kanecvr wrote:

The X1900 series cards were awesome (and still are). The AGP versions (like the X1950 PRO) are great for older AGP machines and windows XP. The x1900 were also the first cards from ATi to support multi-GPU configurations, using an external connector reminiscent of a voodoo card setup. Crossfire required a "Crossfire Edition" master card witch could be mated with a regular card for multi-gpu. The latter X1950 PRO PCI-E used an internal modern CF cable and did not require a special master card for multi-gpu.

Crossfire was available since X800 series... https://www.newegg.com/Product/Product.aspx?I … N82E16814102639

kanecvr wrote:

It was followed by the x1800 series witch was not spectacular, but they are worth having because they are quite rare in some parts of the world (east europe for example) with AGP versions being extremly rare as they were partner designed cards and not ATi designed cards.

The 3850 / 3870 series are great cards. Back in the day they were very good value for money and offered excellent performance. Most notable are the GDDR4 equipped HD 3870 - as one of the few cards to be equipped with GDDR4 - performance is somewhere in between the 8800GTX and 9800GTX (although back in the day it was only on par with the 8800 really, slightly slower in some games, a tiny bit faster in others)...

The HD 2900 gets an honorable mention. Performance wise (today with most recent drivers) it places in between the 8800GTS 640MB and the 8800GTX, and it was usually sold for less then the GTS. It was notably faster then the later when AA and AF were enabled, actually catching up to the GTX. They are also quite rare, witch makes them interesting for collectors.

Few corrections:

1. X1800 AGP doesn't exist. There are X1950 GT/Pro AGP and also some very rare X1950 XT AGP... but X1800 is pci-e only.
2. Good value for money sure (especially HD 3850) but 3870 is no match for 8800/9800 GTX... 3870X2 was similar to these two (same amount of fps, not actual smoothness). Single RV670 is much slower.
3. As Scali pointed out, AA is extremely slow on R600... actually on all R(V)6xx GPUs, including HD 3800. The massive performance drop was fixed later with HD 4000 series. In my tests 8800 GTX is 54% faster than HD 2900 XT at 1600x1200 4xAA, 16xAF. And 39% faster at 1600x1200 16xAF.

RacoonRider wrote:

kanecvr wrote:

The X1900 series cards were awesome (and still are). The AGP versions (like the X1950 PRO) are great for older AGP machines and windows XP. The x1900 were also the first cards from ATi to support multi-GPU configurations, using an external connector reminiscent of a voodoo card setup. Crossfire required a "Crossfire Edition" master card witch could be mated with a regular card for multi-gpu. The latter X1950 PRO PCI-E used an internal modern CF cable and did not require a special master card for multi-gpu.

Crossfire was available since X800 series... https://www.newegg.com/Product/Product.aspx?I … N82E16814102639

And for sure he means multi card setup, because ATI had a multi GPU card in 1999, Rage Fury MAXX.

Some are worthwhile, the x800 series, the x1800 and x1900 series are worthwhile because they had better IQ than comparable Nvidia cards, but be warned they tend to do worse in OpenGL games than Nvidia, but faster in DirectX. A few good values would be such as the x1800XT it was within 5% of the 7800GTX 512, but you can still find an x1800XT on ebay, good luck on the 7800GTX 512. Their DX10 stuff however was not so hot. The 2900 would place between an 8800GTS 320 and 640 but use more power than an 8800 Ultra, the 3870 is about as fast as 9600GT they traded blows, and you can get an 8800GT for the same price as a 3870 on ebay these days so no real reason to even buy an ATI DX10, unless we are speaking HD4xxx those where a vast improvement.

One thing that's cool about the X-era of Radeon cards is that quite a few of them were 'ViVo' cards (video-in/video-out), which means that you can use them as a capture device for analog sources (composite and s-video at least, perhaps some also with component).
Look for an Ati Rage Theater chip on them: http://www.hardwaresecrets.com/ati-radeon-x1800-xt-review/2/
I have an X1800XT and X1900XTX that both have ViVo.

