Nvidia Ion vs Radeon 9700 Pro

[v8envy]

The 8600GT (and lower) were designed for low resolutions and FSAA off. Look at those benchmarks. Except for the UT3 one, they're all at low res. The 8600 was severely crippled by lack of memory bandwidth, although it had some shader power (not enough to run PhysX these days, of course).

For early DX9 titles at better than 1280x1024 the 9700 would easily keep up with a 8500GT and possibly even a 8600GT. It could absolutely push 60 fps in HL2 at 1024x768, 1280x1024 and even 16x12 without FSAA.

In any event, "way faster" is a bit of a stretch. It might be faster in newer games (where it's an ugly, low res slide show anyway) and slower in the old games both cards are able to run acceptably in the first place.

 

[Schmide]

It's a loaded question. As much as the 9700 has more processing power, its shaders are built on a vector model rather than the stream processor. The later being much more flexible. While the vector model of the 9700 has a higher theoretical throughput, the 8400 slightly makes up for this in its flexibility of operation. One of the big sexy tricks added in HL2 lost coast, was the use of HDR lighting and thus some advantage was given even to the lower end cards of the latter generation.

 

[toyota]

Originally posted by: v8envy
The 8600GT (and lower) were designed for low resolutions and FSAA off. Look at those benchmarks. Except for the UT3 one, they're all at low res. The 8600 was severely crippled by lack of memory bandwidth, although it had some shader power (not enough to run PhysX these days, of course).

For early DX9 titles at better than 1280x1024 the 9700 would easily keep up with a 8500GT and possibly even a 8600GT. It could absolutely push 60 fps in HL2 at 1024x768, 1280x1024 and even 16x12 without FSAA.

In any event, "way faster" is a bit of a stretch. It might be faster in newer games (where it's an ugly, low res slide show anyway) and slower in the old games both cards are able to run acceptably in the first place.

yes I know the 8600gt shortcomings but that has nothing to do with I said. I am not saying the 8600gt is a very good card I am sying that it was much faster with newer drivers and thats a fact. after newer drivers it played every game with max settings and no AA at 1024x768 or 1280x1024 back then except Crysis which could easily be run with a DX9 medium settings and high water at 1024. heck i played around with the 8600gt the other day and it was smooth at 1440x900 with those settings now. anyway back then those same settings in DX10 were okay in Crysis except for the snow and carrier levels which brought plenty of gpus to their back then. I have had the card since it was released so I know pretty much what it can and cant do and yes many of the latest games would kill it. lol

 

[T2k]

Originally posted by: toyota

Originally posted by: alyarb
the 9700 is still faster in D3D but it will not help the CPU decode HD video.

actually the gpu in the ION is about like an 8400gs which way faster than a 9700.

Quite the other way around, 9700 Pro is way faster than that overpriced crap called Ion..

 

[cusideabelincoln]

Originally posted by: v8envy
The 8600GT (and lower) were designed for low resolutions and FSAA off. Look at those benchmarks. Except for the UT3 one, they're all at low res. The 8600 was severely crippled by lack of memory bandwidth, although it had some shader power (not enough to run PhysX these days, of course).

For early DX9 titles at better than 1280x1024 the 9700 would easily keep up with a 8500GT and possibly even a 8600GT. It could absolutely push 60 fps in HL2 at 1024x768, 1280x1024 and even 16x12 without FSAA.

In any event, "way faster" is a bit of a stretch. It might be faster in newer games (where it's an ugly, low res slide show anyway) and slower in the old games both cards are able to run acceptably in the first place.

There's no way the 9700 Pro can keep up with the 8600GT. It might be able to hang with the 8500GT, but not in every scenario.

After using an 8600GT, the lower resolution used the better. 1280x1024 is about as high as the card can go, as that's when huge performance dips occur and drops in IQ are necessary. At 1024x768, I'm actually surprised at how well the 8600GT can play games. It can handle Crysis on medium details fine, although in heavy action it does drop to near unplayable. It can play Far Cry 2 pretty well in DX9 mode with HDR and a mix of medium/high settings. But in either of these two games, once the resolution is increased performance drops off significantly.

 

[Munky]

Originally posted by: Fox5
Aren't the specs of the ion like twice the 9700 pro (twice the pixel pushing power, 16 shaders instead of 8, same clock speed), but with significantly lower memory bandwidth? (1/4 to 1/2?)

Edit:

Ok the specs compare like this:

9300 --------------------- 9700 pro
Core speed (Mhz) 450 (900 shader) 325
Pixel fill rate 3600 Mpixels/sec 2600 Mpixels/sec
Shaders 16 8
Memory bandwidth 6.4GB/s to 12.8GB/s 19.8GB/s


The ion is faster in almost anyway, so unless memory bandwidth is killing it, it should win.
The geforce 6600gt's specs are:
Core speed 500mhz
Pixel fill rate 2000 Mpixels/s /4000 Mtexels
Shaders 8?
Memory bandwidth 16GB/s

The ion looks like it should be faster than the 6600gt too.

The x800xt:
Core speed 500 mhz
Pixel fill rate 8000 Mpixels/s
Shaders 16
Memory bandwidth 32GB/s

If we want to go purely by shader power, the ion probably is close to something like the 7900 Ultra. By memory bandwidth, it's a 9600 pro. By pixel fill rate, it's an X700XT.
By transistor count, it's about a 6800Ultra, and that's about where I'd expect it's performance to lie if it's not too hampered by memory bandwidth.

The cards previous to the 8-series used vec4 shaders, so implying that 16 scalar shaders is faster than 8 vec4 shaders is kinda misleading. An x800xt with 16 vec4 pixel shaders would destroy an 8400gs with 16 scalar shaders.

 

[Schmide]

The idea behind the scalar shader is it will run at a higher rate because it is less complex and you can pack a bunch of them in a small space. In addition only certain operations can pack a vector shader to produce maximum throughput, most of the eye candy operations (HDR, filtering, etc) do not fall into this category. (Think 70-80% maximum throughput)

It generally takes double the scaler shaders and double MHZ to equal a vector shader. Although this is not a concrete relationship, with regards to the above 2 cards, (9300) 16 scaler shader 900mhz is about half the speed of a (x800) 16 vector shader 500mhz card give or take.

 

[Fox5]

Originally posted by: Schmide
The idea behind the scalar shader is it will run at a higher rate because it is less complex and you can pack a bunch of them in a small space. In addition only certain operations can pack a vector shader to produce maximum throughput, most of the eye candy operations (HDR, filtering, etc) do not fall into this category. (Think 70-80% maximum throughput)

It generally takes double the scaler shaders and double MHZ to equal a vector shader. Although this is not a concrete relationship, with regards to the above 2 cards, (9300) 16 scaler shader 900mhz is about half the speed of a (x800) 16 vector shader 500mhz card give or take.

The x800 had 16 pixel shaders (were they still vec4?). I don't know how many vertex shaders it had, but cards around that time had less than 5 (from what I recall, all the cards of that generation had either 3, 4, or 5), but yes, it takes more than 1 general purpose shader to equal the old school vertex shaders.

And the geforce 9300 still has the advantage in fillrate over a 9700 pro, as well as double the number of general purpose to pixel shaders running at a far higher clock speed, so it probably makes up for the vertex shaders as well. Memory bandwidth still seems like its only deficit compared to the 9700 pro.

 

[MODEL3]

325MHz 9700PRO has 4 Vertex shaders units and 8 Pixel shaders units and it delivers 34 Gflops.
450MHz 9300 IGP is based in the 9000 series of GPUs and has 16SPs (clocked at 1.2GHz) and it delivers 39Gflops Single-issue and up to 58Gflops (Dual-issue).

Overall the 9300 has more shading power.
Also the MADDs are 24bit precision in 9700PRO case.

9300 IGP has 4ROPs / 8TUs (450MHz)
9700PRO has 8ROPs / 8TUs (325MHz)

And of cource the 9300 is an IGP but the 9700PRO has dedicated mem (128MB, 19.84GB/s 310MHz 256bit mem. con.)

Also the tests that can be found on the web (if anyone can find anything, lol) would show nothing.

In those tests the 9700PRO will do DX9.0 based rendering (no HDR etc...) but in the newer tests the 9300 IGP will do DX9.1 or DX10 rendering. (not exactly equal...)

