ATI 4xxx Series Thread

[bryanW1995]

those prices actually make sense because you know that fuad heard a part of a conversation between intelligent people. His limited vocabulary probably caught: rv770..gibberish...blah blah blah...jensen huang...$229... and he just kind of built his little article.

 

[Quiksilver]

I'm just hoping that 30% faster than 9800GTX is claim is with 4xAA and 16xAF at at least a 1600X1200 resolution.

 

[stepone]

With the 48xx series right around the corner the 3850 & 3870's are starting to go into free fall (price wise)! Could be a good time to get a cheapo 3850 in the next 2 weeks or so.
I saw a powecolor 3850 512mb reduced from £95 to £64.99 ($60 saving to you US types) this week & all prices heading south pretty quickly!
Heck most 8600gt's are still more expensive than that in the UK!

This is a good sign that the 4850 is going to be not much more than the old 3850 launch price & if it turns out to be quicker than a 9800gtx then we should be in for some interesting times in the mainstream price segment.

* How sweet would it be if the 3850 256mb becomes the new low-end!

 

[Janooo]

Some confusion: ...among the 800 SP, 480 SP are responsible for unified rendering, 320 SP are responsible for the physical acceleration.

People close to ATI told us that ATI next-generation flagship product RV770 still uses the basic architecture of R600, but the specific structure is not the same as R600/RV670.

In addition, although both using RV770 chip, the memory controller in Radeon HD 4870 graphics card has more power consumption than the memory controller in Radeon HD 4850, because the GDDR5 configuration.

Rumors suggested that RV770 has 480 SP or 800 SP, but we heard RV770 may adopt a unified rendering + separate physical speeding up rendering mode, among the 800 SP, 480 SP are responsible for unified rendering, 320 SP are responsible for the physical acceleration.

Removing physical acceleration, we can see RV770 has 480 SP, if taking the physical acceleration into account, then RV770 has 800 SP built-in.

 

[Sureshot324]

Originally posted by: Cookie Monster
Prices of HD4800 series

So basically, $349 for HD4870, $229 for HD4850 and $529-549 for the X2.

Well that's kind of disappointing. Much more expensive than the 3800 series. Lets hope the performance justifies the price.

 

[bryanW1995]

I think that it's better if rv770 is MORE expensive b/c that means that it will have better performance. we can still get g92/rv670 for budget buys.

 

[Rusin]

Quicksilver:
Nope..Since it will have either four 64-bit channels or eight 32-bit channels there won't be bottleneck since there are enough "routes" for data to get trough at same time.

HD2900 XT was 512-bit monster, but it had eight 64-bit channels. Biggest single memory channel of all time was on Matrox Parhelia 512 which had one 256-bit memory channel. There was huge latency issue hindering performance since data B had to wait for data A to get trough that single pipe..they couldn't go at the same time.

 

[Quiksilver]

Originally posted by: Rusin
Quicksilver:
Nope..Since it will have either four 64-bit channels or eight 32-bit channels there won't be bottleneck since there are enough "routes" for data to get trough at same time.

HD2900 XT was 512-bit monster, but it had eight 64-bit channels. Biggest single memory channel of all time was on Matrox Parhelia 512 which had one 256-bit memory channel. There was huge latency issue hindering performance since data B had to wait for data A to get trough that single pipe..they couldn't go at the same time.

??? You mean their won't be problems or their will be? I can't tell...

 

[NoStateofMind]

Looks like ATi is trying to beat out nVidia:

Launch date scramble, new names for GT200 *updated*

It seems AMD wasn't very pleased with the fact that NVIDIA was trying to launch GT200 before AMD got a chance to present RV770 to the world. The GT200 launch seem to be set for middle of June, the 18th to be more precise, which would have coincided with AMD's launch of RV770. AMD therefore decided to speed up its launch by moving it to the 16th, which doesn't really change much, but now RV770 will be first on the market. And yes, GT200 has been renamed from the 9900 series to GTX 200 series. The GeForce 9900GTX will now be GTX 280 and 9900GTS will become GTX 260. We're not certain what NVIDIA will do to the already existing 9800GTX at this time.