Scali wrote:

One thing that's cool about the X-era of Radeon cards is that quite a few of them were 'ViVo' cards (video-in/video-out), which means that you can use them as a capture device for analog sources (composite and s-video at least, perhaps some also with component).
Look for an Ati Rage Theater chip on them: http://www.hardwaresecrets.com/ati-radeon-x1800-xt-review/2/
I have an X1800XT and X1900XTX that both have ViVo.

That was an awesome feature, I used my now dead X800XT AiW to grab video off of my old Sony Camcorder with its old 8mm tape.

sprcorreia wrote:

RacoonRider wrote:

kanecvr wrote:

The X1900 series cards were awesome (and still are). The AGP versions (like the X1950 PRO) are great for older AGP machines and windows XP. The x1900 were also the first cards from ATi to support multi-GPU configurations, using an external connector reminiscent of a voodoo card setup. Crossfire required a "Crossfire Edition" master card witch could be mated with a regular card for multi-gpu. The latter X1950 PRO PCI-E used an internal modern CF cable and did not require a special master card for multi-gpu.

Crossfire was available since X800 series... https://www.newegg.com/Product/Product.aspx?I … N82E16814102639

And for sure he means multi card setup, because ATI had a multi GPU card in 1999, Rage Fury MAXX.

X800 crossfire WAS a multi card setup. I didn't know it either and also thought it began with the x1900 series. Here a link

http://www.bit-tech.net/hardware/graphics/200 … sfire_preview/1

havli wrote:

3. As Scali pointed out, AA is extremely slow on R600... actually on all R(V)6xx GPUs, including HD 3800. The massive performance drop was fixed later with HD 4000 series.

Would this mean the HD 4670 actually has an advantage compared to HD 3850 when it comes to AA and massive performance drop?

The R6x0 anti-aliasing situation was interesting to say the least. I'm not sure if I believe they just intentionally crippled MSAA. It seems too ridiculous. 😀 I still wonder if the design was just poor/broken and efficiency dropped off a cliff in various cases. The early drivers were pretty volatile in particular, with performance all over the board as if they just couldn't get the chip to work properly.

This post by Eric Demers (ATI engineer) has always made be question whether the shader resolve for MSAA was actually the problem.
https://forum.beyond3d.com/posts/1019242/

Tetrium wrote:

havli wrote:

3. As Scali pointed out, AA is extremely slow on R600... actually on all R(V)6xx GPUs, including HD 3800. The massive performance drop was fixed later with HD 4000 series.

Would this mean the HD 4670 actually has an advantage compared to HD 3850 when it comes to AA and massive performance drop?

the problem will be finding a proper 4670 AGP, not one with DDR2

but, I guess...
no AA vs AA

2004-2008 is filled with good Radeons, x800, x1K were very competitive, HD 3800 were great value, and HD 4800 was also very competitive.

SPBHM wrote:

Tetrium wrote:

Would this mean the HD 4670 actually has an advantage compared to HD 3850 when it comes to AA and massive performance drop?

the problem will be finding a proper 4670 AGP, not one with DDR2...

Yeah... got one 😀
I tested with Crysis, across two resolutions and medium settings (higher resolution is basicly a SSAA, so it kinda represents how GPU should score with a basic form of AA).
I don't have a proper AGP tester machine at the moment, so MSAA comparison will have to wait.

Crysis @ Medium 1024x768 vs. Full HD :


For QX9770 OC'ed :
HD 4670 1GB drop = 54% (it retained 46% of 1024x768 performance).
HD 3850 512MB drop = 38% (it retained 62% of 1024x768 performance).

Basicly : No matter how good RV7x0 architecture is, it's no match in performance drop comparison for resolution/SSAA when RV7xx has 8ROP/128-bit bus and RV6xx has 16ROP/256-bit bus (at least with the GPU/VRAM clocks I tested with).
There simply isn't enough raw pixel pushing power to make higher resolutions work better on 4670 card.

