backtop


Print 63 comment(s) - last by Snuffalufagus.. on Jan 27 at 5:35 AM


ATI X1900XT Board Layout

Retail X1900XT Board And Cooler
R580 has been the talk of the town for months as what Radeon X1800 should have been. We got an early sample of the card and put it through the paces

ATI’s Radeon X1800 launch came and went in October with very little fanfare.  The R520 ASIC was hampered by delays and redesigns, and the final product was a far cry from the GeForce 7800GTX (G70) killer.  Ultimately, what killed R520 was the lack of commitment to produce quantity – which in turn made the chip difficult to obtain, and even harder to buy for a competitive price.  NVIDIA made the same mistake with their GeForce 7800GTX 512 chip as well, but this is a look at ATI’s newest, not a history lesson.

Even though the official launch of Radeon X1900 is not until next week, one of our writers in Taiwan was able to over-night a card back to our labs.  There are two standalone versions of Radeon X1900; the Radeon X1900XT and the Radeon X1900XTX.  The X1900XTX is essentially the same card as the X1900XT, but with slightly higher clock speeds.  One tier one AIB told us ATI switched from the “Platinum Edition” to the “XTX” naming due to the low quantity stigma associated with Platinum Edition.  Finally, there is also a Radeon X1900 CF (CrossFire) card expected at launch as well.  This card has the same core clocks as the X1900XT, but replaces some of the outputs with a VHDCI connector for CrossFire support.  ATI’s low end cards support CrossFire directly over the PCIe bus, but X1900 will unfortunately require the bulky cable.  Radeon X1900XT and Radeon X1900XTX come with twin DVI outputs, but a dongle comes with the retail card can convert the DVI signal to VGA instead. 

Radeon X1900, or R580, is in many ways what the R520 should have been before all the die respins.  R580 uses the same memory controller found on R520, with the 512-bit internal ring bus.  Externally, the chip can address 256-bits at a time.  Our Radeon X1900XT came with 512MB DDR3 running at 1.45GHz, but the X1900XTX version comes with a default memory clock of 1.55GHz.  As anticipated, the R580 core features 16 pixel pipelines with 48 pixel shader processors.  R580 is produced on a 90nm process.  Unfortunately, all X1900 series cards are double width cards; the additional width is needed to properly cool the GPU.

ATI has an internal presentation to show the distinction between each of the new cards:

Product Name

X1900 CrossFire

X1900XTX  512MB

X1900XT 512MB

Core Speed

625MHz

650MHz

625MHz

Pixel Shader Processors

48

48

48

Memory Speed

1.45GHz

1.55GHz

1.45GHz

Memory Size

512MB DDR3

512MB DDR3

512MB DDR3

Memory interface

256bit

256bit

256bit

       

TV Output

-

S  Video

S  Video

DVI-I

DVI-I x 1,          VHDCI x 1

DVI X 2

DVI X 2

VIVO (Video In/Video Out)

-

YES

YES

HDTV Support

YES

YES

YES

Shader Model

3

3

3

AVIVO Support

YES

YES

YES

Crossfire Support

Master Card

Crossfire Ready

Crossfire Ready

H.264 Support

YES

YES

YES

We ran our X1900XT against our eVGA GeForce 7800GTX 256MB in a few popular game timedemos.  The cards were both benchmarked on an Opteron 165 workstation with 2GB of PC-3200.  We used the latest NVIDIA driver, but the Catalyst 6.1 driver did not detect the Radeon X1900.  Thus, we used a beta driver from late December.  Our tests are no where near extensive, but they should give you a good idea of what to see during ATI’s launch next week.  All settings are default unless noted elsewhere. 

Game Test

GeForce 7800GTX 256MB

Radeon X1900XT 512MB

FEAR 1600x1200 No AA

48 FPS

61  FPS

FEAR 1920x1200 No AA

41 FPS

 55  FPS

FEAR 1600x1200 4xAA

25 FPS

 45 FPS

FEAR 1920x1200 4xAA

 21 FPS

 37 FPS

COD2 1600x1200 No AA

31 FPS

39 FPS

COD2 1920x1200 No AA

28 FPS

 35 FPS

COD2 1600x1200 4xAA

25 FPS

 35  FPS

COD2 1920x1200 4xAA

22 FPS

 29  FPS

BF2 1600x1200 No AA

61 FPS

63 FPS

BF2 1920x1200 No AA

60 FPS

 60 FPS

BF2 1600x1200 4xAA

53 FPS

57 FPS

BF2 1920x1200 4xAA

46 FPS

51 FPS

ATI has told its partners that the X1900 launch will be an actual product launch; meaning we should see inventory in the stores on the 24th.  This may be true for ATI branded inventory, but AIB partners say it will be several weeks before they have reliable shipments of inventory ready for merchants. 



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Nice numbers!
By Assimilator87 on 1/20/2006 8:44:18 AM , Rating: 2
What's the point of creating lots of pixels per clock then limiting the output by the number of ROP units? It makes no sense to me. The output will only be limited further if G71 has the same number of ROP units, but more pixel pipelines.


RE: Nice numbers!
By hcforde on 1/20/2006 10:38:26 AM , Rating: 2
This is for the graphics the next big thing will be the physics processors -Ageia's PHYSX card. Also the 580 is programmable they have encoded video 10 times faster using this chip rather than the CPU. Yes 10 times not 10%. This chip is more than just the sum of its parts

Thanks


RE: Nice numbers!
By Clauzii on 1/21/2006 2:32:03 AM , Rating: 2
There is a difference in calculating physical pixels and pixels with a lot of eyecandy on.


RE: Nice numbers!
By Thalyn on 1/23/2006 1:30:14 AM , Rating: 2
Consider how long it takes to execute a shader program. It's not going to be one cycle per shader, so having an equal number of SMUs and ROps is going to limit the actual pixel output anyway.

Taking it a step further, most (if not all) shader programs are going to take longer than 3 cycles to complete - so even a 3:1 ratio of SMUs and ROps will be limited by the number of shaders (remembering shaders can be rendered in parallel, so 3 can be running simultaneously on for one ROp path). Even a 10:1 ratio would still leave the performance hindered by the number of SMUs.

Multi-texturing is dead - get over it. Very few, if any, new release titles make heavy use of MT over shaders anymore, so the benefit just won't be there if you increase the number of TMUs or ROps; especially at the cost of SMUs. Since not spending money (and realestate) on TMUs can then mean more SMUs, anyone designing a graphics card these days would be daft to make that kind of trade-off.

Think back to the high-clock 4x2 architecture of the FX5800, then compare it to the 8x1 design of the 9700. That's about how long TMUs have been out of style for.

-Jak

[i]SMU = Shader Map Unit[/i]


"We don't know how to make a $500 computer that's not a piece of junk." -- Apple CEO Steve Jobs











botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki