RV570 = X1900 GTO RV560 = X1700

[cleeve]

So there's the result of the name game:

http://www.xbitlabs.com/news/video/display/20060405200748.html

I didn't think they'd go with a '50' at the end... they like to reserve that for OEM stuff most of the time I think.

Still, not sure the X1900 GTO will outperform an X1800 XT...

 

[Bluefinger]

Most likely it will, but I suspect not by much. It will still have more pixel shader processors, so it is geared up on taking on shader-heavy games like FEAR and Oblivion, however, nothing has been said about its clock speeds, so we will just have to wait until ATI make it official and give out all the details we are craving for.

 

[bluntside]

whow. a x1900 gto? Is this a card thats between the x1800xt and the x1900xt? Im betting that the x1900gto will outperform the x1800xt by pretty much, if it has more pixel shading units and higher clock speeds, the definetly.
Have NE of U heard of the upcomming x1950xt card? Its supposed to be a much coller and quieter card than ne of the other series and have insane clock speeds..
Thats crazy 8O

 

[HardWareBoss]

No Direct X10 Graphic cards this year. Why no Vista.

 

[cleeve]

Well, the issue is kid of complex:

X1900 GTO
12 Pipelines
36 Shader Units
128-bit memory interface

X1800 XT
16 pipelines
16 Shader units
256-bit memory interface


So the X1900 GTO has 20 more shaders than the X1800 XT, but the X1800 XT has a 256-bit memory bus and 4 more pipelines. It should be a close race, but like you guys said it'll all come down to clockspeeds.

For comparison, the X1700 will have:
8 Pipelines
24 Shader Units
128-bit memory interface

 

[bluntside]

Im sure that you can unlock the pipelines in the x1900gto, perhaps rivatuner would work??

 

[sojrner]

well, i think that 128 bit memory interface is a big hangnail on it too.

less bandwidth cannot be made up for w/ just higher speeds.

will be interesting to see what happens...

 

[Ian]

Yeah, however look at how well the 7600 GT is managing with 128 bit memory bus.

 

[sojrner]

true, but there is "manage" and then there is "compete"... yes, the 7600 does very well; in it's class.

I was just pointing out that the smaller interface may show that the 1900gto is in a lesser class than the 1800xt.

But we will have to see when the benches roll in.

 

[TheGreatGrapeApe]

Fall!?!

Damn-it! :x

Well I guess the Merom will be out by then an BR/HD-DVD and HDCP should be figured out.

 

[cleeve]

Im sure that you can unlock the pipelines in the x1900gto, perhaps rivatuner would work??

Unfortunately, no.

The X1900 GTO will be a whole new die, the RV570... not a crippled X1900 XT, R580.

There are no pipes to unlock on the RV570.

 

[ltcommander_data]

In this case, I think X-BitLabs is incorrect on the naming. The Inquirer posted a story four days ago, and it's numbering system makes much more sense. The RV570 will come in an X1700XT version and a X1700XL version while the RV560 will be labeled the X1700Pro. This would be consistant with slides that HKEPC have shown where there are 2 planned RV570 models and only 1 planned RV560 model.

http://www.theinquirer.net/?article=30687

I believe that the X1900GTO is a real product that will also have 12 pipelines and 36 pixel shaders but will be based on rejected R580 dies rather than the RV570. It'll probably replace the X1800GTO and hold out until the RV570 arrives since that won't be until Q3. There's also supposed to be a X1900XL in the works which has the full complement of pipelines and shaders but operates at frequencies between 500MHz-560MHz.

http://www.theinquirer.net/?article=30714

What I find unbelievable is the number of cores that ATI will soon have. They include the RV505 80nm TSMC, the RV515 90nm TSMC, the RV516 90nm UMC (The Inquirer reports it as a 80nm TSMC part), the R520 90nm TSMC, the RV530 90nm TSMC, the RV535 80nm TSMC, the RV536 90nm UMC, the RV560 80nm TSMC, the RV570 80nm TSMC (although X-Bit didn't say this themselves), the R580 90nm TSMC, and the R580+ 90nm TSMC. That's 11 cores. With the various GT, GTO, Pro, XL, XT, XTX, etc. configurations I'd hate to see how many models they are planning to put on the shelves. The previous R4xx generation will still be around too.

What I was most disappointed with was the R580+ turning out to be a 90nm part instead of a 80nm part. At least they're going to be redesigning it. The story from X-Bit Labs only talks about the memory controller, but an article by The Inquirer seems to indicate that it'll be a more complete entire core redesign and optimization like the transition between the G70 and G71. I guess even without going to 80nm, they can still reduce die size and power by cutting transistors. The Inquirer was mentioning larger dies for higher clock speeds, but I really don't see how they are a requirement for each other. I don't think ATI will be adding any more pixel shaders or pipelines given that the difference doesn't even justify a number increment only a +.

http://www.theinquirer.net/?article=30775

From the first link I posted, I also noticed that while the RV570 will be receiving 1 more vertex shader to 6 from the RV530, the RV560 seems to lose 1 and will only have 4. The RV560 would have probably benefited from the doubling of the pixel pipelines more it the number of vertex shaders wasn't reduced.

As a bit of detail, X-Bit Labs appears to be wrong in their claim that current X1300 don't support 32-bit memory controllers and the RV516 will add this feature. According to ATI, the current X1300 do support a 32-bit memory bus although perhaps no manufacturer is shipping cards with such as small bus, which I guess is a good thing. I believe the 32-bit bus is mainly for the Hypermemory cards anyways.