PS. GPU-z screenshot of cards used :

swaaye wrote:

The R6x0 anti-aliasing situation was interesting to say the least. I'm not sure if I believe they just intentionally crippled MSAA. It seems too ridiculous. 😀 I still wonder if the design was just poor/broken and efficiency dropped off a cliff in various cases. The early drivers were pretty volatile in particular, with performance all over the board as if they just couldn't get the chip to work properly.

This post by Eric Demers (ATI engineer) has always made be question whether the shader resolve for MSAA was actually the problem.
https://forum.beyond3d.com/posts/1019242/

They couldn't make the hardware work right, it wasn't supposed to be that slow, but they couldn't solve it so for the R700's they went back to dedicated hardware.

SPBHM wrote:

Tetrium wrote:

Would this mean the HD 4670 actually has an advantage compared to HD 3850 when it comes to AA and massive performance drop?

the problem will be finding a proper 4670 AGP, not one with DDR2

2004-2008 is filled with good Radeons, x800, x1K were very competitive, HD 3800 were great value, and HD 4800 was also very competitive.

I checked, mine is DDR3 😁
Too bad I don't have a good AGP board + CPU to put it in 🤣

agent_x007 wrote:

Yeah... got one :) I tested with Crysis, across two resolutions and medium settings (higher resolution is basicly a SSAA, so it […]
Show full quote

Yeah... got one 😀
I tested with Crysis, across two resolutions and medium settings (higher resolution is basicly a SSAA, so it kinda represents how GPU should score with a basic form of AA).
I don't have a proper AGP tester machine at the moment, so MSAA comparison will have to wait.

Basicly : No matter how good RV7x0 architecture is, it's no match in performance drop comparison for resolution/SSAA when RV7xx has 8ROP/128-bit bus and RV6xx has 16ROP/256-bit bus (at least with the GPU/VRAM clocks I tested with).
There simply isn't enough raw pixel pushing power to make higher resolutions work better on 4670 card.

PS. GPU-z screenshot of cards used :

GPU-z 4670 vs. 3850.png

Which HD 4670 AGP do you have? Mine is a Club 3D version, though I think there was a HIS with faster clocks?
A friend of mine also has the exact same AGP card. We both actually each decided to go get one when new...in a real PC parts store 😁

Tetrium wrote:

Which HD 4670 AGP do you have? Mine is a Club 3D version, though I think there was a HIS with faster clocks?
A friend of mine also has the exact same AGP card. We both actually each decided to go get one when new...in a real PC parts store 😁

This one : LINK 😀
There indeed was, it has a bit faster DDR3 memory installed - 1750MHz, instead of 1600MHz [which I got].
I can't OC mine to over 1720MHz, it's simply not stable at all at those settings (750/1750).
I doubt it would make mush difference either way... (it's only 9,4% faster VRAM vs. my stock value).

PS. What's your VRAM frequency ?

agent_x007 wrote:

Yeah... got one :) I tested with Crysis, across two resolutions and medium settings (higher resolution is basicly a SSAA, so it […]
Show full quote

Yeah... got one 😀
I tested with Crysis, across two resolutions and medium settings (higher resolution is basicly a SSAA, so it kinda represents how GPU should score with a basic form of AA).
I don't have a proper AGP tester machine at the moment, so MSAA comparison will have to wait.

Basicly : No matter how good RV7x0 architecture is, it's no match in performance drop comparison for resolution/SSAA when RV7xx has 8ROP/128-bit bus and RV6xx has 16ROP/256-bit bus (at least with the GPU/VRAM clocks I tested with).
There simply isn't enough raw pixel pushing power to make higher resolutions work better on 4670 card.

