RSX = Nvidia 7900???

Here's a few pages of the Nvidia 7900 review. I though of RSX when i read this

Based on the G70 architecture, Nvidia has pushed the performance envelope even further while decreasing the overall size of the silicon. The GeForce 7900 GTX, 7900GT and the 7600GT are all manufactured using a 90-nm process. This shrinkage increases the number of chips per wafer and should decrease the cost to produce each chip.

Replace the PC parts with the word RSX, and it makes sense.

Based on the G70 architecture, Nvidia has pushed the performance envelope even further while decreasing the overall size of the silicon. The RSX is manufactured using a 90-nm process. This shrinkage increases the number of chips per wafer and should decrease the cost to produce each chip.

See. ;)

The size reduction also allows the chip to use less power, which means less heat. The better thermal design also allows the GeForce 7900GT and 7600GT to use smaller and quieter cooling solutions. This will soon translate into an easier transition to the mobile market in which the thermal design parameters are more stringent compared to discrete graphics processor architectures.

Useful for RSX ;)

To compliment this die shrink, Nvidia reengineered some of the circuitry, which allows the G71 and G73 designs to use fewer transistors to do more. The G70 debuted as the GeForce 7800 GTX with 24-pixel pipelines and eight vertex units operating at a core clock speed of 430 MHz inside a 302 million transistor count package. G71 has the same number of core units as the G70 (eight vertex shaders, 24 pixel shaders and 16 ROPs).

Although there is no change to the amount of physical units compared to the previous design, Nvidia has optimized how they can perform."We changed the ROP performance as well as reconfigured some of the pipelines to make sure the card was more optimized over G70," Nvidia said.

IMHO, this has RSX written all over it.
 
If you look at all the specs (ROPs, 128 bit memory bus, clockspeed, price, PLUS no external power connector needed), i thought it was more like a 7600 but with more of the pixel shader pipes enabled. Its likely the 7600 has some pixel shaders disabled just to meet a lineup pricepoint, or redundancy for REALLY good yields.

Only thing throwing it off a little is the 5 vertex pipes on the 7600, not sure if they could enable more of those or Sony felt the cell would obviate the need for 8.
 
Last edited by a moderator:
Remember Nvidia claimed RSX was a seperate development, using 50 engineers, so RSX is not the 7900.

Another obvious thing to consider is the Cell FlexIO Bus that is added to RSX, and certainly not present on the current crop of Nvidia GPU's.

Since RSX will be manufactured at 90 nm, and so should share similar die size, and thermal characteristics to the 7900 series. This is excellent considering the 200 mm2 die size, and reasonably low thermal output.
 
RSX = over 300M transistors
7600GT = 177M transistors (incl. PureVideo)
7900GTX = 278M transistors (incl. PureVideo)

The already known differences are the interfaces (FlexIO/PCI-E, VGA/HDMI)
 
Last edited by a moderator:
expletive said:
If you look at all the specs (ROPs, 128 bit memory bus, clockspeed, price, PLUS no external power connector needed), i thought it was more like a 7600 but with more of the pixel shader pipes enabled. Its likely the 7600 has some pixel shaders disabled just to meet a lineup pricepoint, or redundancy for REALLY good yields.

Only thing throwing it off a little is the 5 vertex pipes on the 7600, not sure if they could enable more of those or Sony felt the cell would obviate the need for 8.

Even Nvidia have already said it FASTER than the 7800

What the improvements to the ROPS, could RSX have this aswel?
 
!eVo!-X Ant UK said:
Even Nvidia have already said it FASTER than the 7800

What the improvements to the ROPS, could RSX have this aswel?

Yep they definitely did.

Interestingly, benchmark wise, the 7900GT is almost identical to the 7800GTX. With the 7900GT being ever so slightly faster.

http://www.firingsquad.com/hardware/nvidia_geforce_7900_gt_gtx_performance/page7.asp

Couple of things to keep in mind:

1. The 7900GT is clocked 100mhz lower than what an RSX is alleged to be.
2. The 7900GT has a 256bit mem bus instead of 128bit in RSX.
 
!eVo!-X Ant UK said:
So RSX could wel be a G71 hybrid GPU???

Sure in the sense, it will be using the same vertex, pixel shaders, and ROPs. Hopefully in a 8 by 24 by 16 arrangement.
 
