TechARP lists G300 series cards and specs

http://www.techarp.com/showarticle.aspx?artno=88&pgno=5
http://www.techarp.com/showarticle.aspx?artno=88&pgno=6
http://www.techarp.com/showarticle.aspx?artno=88&pgno=7

Considering the source, we may have winner on our hands. Techarp is generally reliable, although I find the timing odd...(as if NVIDIA was waiting for ATI to...nah )

If the specs are to be believed:

GTX 285: Vertex Shaders (240), ROP's (32), Pixel Shaders (240)
Core Speed (648, 1.476), Pixel Fill Rate (20,736), Texture Fill Rate (51,840)
Memory Bandwith/Type (512Mbit GDDR3), Memory Speed (1242Mhz), Memory Bandwith (158.98 GB/sec)

GTX 380: Vertex Shaders (512), ROP's (64), Pixel Shaders (512)
Core Speed (700, 1.6), Pixel Fill Rate (44,800), Texture Fill Rate (89,600)
Memory Bandwith/Type (512Mbit GDDR5), Memory Speed (1100Mhz), Memory Bandwith (281.6 GB/sec)

5870: Vertex Shaders (1600), ROP's (32), Pixel Shaders (1600)
Core Speed (850), Pixel Fill Rate (27,200), Texture Fill Rate (68,000)
Memory Bandwith/Type (256Mbit GDDR5), Memory Speed (1200Mhz), Memory Bandwith (153.6 GB/sec)

Feel free to debate.

 

Oh, I'll edit to add the 5870 for comparisions sake in a minute or two.

 

I want one GTX380... No, I mean I want two of those.

 

I am not sure if I take the memory bus width without a cup of salt. Everything else so far has been pointing towards 384bit.

 

Added 5870 to the list for comparison.

Specs look believable to me. Those fill rates and memory bandwith look awesome. Really, if these numbers are accurate, NVIDIA may have just pulled a rabbit out of its hat...

 

Uh, a couple things. Why do they list the vertex and pixel shaders separately when they are the same thing and have been since G80?

Also, if all they did was just double the GT200 core, then the die size is huge and these things are going to be priced at like $5 Billion. I certainty hope these are not the specs.

 

ymm its completely new architechture, not revamped GT200.

 

Interesting. I hope I can buy a 5890x2 or a GTX395 before christmas

 

It doesn't say that in any of those links.

 

You should know that by now that DX10 to DX11 needs architechtural changes.

Tesselation hw support is one of them.

 

Specs i heard are


GTX 380: Stream Pros ???, ROP's ??
Core Speed - 630, Shader - 1500
Memory - 384 bit GDDR5

 

A proper multithreaded scheduler and pixel shader 5.0 are two more.

The reason both are listed is because they list EVERY card, ever, so its there to not mess up legacy cards. I didn't even think of that when I copied the charts (by hand, mind you).

 

Well of course it requires minor architectural changes, but that doesn't make it a revamp of the architecture. The AMD R800 architecture is still extremely similar to the R600 architecture, despite the fact that it also is DX11. It's not like the tessellator is baked into the SP clusters or something.

Yeah I guess that's a good point. But I have a hard time believing that if these are accurate it's using a MIMD architecture for the cores.

 

Changes in DirectX level hardware don't constitute a change in the architecture of the core.

It's like saying the 5870 is a new architecture because it has DX11 hardware and not DX10, when its painfully obvious that it uses exactly the same hardware as used in the 4 series.

 

I've never once believed in the 384 bit bus. Not once.

Simply put, with NVIDIA already running on 512-bit, going back down made no sense whatsoever, even with GDDR5.

 

HAHAHA I am so full of you all dissing GT300 its almost ridiculous.

 

Actually, no. As both M$ and ATI had long since stated, the Tesselator on the Xbox360 (and by extension, 2000, 3000, and 4000 series cards) is only a subset of what is needed for DX11. I'd say extending the Tesselator and adding a updated pixel shader model would constitute a new architecture.

 

I'm off for the night. Lets see how big this thread gets...

 

Uh yeah it does, it decreases the size of the chip. If the bandwidth is not needed, nvidia can make the chips cheaper. And as far as benchmarks with the 5870, it doesn't seem to have trouble with memory bandwidth. Also it's extremely difficult to engineer a GDDR5 bus.

Here is a quote from anandtech:

You mean looking at it with a technical perspective? What you stated is wrong. DX11 doesn't change the architecture of the card. When the AMD guys were spitting out rumors of the 5870 having XDR ringbus or whatever the hell retarded rumor there was, I was also there explaining why that didn't make sense.

What? He didn't say anything about the tessellator. What he said, is that what you guys are saying is comparable to saying "the architecture in the R800 is different then the R600". It's the same thing, just doubled a couple of times with minor improvements. The overall architecture is the same. It doesn't constitute anything.



As you can see here at a high level the architecture is exactly the same between the 5870 and 4870. The difference of DX11 lies much lower, but the overall architecture is the same. Using a MIMD would drastically change the overall architecture. Which is what the original argument was about.

 

If those spec's are real, then the card is a monster.