PS. GPU-z screenshot of cards used :

the Oblivion results I posted I think are from reference cards (the best 4670 with 2GHz GDDR3), your cards it seems the 3850 is an OC model, while the 4670 is running with slower memory, which considering the card only have a 128bit bus it probably is significant, but I'm not aware if 4670 exists with 2GHz memory for AGP?
but as I said, DDR2 4670 AGPs were so common that a 1600 model is already quite nice, it's something to consider most reviews had results from 4670s with 2GHz memory, and with that it did a good job against the 3850 reference, but with slower memory 4670s and 3850s OC not so much.

agent_x007 wrote:

This one : LINK :) There indeed was, it has a bit faster DDR3 memory installed - 1750MHz, instead of 1600MHz [which I got]. I ca […]
Show full quote

Tetrium wrote:

Which HD 4670 AGP do you have? Mine is a Club 3D version, though I think there was a HIS with faster clocks?
A friend of mine also has the exact same AGP card. We both actually each decided to go get one when new...in a real PC parts store 😁

This one : LINK 😀
There indeed was, it has a bit faster DDR3 memory installed - 1750MHz, instead of 1600MHz [which I got].
I can't OC mine to over 1720MHz, it's simply not stable at all at those settings (750/1750).
I doubt it would make mush difference either way... (it's only 9,4% faster VRAM vs. my stock value).

PS. What's your VRAM frequency ?

Mine was this one
1333 instead of 1600 so mine still has a bit slower memory, but at least it's not a HD 4650.
But I'm glad I got it 😀

edit: I'm not even sure it even matters that much, isn't DDR2-800 actually faster than DDR3-1333?

@up
In CPU's it can be, in most cases (if you have a IMC, and low timings on DDR2 with high timings on 1333MHz).

In GPU's... nope.
GPU's actually need all the bandwidth they can get (it's used by ROP's and TMU's primary, but Shaders use it as well, to store programs that are too big for build-in L1/L2 cache).

In case of HD 4670 (with that 8ROP's and 32TMU's along with 320SP's all working at 750MHz), it can get bandwidth starved*.
*Usually it's not a problem, since most CPU's that are used with AGP post aren't fast enough to show it.

@SPBHM
As for "3850 OC" I used... LINK 😉
Still, second fastest "stock" HD 4670 AGP vs. fastest "stock" 3850 AGP 😁
There are no 2GHz 4670's for AGP.
I READ that HIS 4670 AGP w/1750MHz VRAM, can be OC'ed to 2GHz VRAM.

PS. Stock clocks for 3850 AGP are : 668MHz/1656MHz
So my card is +7% on GPU and +9% on VRAM.
In case of 4670, my 1600MHz VRAM is already ~20% faster than Tetrium's Club 3D card with 1333MHz.
Last but not least : Bandwidth wise - 800MHz on DDR2 = 800MHz on DDR3.

I think that's also the reference clock for the 3850 PCIE, in any case the 3850 have more room for OC than a 4670 so that's something to consider as the best possible AGP, even if you can get something up to 2GHz memory.

DDR3 1333 is way faster than DDR2 800 for a GPU
the 4670 I have runs with DDR2 800 I think and it's quite terrible, in some cases it looks slower than a 8600GT (which had 1.4GHz memory but the GPU itself is a lot slower)

My PowerColor HD4670 has DDR3 too. The website says it's 1333MHz. I'll check the ram modules.

EDIT: Ram modules are rated 1600MHz. I guess it's time to power on a machine to check this.

EDIT: And for a comparison my Gigabyte HD4650 has 800MHz DDR2 modules.

To my knowledge HIS had the fastest HD4670 AGP.
750MHz core and 1GB GDDR3 128bit memory at 1750.
They also released one with slower GDDR3.
http://www.hisdigital.com/un/product1-47.shtml
Both have a very nice IceQ blower cooler and run cool and silent.

I have the fast version, they don't appear that often for sale so if you see one for a good price grab it.