About the original HyperZ implementation

[no-X]

Probably isn't. It has actually more bandwidth for it's single pixel pipeline, than Radeon SDR (R100+128bit SDR) for two pipelines. I'll try to find Radeon VIVO (R100 128bit 64MB DDR) and Radeon 7200 (R100 128bit 64MB SDR), clock them equally and run some fast tests...

 

[Nick]

so it does run games that demand hardware tnl ???? (max payne i seem to remember was one of them)

All except the most recent Intel IGPs do the vertex processing on the CPU, yet report 'hardware' processing. Software processing in the Direct3D sense really means using the runtime's built-in vertex processing. Otherwise it's up to the driver to deal with it.

Besides, the Radeon VE runs at 183 MHz, while the average CPU of that time ran at 1.5 GHz. So with 3DNow! or SSE their vertex processing capabilities were nothing to sneeze at.

 

[no-X]

R100 SDR vs. R100 DDR, both @166MHz

RV100 (64bit DDR) has about the same BW as R100 SDR, which seems to be sufficient...

 

[msxyz]

Interesting results; looking at the figures it seems that the extra bandwidth of the Radeon DDR isn't really needed in most situations. On the other side, HyperZ has probably little impact on a Radeon DDR.

I wonder if ATi gambled on Hyperz because it was unsure about DDR RAM price/availability at the time or because DDR RAM was a late addition to the project

 

[no-X]

R7200 (183/183) doesn't support assynchorous core/memory clocks (supports, but is unstable). But R7500 (290/230) supports it correctly. This is just a guess, but couldn't it be just a HW bug (pre-R100 chips were asynchornous, next too)? R100 has a lot of OC headroom and I read somewhere, that ATi initialy planned >200MHz core and (cheaper) 166MHz RAMs (final Radeon DDR was 183/183). It seems, that slower memory modules wouldn't hurt performance because of HyperZ, but price would be significantly lower and faster core could increase performance at least by 10%. Final Radeon DDR had quite bad price/performance ratio, but later R7500, which brought asynchronnous mem. controller, was very competitive product (firstly competed with GF2Ti, than with GF4MX440). Maybe cheaper VRAM was one of the former intentions.

 

[Davros]

I wonder if ATi gambled on Hyperz because it was unsure about DDR RAM price/availability at the time or because DDR RAM was a late addition to the project

Im not sure hyperZ seems to of benefitted the 8500 quite a bit and no doubt was just passed down to the slower chips even if they didnt need it much

 

[msxyz]

The Radeon 8500 employed second generation HyperZ technology and there's no doubt that it helped save some bandwidth. Like NVidia chips of the time, it was a 4x2 architecture with four pipelines racing to write a lot of pixel and Z data.

No-X's speculation that the classic Radeons were intended to reach higher speeds might be true. ATi initially claimed 200 MHz clock operation and I've heard a few people saying that the original Radeon was so cool running that it really didn't need a fan on it. And its descendant, the Radeon 7500, could be pushed to insane clock speeds. Most reviewers of the time seemed to be able to reach 340-350MHz without problems.

If I'm able to snatch off eBay a cheap Radeon 7xxx, I'll promise I'll do some more tests. Stay tuned! :smile:

 

[t0y]

I'd trade my VIVO for a 2900XT....





Actually, I still use it and although it's my second one (after a RMA) it was money well spent. Never regretted buying it even watching it being beaten by nv cards.
It says DX8 compatible in the box, *exxxcelent value*!

 

[Humus]

With HyperZ enabled: 32 With HyperZ disabled: 31

IIRC the Radeon VE didn't support HyperZ. You could try testing with my old GL_EXT_reme benchmark I wrote back in the days. The overdraw tests should give a pretty good hint on whether it's there or not.

 

[msxyz]

Are you sure ?

HyperZ is mentioned in the official product brief:
http://ati.amd.com/products/radeon7000/radeonve/index.html

 

[no-X]

OD factor 8

R7000 - RV100 DDR 183/183MHz 64bit: 41,3 / 44,8 / 43,9
R7200 - R100 DDR 183/183MHz 128bit: 80,1 / 82,3 / 81,8
R7500 - RV200 DDR 270/200MHz 128bit: 106,0 / 133,8 / 125,7

 

[swaaye]

I have a 7500 in the drawer at home. I bought it off ebay not long ago because I wanted to mess around with a 150 nm R100. I tried some overclocking; it does 310/330. That's a 100 MHz overclock on the RAM! I played Guild Wars on it for a bit and it ran it really quite well.

3DMark01SE results on a 1438 MHz Tualatin
http://service.futuremark.com/compare?2k1=8715147

 

[msxyz]

Hi,

so I purchased on ebay a really cheap 7500. It turned out it isn't really a 7500 but rather a mobility M7 placed onto a half height board, similar to those of many budget videocards. Brand is unknown, DeviceID returns 1002&4C57 which is the value corresponding to an ATi inc. Mobility M7 7500 series.

It has 64 bit bus and DDR memory It's factory clocked at 200 MHz / 143 (286DDR) but it overclocks really well with the core going up to ~300 MHz before artifacts start to appear. It should be mentioned that this thing has also a small passive heatsink which remains barely warm even under load.

Out of curiosity, I clocked the board and RAM at the same speeds of my Radeon 7000 which are 166 MHz both. Bandwidth should also be the same (2.6 GB)

I obtained 3300 points at 3DMark 01, 56 fps at villagemark (1024x768x32 no trilinear). Again disabling HyperZ in the drivers or with radlinker doesn't change the results which leads me to believe in late Catalyst this feature cannot be altered.

As an interesting side-note, the 3X Antialiasing mode it was hidden (and working) in the R7000 seems to be gone Only 2x and 4x works.

 

[swaaye]

I have a Radeon LE that has defective HyperZ. With the more recent drivers, HyperZ comes enabled for this board (originally, HyperZ would be disabled). With it enabled, the board draws rendering errors that get worse as I overclock it. W/O it's semi-bugged HyperZ, it can clock quite high w/o errors.

Radeon LE is a full R100 sold cheap, btw. They come clocked at 150 Mhz and do usually have HyperZ problems. Anandtech had a review of it once.

With 7500, I noticed that the AF slider was more granular. I could pick 2x, 4x, 8x, 16x. With the original Radeon, the drivers only offer Off and 16X, I believe.

 

[no-X]

With 7500, I noticed that the AF slider was more granular. I could pick 2x, 4x, 8x, 16x. With the original Radeon, the drivers only offer Off and 16X, I believe.

I don't have R7500, so I can't compare, but R100 gives quite strange results, when you choose (via tweaker) AF2x/4x/8x:

trilinear


AF 2x


AF 4x


AF8x


AF16x

 

[swaaye]

The outer circle seems to shrink with each further increase in AF level. Whatever that means. R100 and R200 have extremely angle dependent AF, btw. And they don't do AF w/ trilinear at all. So that is bilinear + "AF" there.

 

[no-X]

The real problem of R100 isn't angle dependancy, but bad mip-map levels... look at AF4x screenshot. "Red" level i missing for 0/90/180/270°

R200/R8500 fixed this. Not sure about RV200/R7500...