NVIDIA GF100 & Friends speculation

[Ailuros]

Ok, yes in that case 625/1296 doesn't make sense.

Could be a typo. In any case 1250-1300 hot clock for the 470 sounds within so far expecations.

 

[jimmyjames123]

And it seems its showing 1296MHz for shader clock..

No, it doesn't look like 1296 MHz, it looks like 1250 MHz. Half this frequency would be 625 MHz, which looks like the number above it.

Seems reasonable (maybe a bit lower hot clock frequency than expected?), with the 480 most likely having ~ 15-20% higher processor clock frequency.

 

[jimmyjames123]

Yep, that was precisely why I was questioning your earlier comment about 1Ghz being the lowest bin. Dropping down to 1.35v from 1.5v may result in considerable power savings (and they might be cheaper to boot).

Yes, the savings in power consumption may be significant. Also, the 470 and 480 have a much wider memory interface than HD 58xx, so there is much less need to have GDDR5 mem with ultra high clocks.

 

[fellix]

The lower GDDR5 rates could be linked to the interface buffer capacity in the GPU that may break the thermal envelope of the device and the cooling solution. I remember RV770's MEMIO thermal sensor readings going wild under most load conditions for some reason.

 

[Sontin]

I think the cache system should helps them to save bandwidth.

BTW: nVidia sends out cards with different clocks? Or is it in to "blur" out some information?

 

[compres]

OK so it's March 2010. Whare is Fermi? Am tired of waiting...

 

[mboeller]

Yes, the savings in power consumption may be significant. Also, the 470 and 480 have a much wider memory interface than HD 58xx, so there is much less need to have GDDR5 mem with ultra high clocks.

much wider? 320bit <-> 256bit = 25% more for the GTX470.

 

[itsmydamnation]

i wouldn't call 1000-1200mhz ram ultra high either

 

[Alexko]

Digital circuits have two main building blocks: transistors and wires. Transistors do the switching and store information, but they must be connected with wires.
In general, transistors switch faster as they get smaller. But wires get slower as they get smaller, for two reasons. Firstly, it's just hard to push current through small wires - this is called resistance. Secondly, the wires get packed closer together, which increases capacitance. Roughly speaking, with every new process generation, the transistors get a little faster and the wires get a little slower, and we're now at the point where the wires can dominate the time and power characteristics of a circuit.

You can make a wire propagate a signal faster many ways, here are a few:
1. Move the wire to a bigger metal layer. Generally, the lowest metal layers in the stack are the smallest, and are used for short connections. As you go higher in the metal layer stack, the wires get bigger (but there are fewer of them), which makes for less resistance and capacitance.
2. Cut the wire into pieces. The time it takes a signal to travel through a wire superlinearly related to its length. If you break the distance up into smaller hops, the sum of the delay for the entire path can be quicker. At the end of each hop, the signal has to jump back down to the transistor layer in order to be reboosted back to full strength and sent out to the next hop. This is called "buffering". You can only do this with a metal spin if you have extra buffers lying around which aren't being used - to add new buffers that aren't there would require a full silicon respin.
3. Make the wire shorter. Depending on the flexibility of your design, you can change the physical placement of the transistors it's connecting to make the path shorter. This is hard to do in a metal spin, and often introduces problems of its own: if you make one wire shorter, you often make another wire longer.

Thanks, that's crystal clear!

 

[jimbo75]

GTX470:

Vantage P16600
crysis,1920X1200 veryhigh+8x aa,24fps
crysis,1920X1200 veryhigh+No aa,53fps


Rumor from Chinese forum~
http://mylab.q.yesky.com/thread-121-1-1.html

So going on that, and taking THG's benchmarks - http://www.tomshardware.com/reviews/radeon-hd-5970,2474-8.html -

The 470 beats crossfired 5870's without AA, but loses to the 5850 with 8xAA.

This seems a tad unlikely but if true at least we know what kind of benchmarks to expect come March 26th.

 

[fellix]

Non-public variant of a GF100 SKU

More featured VRM circuity, no sign of vent openings in the PCB, 1xDVI, 1xHDMI and 1xDP on the backplate.

 

[Ailuros]

So going on that, and taking THG's benchmarks - http://www.tomshardware.com/reviews/radeon-hd-5970,2474-8.html -

The 470 beats crossfired 5870's without AA, but loses to the 5850 with 8xAA.

This seems a tad unlikely but if true at least we know what kind of benchmarks to expect come March 26th.

Granted Tom's test seems to be w/o AF in both cases which is besides the point. In 1920 the GTX285 goes from 1xAA= 24.9 fps to 8xAA= 19.5 fps which equals a drop of 22%.

Now according to that benchmark the 470 might turn out even faster than 2*5870 with 1xAA but drops by 53% with 8xAA. Tsktsktsk....

 

[FrameBuffer]

Non-public variant of a GF100 SKU

More featured VRM circuity, no sign of vent openings in the PCB, 1xDVI, 1xHDMI and 1xDP on the backplate.

google engrish transration:

From the M-one non-public version of this screenshot GF100 PCB board point of view, this piece of M-one non-public version of GF100 for 10-layer PCB design, length of about 10.5 INCH, about 266.7mm. 6PIN, and one each in 8PIN external power interface is expected to power more than 360W (very strong). The core power supply 12-phase, output 18 of tantalum capacitor, the input six solid capacitors. 2-phase memory power, output two solid tantalum capacitors and two capacitors, two solid-state input. Interface portion of DP + HDMI + DVI + VGA 4 interface design. Judging from DX 11 areas of ATI "lonely" days coming to an end

original pic:

 

[Ailuros]

My tantalum has just been tantalistically tantalized

 

[Rys]

A quick pixel counting exercise makes it ~500mm/sq, using the width of the longest (second) section of PCIe x16 connector as a reference. I make that 71.65mm according to wikipedia, so ~0.746mm/pixel (96 pixels wide), and I make the die area to be 30x30 pixels.

That's a five minute go for cheap thrills anyway

 

[Sontin]

The IHS has the same size like the G80 one. So, a die size over 529mm^2 is wrong.

 

[Dave Baumann]

The IHS has the same size like the G80 one. So, a die size over 529mm^2 is wrong.

How do you figure that?

The IHS is dictated by the package size and the package sizes can handle variable die sizes to some degree. Packages are more dictated by the number of pins they have to handle and the like.

 

[Sontin]

How do you figure that?

I compare the pictures of G80, GT200 and GF100. And i saw this:

http://www.xtremesystems.org/forums/showpost.php?p=4268014&postcount=1651

I think GF100 has a g80 die size.

 

[Ailuros]

495! Do I get a cookie?

 

[Rys]

I think GF100 has a g80 die size.

The die is under that IHS in all cases. As in, it's not that size, and you can't tell just by looking at it.