AnandTech's R520, RV530 and RV515 Pre-NDA Extravaganza
by Kristopher Kubicki on September 13, 2005 9:24 PM EST- Posted in
- GPUs
The part that everyone wants to hear about is, of course, the Radeon X1800 based on ATI's long awaited R520 GPU. Due for introduction later this quarter, the 90nm R520 will be a 16-pipe, 16-shader processor design with a number of different SKUs based on the GPU. Internal ATI documentation specifically claims that the R520 series will ship at launch, just as NVIDIA's 7800GTX and 7800GT series shipped and launched on the same dates.
ATI R520 Roadmap and Pricing
|
||||
Card | Pipes | Std Core Clock | Std Memory | MSRP |
X1800 XT | 16 | 600MHz | 700MHz 512MB GDDR3 | $599 |
X1800 XL | 16 | 550MHz | 625MHz 512MB GDDR3 | $499 |
X1800 Pro | 16 | 500MHz | 500MHz 256MB GDDR3 | $449 |
X1800 LE | 12 | 450MHz | 450MHZ 256MB GDDR3 | $349 |
Common features to all R520 based boards include the new 90nm lead free manufacturing process, a Xilleon based TV encoder, SM3.0, H.264 decode acceleration and CrossFire support. Also expect to see HDTV options for all 90nm ATI cards in the near future, although they may be limited to the All In Wonder series for R520.
At the top end is the Radeon X1800 XT; this 16-pipe R520 will feature a 600MHz core clock, with a 256-bit memory bus connected to 512MB of GDDR3 memory clocked at 700MHz. The 600MHz core clock will give it a lower fill rate than the GeForce 7800 GTX (24-pipes at 430MHz), while the 700MHz memory clock will give it more memory bandwidth than the stock GTX (600MHz). Much like the GTX, the X1800 XT will be priced at $599. The X1800 XT will feature two DVI outputs with HDCP support. The lower fillrate seems alarming at first, but consider several factors. First of all, ATI's traditional core design can do "more" per clock cycle (at least on the R420 design) than NVIDIA. Secondly, R520 has a lot of little tweaks including hardware asissted H.264 decoding. Just last week, we also received details about ATI's revamped memory controller which operates on an internal 512-bit ring bus. There is a lot to speculate about performance, but even with similar fill rates as NVIDIA, there is a strong possibility that other workings in R520 will differentiate the card on a real world performance level.
Next up is the Radeon X1800 XL, which is positioned between the
GeForce 7800 GTX and the 7800 GT. The XL drops the core clock down to
550MHz, and the memory clock down to 625MHz. Other than the lower clock
speeds, the XL is identical to the XT, meaning it still has 512MB of
GDDR3 memory connected to a 256-bit memory bus. The X1800 XL will be
priced at $499. Both the X1800 XT and X1800 XL appear to be dual-slot
designs from previous roadmaps and existing box art. The roadmap also details that there will be HDCP support for the X1800 XL
and X1800 XT via Texas Instrument's TFP513PAP DVI transmitter.
Priced at $449, we have the X1800 Pro, once more a 16-pipe R520 design
but this time the core runs at 500MHz. The Radeon X1800 Pro only has
256MB of memory, also running at 500MHz, but still retains the same
256-bit memory bus. What is interesting about the Radeon X1800 Pro is
that its fill rate and memory bandwidth appear to be identical to that
of NVIDIA's GeForce 7800GT; coincidentally, so does its price.
The reference design for the X1800 Pro features a single VGA and a
single DVI connector, with no HDCP support.
The last member of the R520 family is the Radeon X1800 LE, which disables four of the pipelines of the R520 taking it down to a 12-pipe design. The LE runs at 450MHz with 256MB of 450MHz GDDR3 memory. Once again we're dealing with a 256-bit memory bus, and this time a $349 price tag. The outputs are identical to the X1800 Pro. Both the Pro and LE cards are single slot cooling design, thanks to their lower running clock speeds.