The logic thing is for older games the 9700PRO to produce more fps but i guess we can find (i suppose very few) games (2007 and beyond) that probably the 9300 IGP will be a little bit faster. (it depends from the resolution, quality options and AA & AF levels)

I suspect COD4 (which is not the most demanding game) it can be one possible candidate.
(For example:
a Q1 2006 $500 GPU like the 650MHz 7900GTX 16ROPs 51.2GB/s bandwidth
has nearly the same performance (1280X1024 2AA 8AF) with:
a 550MHz 9500 8ROPs 25.6GB/s bandwidth)

 

[AzN]

Originally posted by: dguy6789

Originally posted by: Azn

Originally posted by: toyota
just for AZN http://en.wikipedia.org/wiki/Source_%28game_engine%29


The release of Half-Life 2: Episode One and The Orange Box both introduced new versions of the engine that could not be used to run older games or mods without the developers performing upgrades to code and, in some cases, content. But both times the work required to move from the older version to the newer was significantly less than what one might have come to expect from other engines - where such an action may well not even be possible without total re-development.

Guess what I also played episode 1 on my 9700pro. Absolutely no fps difference between half life 2. You just love to point out gimmick and marketing BS.

There is a gigantic performance difference between Half Life 2 and Episode 2. Stop saying there isn't, you have obviously never played it otherwise you wouldn't be saying that.

I played all half life 2 and episodes fyi

 

[AzN]

Originally posted by: toyota

Originally posted by: Azn

Originally posted by: toyota

Originally posted by: Azn

Originally posted by: toyota
yeah I was waiting on you to start being an ass. you already know that wasnt what i was trying to say back then. should we put the little immature pm you sent me up so people can see how you really act? grow up.

if you dont know that the Source engine was modified for Ep 2 then you dont have a clue about the HL 2 series, the Orange Box for consoles got the full engine update even for HL 2 which supposedly we pc users will get one day. the Black Mesa mod even had to go back and redo stuff to incorporate the newest version of Source engine since so much was improved with Ep 2 version of the engine. get off your lazy ass and look it up yourself.

Actually that's exactly what you said!

I know source engine wasn't modified anything. It was the exact same engine.

http://www.tomshardware.com/ch...e-2-Episode-2,765.html

Notice the benches don't have 9700pro or 9800pro but we have 1600pro that is exactly the same shit as 9700pro with slightly lower memory bandwidth.

x1600pro 27.1fps
8500gt 27fps
6600gt 24.4 fps
8400gs 21.9 fps

yeah you are right. they made no changes to the Source engine for Ep 2. it was all just a dream...:roll:

Yeah you are dreaming. Just because certain web sites have certain fps while another is getting 100fps does not mean that's the actual fps as setup and testing procedures are different. Also that is testing 1280x1024.

So you've been talking about the 8400gs and ion that beating 9700pro. Guess what? IT DOESN'T and we've been trying to tell you all this time.

I still think the 8400gs could beat the 9700pro in newer games. not that it would matter since they would be too slow for most of them anyway.

No it doesn't. R300 had fragmented pipelines so the were never really shader limited cards like how Nvidia cards are.

 

[AzN]

Originally posted by: MODEL3
The logic thing is for older games the 9700PRO to produce more fps but i guess we can find (i suppose very few) games (2007 and beyond) that probably the 9300 IGP will be a little bit faster. (it depends from the resolution, quality options and AA & AF levels)

I suspect COD4 (which is not the most demanding game) it can be one possible candidate.
(For example:
a Q1 2006 $500 GPU like the 650MHz 7900GTX 16ROPs 51.2GB/s bandwidth
has nearly the same performance (1280X1024 2AA 8AF) with:
a 550MHz 9500 8ROPs 25.6GB/s bandwidth)

The thing is Radeon never had a problem with these modern shader bloated games. Nvidia sure did. Their whole 7x00 performing much worse than the same generation of Radeon GPU as time progressed.

Just because Ion is based on a 8400gs core it's still slower than a physical card with actual memory on board.

When you talk of games being shader limited on 9700pro. it goes both ways. Ion is severely bottlenecked by bandwidth that hinders performance severely. Those theoretical fillrate never hitting peak because of bandwidth.