The only problem is that the chips are targeting two different price and market segments. RV770 is, and has always been, a mid-range chip, and GT200 the extreme high-end. AMD will compete with GT200 with its dual-chip 4870X2 card, while NVIDIA will compete with RV770 by stripping down GT200 into several more or less crippled versions. Just like it did with G92.

 

[dv8silencer]

interesante!

 

[bryanW1995]

Originally posted by: PC Surgeon
Looks like ATi is trying to beat out nVidia:

Launch date scramble, new names for GT200 *updated*

It seems AMD wasn't very pleased with the fact that NVIDIA was trying to launch GT200 before AMD got a chance to present RV770 to the world. The GT200 launch seem to be set for middle of June, the 18th to be more precise, which would have coincided with AMD's launch of RV770. AMD therefore decided to speed up its launch by moving it to the 16th, which doesn't really change much, but now RV770 will be first on the market. And yes, GT200 has been renamed from the 9900 series to GTX 200 series. The GeForce 9900GTX will now be GTX 280 and 9900GTS will become GTX 260. We're not certain what NVIDIA will do to the already existing 9800GTX at this time.

The only problem is that the chips are targeting two different price and market segments. RV770 is, and has always been, a mid-range chip, and GT200 the extreme high-end. AMD will compete with GT200 with its dual-chip 4870X2 card, while NVIDIA will compete with RV770 by stripping down GT200 into several more or less crippled versions. Just like it did with G92.

I thought that nvidia was going to compete against 4xxx with g92-based cards, at least for now...

 

[taltamir]

people keep on going on and on and on about "maybe ATI can afford to sell those for much less then they are worth".
Are you out of your minds? forget the arguments of weather they can or cannot afford it. It is NOT about market share, it is about MAKING MONEY!

Economics rule #1: "Something is worth as much as people are willing to pay for it".
AMD is going to price these to maximize their PROFIT.

The only exception I can see is if AMD is feeling desperate with nvidia's incorporation of physX. So they offer a significally cheaper chard (perhaps at less then what it costs nvidia to produce, since AMD uses a better process) as a strategy of demolishing nvidia's market share to prevent physX from catching on while AMD is developing their own implementation (nvidia opened the standard, but is holding on to the original source code. So they are years ahead in coding it).

This means a very interesting choice, do you go with physX, or do you go with a much cheaper and faster card and give physX up...
But that will be the ONLY case where AMD is vastly underpricing their hardware. If that is not the case, then "the specs show it should kill nvidia's 500$ yet it's priced at 200$" just means that it is actually only performs as well as a 200$ card.

 

[Extelleron]

Originally posted by: taltamir
people keep on going on and on and on about "maybe ATI can afford to sell those for much less then they are worth".
Are you out of your minds? forget the arguments of weather they can or cannot afford it. It is NOT about market share, it is about MAKING MONEY!

Economics rule #1: "Something is worth as much as people are willing to pay for it".
AMD is going to price these to maximize their PROFIT.

The only exception I can see is if AMD is feeling desperate with nvidia's incorporation of physX. So they offer a significally cheaper chard (perhaps at less then what it costs nvidia to produce, since AMD uses a better process) as a strategy of demolishing nvidia's market share to prevent physX from catching on while AMD is developing their own implementation (nvidia opened the standard, but is holding on to the original source code. So they are years ahead in coding it).

This means a very interesting choice, do you go with physX, or do you go with a much cheaper and faster card and give physX up...
But that will be the ONLY case where AMD is vastly underpricing their hardware. If that is not the case, then "the specs show it should kill nvidia's 500$ yet it's priced at 200$" just means that it is actually only performs as well as a 200$ card.

What makes you think PhysX is important at all?

