ATI 3800 Specs Released, Much Faster Than NVIDIA

http://ati.amd.com/products/mobilityradeonhd3800/specs.html

It looks like ATI managed to fit 320 stream processors (compare that to 112 in the 9800M GTX) and a 256-bit memory bus into a 55mn notebook video card. I'm sure that this thing will dominate the 9800M GT/8800M GTX and the 9800M GTX in benchmarks. I really hope that we see this in a lot of notebooks, especially 15" notebooks. According to ATI, you can run up to four of these. Imagine four video cards in a laptop... I wonder which notebook manufacturers tries it first ?

 

Well you can´t compare by stream processor numbers. It doesn´t work that way. It will not dominate but probably be equal/close to 8800m GTX. But it´s nice to see ATI back in the game again in the notebook market with a high end GPU.

 

I figure since it has three times the stream processors with similar specs for everything else, it'll have to be at least a little faster.

 

Nvidia's gpu's generally use 5-dimensional spu's and ATI use single dimensional one's. It basically works out at 5x the performance for nvidia.
Clock speed and "effeciency" is a big part of it aswell though

 

what do you think about 2 3870's in crossfire in the ocz whitebook? how much better would those be compared to 1 9800M GTX?

 

Go for the fastest single card config you can afford. Sli\crossfire is a mixed bag and not worth it imo.

 

Hmmmm.....Letz c

We need real benchmark numbers (game benchs)

 

well alot of games can now take advantage of SLI and crossfire, so its always better now. i can choose between 2 3870 in crossfire or 1 8800M GTX, what would you say?

 

Wouldn't be fair to make judgments based on the number of stream processors. For instance, the Desktop 3870 has more stream processors than the 8800 GTX, but the 8800 GTX outperforms it in an overwhelming number of gaming benchmarks. Also, the SPs in the ATi cards run at half the clock than the NVidia SPs.

However, on average, CrossfireX does scale better than SLi (and offers less hassle from what I've heard), which might be a plus for the 3800 cards.

 

As already mentioned, Nvidia and Ati count their stream processors differently with Ati typically using more stream processors in their specifications. Any benchmarks available from retail machines yet?

 

Yes. Never compare stream processors between Nvidia and ATI, they count them differently. The new 4870s have, what, something like 800 stream processors compared to the GTX 280s 240 stream processors.

The Nvidia GTX280 is certainly faster in many cases, so as you can see, comparing stream processors, frequency, pretty much ANYTHING besides actual FPS numbers is useless. Even the width of the interface will become irrelevant as GDDR5 makes its way into notebooks.

Otherwise, FINALLY. How long have we been waiting for this?

 

ATI's cards have been fantastic on paper for a while, but it never quite works out that way.

 

well its really hard to compare the 9800m gtx to the ATI 3800s since the ATI cards are a generation behind. Its more comparable to Nvidia 8800s and it was very competitive with those cards, mainly because of its price and better crossfire scalability.

ATI has the 4800s out now, and it dominates 9800ms.

 

9800m GTX isnt even out, and the 9000 series except for that card are just a reworked, and rebranded line, so they are not a generation behind at all.

 

Technically they are not, but people don't care about that

Desktop HD4850 easily beat 9800GTX card, so i guess same will happen with mobile parts

 

I do believe the 48xx desptop series looked good pn paper and is now leading the desktop market so I'm not sure if you've paid any attention sinse the hd 2xxx series which is when this was true.

 

I think you have that mixed up. nVidia's stream processors process independent instructions so if you have 128 SPs you always have 128 instructions being processed. In contrast, ATI groups there SPs in groups of 5 and requires heavy driver optimization so fill all 5 SPs in a group. So in the worst case, if only 1 SP in a group is filled ATI's 320 SPs perform like 64 of nVidia's SPs. ATI has probably got average utilization up to 3 SPs out of 5 so 320 ATI SPs probably perform like 192 of nVidia's SPs, but as other's mentioned nVidia clocks their SPs higher and there are other architectural differences. Which is why the desktop 3870 can't outperform the 8800GT even though the desktop 3870 has 320 SPs and the desktop 8800GT has 112 SPs.

http://www.anandtech.com/video/showdoc.aspx?i=3341&p=13

The new 55nm 9800GTX+ that nVidia released specifically to combat the HD4850 can actually trade blows quite successfully with the HD4850. If ATI's high-end mobile versions of the 4800 series is based around the desktop HD4850 and nVidia manages to release 55nm mobile refreshes based around the 9800GTX+ then things are actually pretty competitive. And if ATI's mobile parts end up winning in raw performance, nVidia can always compete on price seeing that the 9800 series is now mature and the 55nm shrinks will be even cheaper for them.

 

ATI has the problem that thier cards are bottlenecked by other parts of the card. There are many shader units, but not enough ROPs or TMUs.

The ATI cards have the rendering horsepower but were bottlenecked. ATI realized this with the HD 4800 series when they doubled the ROPs and TMUs.

The fact here is that there are alot more factors that determine a video cards performance than just the memory speed, and number of unified shaders.

 

crossfire can also have support for two different cards

 

I was expecting Something over 500 SPs, with those specs, i think the ATI card will lose, check out the desktop parts... the newest ATI card has 900SP or something.. and still could be beat by much lower SP card

 

Well, technically the shader power to texture power bottleneck still exists since the ratio of stream processors to texture units has remained the same at 4:1. The number of ROPs is completely unchanged holding at 16 as has been since the X800. It's just that the number of stream processors has increased 2.5 times to 800 and the TMUs correspondingly increased 2.5 times to 40, so the raw TMU number is high, but there isn't anymore texture power compared to shading power than the previous generation.

 

Unless a game is completely shader dependant like Crysis or Oblivion, then the SP's are not going to be as large of a factor, mem bandwidth is still the biggest bottleneck and no one can seem to get that right. the 2900XT had a 512bit bus, but it sucked as a card, and potentially could have destroyed the 8800GTX had it had proper driver support.