Well it has 256 MB GDDR3 VRAM clocked at 700 MHz, so that's 22.4 GB/s, and Flex IO bus to CELL providing 20 GB/s (write), 15 GB/s (read). CELL's memory bus is 25.6 GB/s, so RSX total bandwidth is 48 GB/s.

Expect some kind of "turbo cache" to help with data transfers between CELL's main memory and RSX. My guess 512 KB to 1 MB of SRAM cache.
 
Last edited by a moderator:
If data is generated directly from Cell RSX could get up 57.4Gb/s instead of being limited to how much bandwidth the XDR memory pool would allow for.

Generating data procedurally also would stave off RSX soaking up large amounts of Cell's bandwidth.
 
one said:
RSX = over 300M transistors
7600GT = 177M transistors (incl. PureVideo)
7900GTX = 278M transistors (incl. PureVideo)

The already known differences are the interfaces (FlexIO/PCI-E, VGA/HDMI)

Ahead of the game as usual ;)
 
Edge said:
Well it has 256 MB GDDR3 VRAM clocked at 700 MHz, so that's 22.4 GB/s, and Flex IO bus to CELL providing 20 GB/s (write), 15 GB/s (read). CELL's memory bus is 25.6 GB/s, so RSX total bandwidth is 48 GB/s.

Expect some kind of "turbo cache" to help with data transfers between CELL's main memory and RSX. My guess 512 KB to 1 MB of SRAM cache.

What is the memory bandwidth that would be equivalent to the 42 GB/s on the 7900GT? I dont think the bandwidth to Cell is relevant in comparing RSX/7900.
 
expletive said:
What is the memory bandwidth that would be equivalent to the 42 GB/s on the 7900GT? I dont think the bandwidth to Cell is relevant in comparing RSX/7900.

I don't know what the equivalent bandwidth would be, as read/writing anything from CELL's main memory involves greater latencies, than to it's own local pool of GDDR3 memory. CELL's bandwidth is relevant, as Nvidia said rendering can be done to that memory.
 
Edge said:
I don't know what the equivalent bandwidth would be, as read/writing anything from CELL's main memory involves greater latencies, than to it's own local pool of GDDR3 memory. CELL's bandwidth is relevant, as Nvidia said rendering can be done to that memory.

Understood but youre using a 48GB/s number to acces the 256MB of Cell's main memory, thats only half of what the PS3 will need. If we just take the 256MB of RSX to VRAM, its 24GB/s?
 
Edge said:
Well it has 256 MB GDDR3 VRAM clocked at 700 MHz, so that's 22.4 GB/s, and Flex IO bus to CELL providing 20 GB/s (write), 15 GB/s (read). CELL's memory bus is 25.6 GB/s, so RSX total bandwidth is 48 GB/s.

Expect some kind of "turbo cache" to help with data transfers between CELL's main memory and RSX. My guess 512 KB to 1 MB of SRAM cache.
Can RSX's framebuffer exist outside of the GDDR3 RAM? I understand it can use Cell to access XDR RAM, but there's more latency in that and would require even more significant reworking of the core logic to do that.

From what I understood, the framebuffers must exist in the GDDR3 RAM pool while it can access all other kinds of data (vertex, pixel, etc) from both RAM pools.

So for things like anti-aliasing, throwing a figure around like 48 GB/s becomes a red herring.
 
one said:
RSX = over 300M transistors
7600GT = 177M transistors (incl. PureVideo)
7900GTX = 278M transistors (incl. PureVideo)

The already known differences are the interfaces (FlexIO/PCI-E, VGA/HDMI)
This is the first time I've heard 278M transistors for the 7900GTX?

The 7800GTX was widely considered to have 302M transistors ("over 300M") (http://en.wikipedia.org/wiki/GeForce_7_Series), so if the 7900GTX only has 278M, what happened to the rest?

The number looks even more curious when we consider that the X1900 has 384M transistors.
 
Asher said:
So for things like anti-aliasing, throwing a figure around like 48 GB/s becomes a red herring.

True, but I don't think we were talking about bandwidth available to one single activity, but the GPU in total.

Asher said:
The 7800GTX was widely considered to have 302M transistors ("over 300M") (http://en.wikipedia.org/wiki/GeForce_7_Series), so if the 7900GTX only has 278M, what happened to the rest?

They "repipelined" the architecture to remove some unnecessary redundancy, apparently.
 
Back
Top