From what I see graphics-accelerated physics is as stupid as hardware-accelerated physics was; it's just not practical. Games don't need dedicated physics hardware, it can be performed on CPUs. With quad-core and soon octal-core CPUs, there is plenty of CPU power that is currently going unutilized that could be put towards physics. And with games being GPU limited already, what makes anyone think that shading resources will be able to be diverted from graphics processing?

 

[taltamir]

no it can not. intel nehalem demostration of physx got 15fps on an octa core... an 8800GTS 512 got 300fps (not the same code, but the same end result), and nvidia says the G200 would get twice that... 600fps.

You already have phyx CPU acceleration, but even the fastest quad core onthe market today chokes when trying to render cloth and other such intensive tasks resulting in horrible FPS. CPUs simple do not have the power.
A GPU physx will allow you to play at playable rates.

Also, for intensive calculations there is tesla, nvidia GPU's without a DVI/VGA port, it just runs C code with Cuda and is used where intensive calculations are needed, it is much more powerful then any quad core CPU.

The only question is, will there be enough games to justify it, (and will they be good enough games).

The key difference here is, you don't have to buy a whole new 100+$ expansion card that works with only a few games, you simply have to buy a different companys video card and know that it does physX as well..

 

[Extelleron]

Originally posted by: taltamir
no it can not. intel nehalem demostration of physx got 15fps on an octa core... an 8800GTS 512 got 300fps (not the same code, but the same end result), and nvidia says the G200 would get twice that... 600fps.

You already have phyx CPU acceleration, but even the fastest quad core onthe market today chokes when trying to render cloth and other such intensive tasks resulting in horrible FPS. CPUs simple do not have the power.
A GPU physx will allow you to play at playable rates.

Also, for intensive calculations there is tesla, nvidia GPU's without a DVI/VGA port, it just runs C code with Cuda and is used where intensive calculations are needed, it is much more powerful then any quad core CPU.

The only question is, will there be enough games to justify it, (and will they be good enough games).

The key difference here is, you don't have to buy a whole new 100+$ expansion card that works with only a few games, you simply have to buy a different companys video card and know that it does physX as well..

How do you plan on diverting shading resources needed for graphics processing to physics processing without reducing performance?

And I'm sure code optimized for an Octal-core processor would fare well against a modern video card.

 

[taltamir]

you are wrong about the octa-core. Nehalem octa core simply can not compete in raw math capacity, it is meant to do entirely different tasks then a GPU, which is basically a number cruncher.
And as for diverting shading resources... you simply do, shaders are disproportionally fast on nvidia's latest gen cards compared to the rest of the GPU, so there are shaders to spare. regardless, you simply take a little hit to shading to gain a huge benfit in physx. Giving you overall better performance. It is a matter of balancing what you do with the GPU. Just like how you can lower AA and increase the resolution (or vice versa). You balance the various aspects, sacrificing "expensive" operations that give a minimal benefits for "cheap" operations that give significant benefits.
You could also always SLI.

 

[bryanW1995]

can't you just run a 2nd gpu as a physX ppu?

 

[ronnn]

Originally posted by: bryanW1995
can't you just run a 2nd gpu as a physX ppu?

For that you need games coded for it. Actually we need games coded for a computer.

 

[superbooga]

Originally posted by: Extelleron
How do you plan on diverting shading resources needed for graphics processing to physics processing without reducing performance?

Just think of physics as another graphical option. When you set shader and object quality of a game from Medium to High, you increasing the amount of shader resources needed. How is physics any different?

 

[taltamir]

well, considering that a single 8800 GPU has several times the performance of an ageia PPU, it will be a while until such a thing is required, at first only a few percent will be dedicated to PPU while the rest double up on graphics.
Personally I predict that in a couple of years you would be running multiple cards for phisics, and a few percent will be dedicated to graphics...

Originally posted by: superbooga

Originally posted by: Extelleron
How do you plan on diverting shading resources needed for graphics processing to physics processing without reducing performance?

Just think of physics as another graphical option. When you set shader and object quality of a game from Medium to High, you increasing the amount of shader resources needed. How is physics any different?