Perhaps you missed the x1600pro that was just another wannabe 9700pro with 128bit bus vs these modern cards.

http://www.tomshardware.com/ch...e-2-Episode-2,765.html

x1600pro 27.1fps
8500gt 27fps
6600gt 24.4 fps
8400gs 21.9 fps

 

[schneiderguy]

Originally posted by: Azn

Originally posted by: dguy6789

Originally posted by: Azn

Originally posted by: toyota
just for AZN http://en.wikipedia.org/wiki/Source_%28game_engine%29


The release of Half-Life 2: Episode One and The Orange Box both introduced new versions of the engine that could not be used to run older games or mods without the developers performing upgrades to code and, in some cases, content. But both times the work required to move from the older version to the newer was significantly less than what one might have come to expect from other engines - where such an action may well not even be possible without total re-development.

Guess what I also played episode 1 on my 9700pro. Absolutely no fps difference between half life 2. You just love to point out gimmick and marketing BS.

There is a gigantic performance difference between Half Life 2 and Episode 2. Stop saying there isn't, you have obviously never played it otherwise you wouldn't be saying that.

I played all half life 2 and episodes fyi

With HDR and the dynamic flashlight shadows on Episode 2 definitely does NOT perform at the same level as the original HL2 did. Not to mention I'm sure it has higher res textures, higher poly models, etc, than the original did.

It would probably be possible to test the difference by setting Episode 2 to all "medium" since that would disable HDR, dynamic shadows, and have lower res textures and models, but the game is so CPU limited on modern cards it probably wouldn't have a performance difference. However, on a slower card like the 9700 these settings would definitely make the game run slower.

 

[MODEL3]

Originally posted by: Azn
The thing is Radeon never had a problem with these modern shader bloated games. Nvidia sure did. Their whole 7x00 performing much worse than the same generation of Radeon GPU as time progressed.

Sure, 7000 based parts usually had worst shader performance in games than X1000 based parts.
This doesn't say anything by itself in the discussion that we are having about Q3 2002 9700PRO.

Originally posted by: Azn
Just because Ion is based on a 8400gs core it's still slower than a physical card with actual memory on board.

Ion is based on a 9000 series not on 8000 series but the performance difference is small anyway (+10% same specs)
But anyway this is not important.

Originally posted by: Azn
When you talk of games being shader limited on 9700pro. it goes both ways. Ion is severely bottlenecked by bandwidth that hinders performance severely. Those theoretical fillrate never hitting peak because of bandwidth.

Yes I agree.
Certainly 9300 IGP is bottlenecked by bandwidth that hinders performance in a large degree. (the degree it depends from the game engine and from the resolution, quality options and AA & AF levels...)
It is obvious...

Originally posted by: Azn
Perhaps you missed the x1600pro that was just another wannabe 9700pro with 128bit bus vs these modern cards.

http://www.tomshardware.com/ch...e-2-Episode-2,765.html

x1600pro 27.1fps
8500gt 27fps
6600gt 24.4 fps
8400gs 21.9 fps

x1600pro was just another wannabe 9700pro with 128bit bus?

lol

x1600pro has 2,3X the shading power of 9700pro and
2,3X the texel fillrate of 9700pro.
-25% the pixel fillrate of 9700pro.
-35% the memory bandwidth of 9700pro.

And the 2 parts are not so easily comparable.

In the link that you provided the X1600pro was only -11% slower
in relation with a Q3 2004 $300 6800GT GPU (350MHz core / 500MHz mem 16ROPs)

6800GT was around twice the speed in relation with the Q3 2002 9700PRO (325MHz core / 310MHz mem 8ROPs)

So (depending on the game engine and from the resolution, quality options) the X1600pro could be even 2X faster than a 9700PRO.

But anyway let's focus on the subject:

I really don't know where you disagree, since i said that the 9700PRO will be faster than 9300 in all the games of 2006 and before
and that for 2007 and beyond we can find i guess (i suppose very few) (that's what i said) games that the 9300 can be faster.

So if i understood your only point was that:

it is not possible to find any game
in any resolution
and in any quality option
that the 9300 will be a little bit faster (that's what i said) than 9700PRO.

Originally posted by: MODEL3
The logic thing is for older games the 9700PRO to produce more fps but i guess we can find (i suppose very few) games (2007 and beyond) that probably the 9300 IGP will be a little bit faster. (it depends from the resolution, quality options and AA & AF levels)