 

[TheGreatGrapeApe]

In this case, I think X-BitLabs is incorrect on the naming. The Inquirer posted a story four days ago, and it's numbering system makes much more sense. The RV570 will come in an X1700XT version and a X1700XL version while the RV560 will be labeled the X1700Pro. This would be consistant with slides that HKEPC have shown where there are 2 planned RV570 models and only 1 planned RV560 model.

I trust Xbits sources in this case. It also makes more sense IMO, even though for some it's non-sensical. IT reminds me of the X800XL.

What I find unbelievable is the number of cores that ATI will soon have... the R520 90nm TSMC,

Hasn't the R520 already been scheduled for removal next month? I rmemebr reading something like that.

That's 11 cores. With the various GT, GTO, Pro, XL, XT, XTX, etc. configurations I'd hate to see how many models they are planning to put on the shelves.

True it will get messy, but think you have the X1800 and X1900 to do all those naming conventions whereas last ime it was X8xx be it X800 or X850.

The previous R4xx generation will still be around too.

Yeah but not for long, they aren't even economical to produce.

What I was most disappointed with was the R580+ turning out to be a 90nm part instead of a 80nm part.

What would the motivation be? They got their second 90nm part to market before nV go their first. It's like the R9700->9800 era all over again where everyone was expecting ATi to migrate it over to 130nm like the R9600, and it never happened. Just ask Crash, he wasn't too happy and quite vocal about it, as was Eden and Cleeve IIRC.

At least they're going to be redesigning it. The story from X-Bit Labs only talks about the memory controller, but an article by The Inquirer seems to indicate that it'll be a more complete entire core redesign and optimization like the transition between the G70 and G71. I guess even without going to 80nm, they can still reduce die size and power by cutting transistors.

And by leveraging the X-over technology they aquired at end of 2004.

The Inquirer was mentioning larger dies for higher clock speeds, but I really don't see how they are a requirement for each other.

Well the only thing I can think of is surface area for the HSF to help disipate heat. The small/tighter the package the more heat must be disipated over a smaller area. That's the only thing that makes sense to me, but it does go contrary to trying to get small chips for cost/chip factors.

I don't think ATI will be adding any more pixel shaders or pipelines given that the difference doesn't even justify a number increment only a +

No but you may see them add vertex engines, but that's unlikely too. ROPs would be nice, but even less likely.

From the first link I posted, I also noticed that while the RV570 will be receiving 1 more vertex shader to 6 from the RV530, the RV560 seems to lose 1 and will only have 4. The RV560 would have probably benefited from the doubling of the pixel pipelines more it the number of vertex shaders wasn't reduced.

True, I';m not covinced yet that the pixel load will always be THAT much more than the vertex load, but unlike the pixel load, the vertex load can be shared by the CPU far more efficiently, and maybe more so with dual/quad core CPUs.

As a bit of detail, X-Bit Labs appears to be wrong in their claim that current X1300 don't support 32-bit memory controllers and the RV516 will add this feature. According to ATI, the current X1300 do support a 32-bit memory bus although perhaps no manufacturer is shipping cards with such as small bus, which I guess is a good thing. I believe the 32-bit bus is mainly for the Hypermemory cards anyways.

Yeah I'm not sure, it may be PER chip, which would allow for larger size smaller number of chips to make-up the board. The way the ringbus works it parcels it out, and perhaps 32 bit wasn't supported for each chip.

I don't know, guess we'll see.

 

[nottheking]

Bah, this is becoming perhaps even more confusing than the whole X800/X850 mess. But at least it will be much more clear once the cards are actually out.

I'm honestly quite intrigued to hear about the RV560, as before, I had only heard of an RV570. I wonder what it might be? Having two separate chips for the X1700 series makes little sense to me, just as would making one of them for the X1900GTO. The numbering scheme implies that it's not, say, a new low-end chip, like an X1400. Then what could the separate chips be for?

I'm crossing my fingers (but not too tightly) that it might be that the RV570 might finally be the chip that brings the 256-bit memory interface to the mid-range market. ATi faked people out with the X1600's "ringbus" (and most people STILL don't understand it ) and nVidia dragged their feet with the 7600GT's memory configuration. (which, to its credit, still performed quite admirably in spite of its 128-bit interface)

And if it has 8 TMUs, 8 ROPs, and 24 PSUs... How will it handle compared to the X1800 series? I doubt that it might actually best it in real-world tests, though it might come close in some titles, such as the shader-hogging Oblivion.

Then that brings us to the R580+. Personally, I feel that we'll see anything changed aside from the memory controller. POSSIBLY, the threading arbiter might change some, or some tweaks to the general layout and processing, (lessons learned from before) but largely, I'd expect to see it as simply an X1900 with faster memory, and likely higher clock speeds, due to some possible trimming of transistors, much like the R423->R480 move, and more recently the G70->G71 move.

I'd still, of course, love to see a 512-bit memory interface, but the number of pins required on the GPU for such probably means we're quite a ways from that. Then again, it could possibly be worked around, and the benefits could be worthwhile; pounding the opposition cards into dust while using cheaper memory. And I'd just like to see cards finally pass the 100GB/sec line, which will likely never come to pass unless they do use a 512-bit interface.

(oh, and finally got to post #100, so I can finally get my avatar enabled!)

 

[cleeve]

Maybe you're right. The X1900 GTO spec would look alot like the X1700 XT spec... just with a different memory interface, I reckon. Maybe it's like this:

X1900 GTO
12 Pipelines
36 Shader Units
256-bit memory interface

X1700 XT
12 Pipelines
36 Shader Units
128-bit memory interface