Two new (conflicting) Rumors

Saying NV30 will be faster than R300 is a very broad statement. This alone could be referring to the Pixel Shading capabilities and nothing else.

 

I just read somewhere (the Inquirer, no doubt) that ATi is transitioning their R300 to 12" wafers and .13micron? If that last long shot is true, I'd think it'd be safe to assume they'd move to DDR2, also.

 

quite conflicting indeed. I think that its really not prudent to discuss "speed" of the NV30 right now --- we know it will be DX9 compliant and more. I think we could leave it at that and just wait and see. The surprise is half the fun

 

I found something interesting in this slides (NV30 OpenGL extensions). In page 8, 'NV_vertex_program2 Performance':

As I think vertex shader instructions are executed (fetched) once per cycle per vertex pipe in both NV20 and NV25 this seems to refer to the number of vertex shader units. NV20 has one vertex shader unit and NV25 two. Would this mean that NV30 would have 3 vertex shaders rather than 4 as the R300?

 

Seams likely.

 

the "> 800MHz memory" rumour points towards RDRAM. But up until now everyone says they are using DDRII.
If the RDRAM rumour would be correct then an 128bit bus + 1200MHz RDRAM would give an bandwidth of 19.2GB/sec.
Would be nice (or strange) to see an GPU using RDRAM.

 

If you mean something like double channel 64 bit RIMMs or 4 channel 32 bit RIMMs at 150 MHz quad bumped (if it even exits), that would only mean 9.6 GB/s. I don't remember any anouncement about faster than 150 MHz RDRAM, and, in any case, it would be even expensive than DDR or DDRII.

 

Standard PC800 DRDRAM is 400MHz using double datarate transfers, not 150. 533MHz exists already. With memory mounted on a one-chip-per-channel basis like in the PS2 for example instead of serially on RIMMs with up to 32 devices per channel there's nothing to say the thing couldn't be clocked considerably faster, like 650MHz+.

But it isn't RDRAM because Nvidia said they're using DDR2. So let's end the speculation here mmkay?

*G*

 

What actual clock-speed DDR-II will be available when NV30 ships? If it's 450-500MHz then even with a 128-bit interface, they will have a fair bit of bandwidth to play with. With an improved occlusion technique over the R9700 is it possible that they could 'catch up' with the extra bandwidth the 256-bit interface provides?

In the past, NVidia chips have used memory bandwidth more efficiently with the various chips outperforming those from ATI with similar bandwidth. I'm not saying that this is definitely still the case, but it wouldn't surprise me if they could just about match performance of the R9700 with a 128-bit inferface.

As always, time will tell.

 

I don't think so, I don't see any possible improvements over the R300 occlusion culling by NOT switching to TBR or game / API calls.

 

Now the .13 micron for the R300 is a very interesting rumor. Sounds like an otherwise overclockable core will get another speed bump shortly. Now could the .13 micron vpu be the Radeon 9500 chip (R???) vice the Radeon 9700 (R300)? Simpler design with less transistors and would give the .13micron process time to mature before making the R300 chips on that process.

Still no rumors of working NV30 cards out which isn't good.

 

The Radeon 9700 at the .13 rumor is interesting. But how far along could ATI be in this process? I mean this is the first rumor that I have read that suggests that ATI is working on the .13 micron process, so I would call this particular rumor into question but it stands to reason that ATI would be working on a .13um solution.

 

Consider that 1.5x faster than NV25 = 2.5x as fast as NV25. If this is not what they meant, nVidia folks must totally suck in algebra

 

Well if Matrox can pull off a 256-bit bus with their comparitively limited engineering resources how hard can it be?

Unless they teamed up with ATi to produce the 256-bit bus? Anyone know if this is the case?

From what Nvidia's been saying it seems like they're trying to hype the features of their cards over speed. Like back in the early T&L days. They managed to convince everyone to buy a GF SDR and dump their TnT2 Ultra to get this mythical T&L which was going to revolutionize gaming down the road.

Of course this was total marketing.

If the Nv30 isnt much faster,or even slower, than the R300. I expect to see a blitz of information about features the Nv30 has that will revolutionize gaming.

 

I thought the concensus had been that NV30 was going to use Samsung's new 1 GHz DDRII...

 

Yeah, that was it...

But current info seems to suggest that Samsung will be unable to deliver 500mhz DDRII on time, so nVidia may use a bit slower memory for the time being (400-450mhz).

To avoid further criticism, that's all my assumptions.

 

Hmmm interesting to say none the less.

I dont doubt the nV30 should be faster than the R9700. But as we all know paper specs are worthless. Yes the .13u process should help...but then again it takes more than an "better" process to make a faster part.... No matter what it should be interesting...

 

Ati, Matrox, Nvidia, etc. don't team up on anything except open standards like OpenGL.