That is exactly it... AA, Shadows, View range, Texture sizes, Filtering (of the non AA kind), Light sources, Reflections, viewable distance, etc etc etc. In many games you can increase and decrease those various settings, and they all compete for GPU resources. With physx you just add another calculating competing for said resources. Turning it on or off (or to different quality levels) would be done to balance visual benefits with overall FPS.

 

[AshPhoenix]

RV770XT fan gets pictured

 

[AshPhoenix]

RV770PRO engineering sample works at 625MHz

Product part to be faster

The RV770PRO is a GDDR3 version of the RV770 chip and we?ve found out that the samples out in the wild are working at 625MHz. The samples are based on revision a12, which is usually ATI?s production revision.

The sample board was equipped with GDDR3 memory clocked at 2,000MHz and we?ve heard that GDDR5 samples are expected any day now.

Both RV770XT and RV770PRO are scheduled for a June 16th launch, just two days before the Geforce GTX 280 / 260 launch.

 

[biostud]

Originally posted by: taltamir
well, considering that a single 8800 GPU has several times the performance of an ageia PPU, it will be a while until such a thing is required, at first only a few percent will be dedicated to PPU while the rest double up on graphics.
Personally I predict that in a couple of years you would be running multiple cards for phisics, and a few percent will be dedicated to graphics...

Originally posted by: superbooga

Originally posted by: Extelleron
How do you plan on diverting shading resources needed for graphics processing to physics processing without reducing performance?

Just think of physics as another graphical option. When you set shader and object quality of a game from Medium to High, you increasing the amount of shader resources needed. How is physics any different?

That is exactly it... AA, Shadows, View range, Texture sizes, Filtering (of the non AA kind), Light sources, Reflections, viewable distance, etc etc etc. In many games you can increase and decrease those various settings, and they all compete for GPU resources. With physx you just add another calculating competing for said resources. Turning it on or off (or to different quality levels) would be done to balance visual benefits with overall FPS.

The thing is most people prefer good loking graphics compared to intense physics calculation. The problem is that even though a videocard is better for physics than a CPU, then when you add more debris, destructable environment etc. it also need more GPU power. As it is now it seems that if you make an environment that requires a PPU then the GPU will be to slow to render playable frame rates. However, if you use the amount of physics calculations a CPU can proces, then the GPU can be used solely for graphics.

Of course gaming could take the route of better physics, and focus less on graphics, but somehow I don't see that happen. The question you have to ask is: When you ad a 2nd GPU what will give me most fps, if it works as a GPU or a PPU?

I think we are a long way to the answer beeing a PPU.

On the other hand the CPU has spare core's to run physics. They may not be as advanced as physX, but it's not like you have a lot of GPU power available.

 

[Continuity28]

Originally posted by: Extelleron
How do you plan on diverting shading resources needed for graphics processing to physics processing without reducing performance?

And I'm sure code optimized for an Octal-core processor would fare well against a modern video card.

The point is, running such physics calculations on the underloaded CPU would result in less FPS than running it on the GPU, even when the GPU is already fully loaded with tasks. It's because the CPU is very slow at this type of task. When the GPU does it, you may drop in FPS from 65 to 62, or hell even for argument, 45. When the CPU does it, you may drop from 65 to 10 because the CPU can't do that level of physics any faster than that.

You can't really optimize the code for the CPU, the issue starts at the hardware level, the hardware isn't built for it. There are things that the modern GPU can do better than a CPU, and things a modern CPU can do better than a GPU, and that's why they have their respective uses and roles in a system. We don't run our operating systems and programs on a GPU, we don't render modern games on our CPU and expect speed. They are drastically different pieces of hardware.

It's like comparing an elephant and horse as modes of transportation. The elephant can carry more for sure, but it's not as fast regardless of it's capacity. If you think the horse will always be faster, try loading it as much as the elephant and see how fast it runs. They aren't built the same way, and won't be used in the same ways.

I don't care if anyone hates my analogy.

 

[Keysplayr]

Actually, I thought that was a pretty good analogy.