According to our roadmaps, it looks like ATI will abandon the "vanilla" nomenclature for future products. For example, instead of a plain X1800, instead we will get an X1800 LE. Likewise, on our previous roadmaps components that were named with the non-XT non-LE non-Pro non-XL name will thus become "LE" parts. Certainly a good move on ATI's behalf, as "vanilla" X800 cards are hard enough to explain to readers.
The roadmap also refers to R580, and that the card is working in-house at the moment. R580 is essentailly a clock ramp and pipe ramp of R520, but both of those details have not been disclosed yet (even to AIBs). Unforunately, the R580 will not ship at the same time as R520.
65 Comments
View All Comments
yacoub - Thursday, September 15, 2005 - link
And me still waiting for my four-speed, dual-quad, positraction 409...JarredWalton - Wednesday, September 14, 2005 - link
I asked Kris this exact question. The answer: we don't know. The roadmaps/PDFs don't say anything about vertex pipelines. However, consider this:G70: 8 vertex @ 430 MHz / 4 = 860 MV/s
R520: 6 vertex @ 600 MHz / 4 = 900 MV/s
Basically, with the higher clock speed there's no point in having more than 6 vertex pipelines. With R580, if they move to 24 pixel pipelines, it would make more sense to go to 8 vertex pipelines. 32 pixel pipes would probably need 10 vertex pipes. Then there's the whole "unified architecture" that we're moving towards.
Anyway, the main point is that I have yet to see anything officially stating that R520 has 6, 8, 10, or whatever pipes. Everything is pretty much a guess, and Ocham's Razor suggests that if 16/6 was good for last generation, and R520 has 16 pixel pipes, it probably has 6 vertex pipes again. :p
Questar - Wednesday, September 14, 2005 - link
What if the 16 pipe card performs like a 24 pipe 7800?Is an eight cylinder engine always better than a six?
jkostans - Wednesday, September 14, 2005 - link
Exactly, this is a new architecture we're talking about. Chances are it's a good deal more efficient than the last generation of cards.nserra - Wednesday, September 14, 2005 - link
"First of all, ATI's traditional core design can do "more" per clock cycle (at least on the R3xx design) than NVIDIA."Ati 9700 with just 275Mhz core speed, and ONLY 270Mhz(540Mhz) memory speed killed any card, even the ones that worked at 500Mhz core and 500Mhz(1000Mhz) memory speed (nvidia 5800).
Put all these new cards at 275Mhz (memory and core) if possible (under clock them) to see who does more work.
I don’t think the phrase is correct for the R4xx design since it as higher memory and core speed than nvidia 6xxx and 7xxx.
arturnow - Wednesday, September 14, 2005 - link
"First of all, ATI's traditional core design can do "more" per clock cycle" - I have to disagree. X850XT has higher fill rate and memory bandwidth then GeForce 7800GT but slower in most games...Griswold - Wednesday, September 14, 2005 - link
The X850XT also has 16 pipes vs. the 20 pipes of the 7800GT. And the GT is only faster in "most" games, not all and then also not by that much. So, all in all, what they said sounds right.arturnow - Wednesday, September 14, 2005 - link
of course not. Who cares if it's 20 pipelines oraz 16 pipline. It all depends on core clock and pipline. X850XT PE is 540/1180 MHz, 7800GT 400/1000MHz theoretical Radeon is faster but in games GeForce prevail :-]Griswold - Wednesday, September 14, 2005 - link
I dont think you understand what you're talking there. The 850XT PE is faster in a few games under certain resolutions with AA/AF cranked up. That is due to clock speed and weaker AA/AF modi. But in most games under normal conditions, the 7800GT is faster - due to the 4 more pipes...Jep4444 - Wednesday, September 14, 2005 - link
the pipeline performance is theoretically negated by the higher clock speed so you are actually wrongthe reason the X850XT PE is slower is becuase its based off older technology and hasn't had any improvements made to its effeciency from the Radeon 9700 days which is why GeForce 6800 cards operate at lower clock speeds for the same performance