Post your breaking NV30 news links here!

Why are you glad?

 

I would be glad too...

in the sense that if it is in fact 20-30% faster (across the board) than the Radeon 9700, and it's doing that with significantly less raw bandwdith....think of what that means for the NV3x architecture on a 256 bit bus.

 

I was asking the same question. But "only" 25-50 % faster is a bit of a funny thing to say as well. I can only presume that he infers that the NV30 would be considerably faster if it had a 256 bit bus. Of course assuming that the NV30 was designed with the 128bit bus in mind there should be no 256bit bus NV3X anything ever.

 

I meant the 2x speed increase of a 256bit bus is still to come in the future...
Maybe nv35 or maybe nv40, doesnt mather, but it hasnt been used yet
Anyway it would be a disapointment if it would be only 25-50percent faster with almost twice the bandwidth as the radeon 9700 if it would have a 256bit bus.

 

http://www.theregister.co.uk/content/3/28153.html

"First up, a faster processor: 500MHz v. 325MHz in used in the TI4600. Next, there’s more memory bandwidth 1GHz DDRII DRAM (600MHz DDR1); new anti-aliasing technology, dubbed Intellisample, designed to double the efficiency of memory reads and writes; and AGP 8X bandwidth (double that of AGP 4X). Built using 0.13micron production technology, the GeForce FX also handles eight pixels per clock cycle, against 4 for its immediate predecessor, by definition doubling the theoretical fill rate."

 

It isn't likely that nvidia would release a new core anytime soon that is compatible with a 256bit bus.....period. Hence my conclusion assuming that if the NV30 is only a 128bit bus then you won't see a NV3X anything that is compatible with a 256bit bus. Now on the other hand the NV40 core *should* be 256bit bus unless of course nvidia has even more bandwidth saving methods.

 

Well i Don't think that *only 25-50%* is a funny phrase.

You guys dont get it. do you. They are delaying even longer in order to make changes that wil allow them to operate at 500mhz core with 500mhz ram. Yet a 200mhz core increase and 200mhz ram increase is only netting them 25-50%. And i will bet you all right now that this 25-50% is not even accross the baord. Lets wait to see the final benchmarks becuase marketing ALWAYS exagerates the truth. Ill bet that its only their new FSAA method that even allows this much of a lead (when it happens).

Supposedly no numbers are going to be released today. Why do you think that is??? becuase as of today there are no Nv30's running at 500mhz.

I am going to wait till the rubber meets the road on this one. As should all of you.

 

Hellbinder can you possibly BE more of an ATI fan-boy? Your Anti-Nvidia "everything" is so hilarious! To be fair I do agree that final performance figures are all that I will take seriously.

 

I believe we do. We're just not overly dramatic about it.

No, that's a 200 Mhz core increase, and a 4 GB/sec decrease in raw bandwidth. The latter being the main reason for being "impressed" if the performance rumors pan out.

Quite possibly true. No one said otherwise. You'll notice all of the "IF" qualifiers in our statements.

I agree with that. IF there are no benchmarks run on actual sample hardware, that only means one thing: nVidia's current sample hardware is not running at target speed. And if it's not running at target speed now, there is no guarantee that the shipping product will reach target speed.

What will be interesting to see, is if nVidia officially states the target speed (MHz). If they do, that at least indicates their confidence level in reaching that speed. (Again, still no guarantee they'll make it though.). If nVidia remains tight-lipped about Mhz...then all bets are off.

We are.

 

That would involve redesigning (at least) the memory controller, wouldn't it?

 

Don't get bent out of shape of zero in on this 25-50% number.

25 - 50% increase was used in a context that did not specify what was 25-50%, but if you look at the clock speeds it fits right in. (i.e. its a mhz to mhz comparison and not a performance comparison)

As for waiting for the rubber hitting the road, you should also. You're already making statements of fact based on nothing but your own opinions.

 

Although I am no chip enginer, I think, yes it would "at least" requirer a new memory controler.

 

Not sure. It's possible the NV3x may be designed with both in mind. For example, the GeForce256 was designed to use both 128 bit SDR and 128 bit DDR. (And their MX parts I believe could also use 64 Bit DDR). So, nVidia could have designed the core (cache size etc.) to be a little "overkill" or even non-optimal for 128 bit. They would sacrifice a bit of cost for the 128 bit version, with the gain of not having another memory interface design to do to move to 256 bit...

 

http://www.hwupgrade.it/articoli/734/1.html
its italian, i believe

 

Joe I understand what you are saying with regards to the original Geforce 256 design. But to move the NV30 to the 256 bit bus would require an increase in pin count wouldn't it? In regards to the move from SDR to DDR it is simply a matter of increasing the number of bits per pin isn't it and not rely requiring an increase in pins as a result of still using the 128bit bus. Please correct me if I am wrong here.

 

Hmm. Since when does the Ti4600 sport a 325MHz chip clock? Typo in the press release? Or mistaking CPU clock with RAM clock?

ta,
-Sascha.rb

 

Time for http://world.altavista.com translation engine.

The connector is a power connector.

 

Speaking of bandwidth, over the weekend I was surfing for info on DDR-II and found Ace's memory guide part 3. Interesting read, but at the end was a note about "yellowstone" technology from Rambus. It's an octal data transfer scheme that has been demoed. Ace's was speculating use in graphics cards. Back to bandwidth, because of the 1-byte transfer rate, Ace's was quoting a peak data transfer using 128-bit bus and 400 MHz of 100 GB/sec. To all those who say 256-bit is a necessity, there are other ways. Don't get caught thinking there is only one way to do things.

 

Even though that the NV30 is rumored to use only a 128bit bus that card looks expensive to build ... relatively speaking of course.

 

Considering the crossbar nature of the controller, I'm not so sure it would be that easy. I mean, they have 4*32 bit right now, so would they add another 4 controllers, or increase the width to 64? None of the choices are easy to implement AFAIK...