is there a Windows wizard in the house?
Replies
14
Views
4K
Grall
G
P
Portable console for all PC games - PGS
2
Replies
31
Views
13K
Cyan
Cyan
Back
Top