RV350: 4 pipes, 256-bit DDR? (Inquirer)

[Pete]

http://www.theinquirer.net/?article=8003

This card should be much faster than a Geforce 4 TI4400 and than the currently shipping DX 9 Radeon 9500 PRO, the cut down version of the R300 chip. Clock speeds for the new chip should be somewhere between 350 to 375 MHz but are still unconfirmed. All we can say clearly for the memory is that it will use DDR with 256 bit bus.

That seems odd, to say the least. Would a 4-pipe card benefit from 256-bit memory bandwidth? My impression was no, and that the extra overhead that 256-bit incurs over 128-bit would be an unnecessary burden. And is 350+ MHz DDR still cheaper than the equivalent DDR2? Or does ATi have to stick with DDR to avoid working out (optimizing) latency issues with DDR2, if they exist?

The high clock speeds are encouraging, tho, and hopefully RV350 will compete well with the 9500 Pro. I hope the 9500 Pro's currently inflated $200+ prices are only temporary, too--or maybe I've missed my chance at a $180 9500 Pro. Will we see R300 phased out with the introduction of R(V)350, or will the 9500's and 9700's compete with their successors a while longer?

 

[antlers]

As we've learned from the pre-NV30 hype, the phrase '256-bit bus' is inexact. With DDR, a bus with 128 data lines is sometimes described as 256-bit because each data line sends two bits per cycle.

I suspect that, like the NV30, the RV350 is 128-bit DDR.

[Edit for typo]

 

[mboeller]

IMHO;

Yes it would help, not with performance but with price.

Instead of using an 400MHz DDRII-memory with an 400MHz 8-10 Layer PCB with 128bit

you can maybe use

an 200MHz DDR-memory with an 200MHz 6 Layer(?) PCB with 256bit

but I really doubt it. I don't think ATi will choose this route.

 

[Pete]

antlers4, I should hope Fuad would be able to distinguish between a 128- and 256-bit bus. He is a real person, right? With a real brain? The kind that calculated the 9700's bandwidth as less than the 4600's b/c he forgot about that tiny bus width distinction? Perhaps that's why TR referred to him as "Fraud." Or perhaps it's a joke I'm not privy to.

mboeller, 200MHz DDR at 256-bit sounds very interesting. One would think it would be about to get very cheap, as CPU RAM is transitioning to it (with AMD's and Intel's respective 400MHz and 800MHZ FSB's). Or maybe ATI would be better off with the economies of scale by moving their whole line to DDR2?

 

[SteveG]

antlers4, I should hope Fuad would be able to distinguish between a 128- and 256-bit bus. He is a real person, right? With a real brain? The kind that calculated the 9700's bandwidth as less than the 4600's b/c he forgot about that tiny bus width distinction? Perhaps that's why TR referred to him as "Fraud." Or perhaps it's a joke I'm not privy to.

I wonder if Fuad is a real person, or just the pseudonym that Inquirer writers use when they are spreading fuad - fear, uncertainty, and doubt. Fuad seems to write most of their "speculative" (fabricated?) articles, most of which turn out to be completely wrong.

 

[Arun]

IMHO;
Instead of using an 400MHz DDRII-memory with an 400MHz 8-10 Layer PCB with 128bit

Are you forgetting that DDR-I limit is 400Mhz, and there are very few ( if any ) reasons to use DDR-II instead of DDR-I beside clockrate?
So, if ATI target for the RV350 was 12.8GB/s, they don't need either a 256-bit memory bus or DDR-II.

Why didn't nVidia use DDR-I for the GFFX Regular then? Because the memiry controller was made with DDR-II in mind. And 2 chips, instead of 1, cost a *LOT* !


Uttar

 

[Joe DeFuria]

So, if ATI target for the RV350 was 12.8GB/s, they don't need either a 256-bit memory bus or DDR-II.

No, but they would still need "high-speed" DDR-I. You can achieve 12.8 GB/sec with 400 MHZ DDR-I at 128 bits, but that memory is probably awfully expensive. NO ONE is using it AFAIK, for any product at any price. It's not that it isn't available....just what cost?

We are all expecting at least one variant of the RV350 to be in the 375-400 MHz range. And to be "balanced," (edit: assuming 4x1 architecture) we are expecting bandwidth in the neighborhood of 12 GB/sec. That means 375+Mhz DDR.

Without being on the "inside" it's impossible to know what's more cost effecitve: 183-200 Mhz DDR on a 256 bit bus, or 375-400 Mhz on a 128 bit bus.

Looking at the market today...I woud tend to think that the 256 bit bus with 200 Mhz DDR would be more cost effective. However, the decision on design was made probably well over a year ago, and it could have been forecast differently....I only hope that the RV350 doesn't turn out to have clocks that are imbalanced, like 400/310, on a 128 bit bus...

 

[Bambers]

Are you forgetting that DDR-I limit is 400Mhz, and there are very few ( if any ) reasons to use DDR-II instead of DDR-I beside clockrate?
So, if ATI target for the RV350 was 12.8GB/s, they don't need either a 256-bit memory bus or DDR-II.

Why didn't nVidia use DDR-I for the GFFX Regular then? Because the memiry controller was made with DDR-II in mind. And 2 chips, instead of 1, cost a *LOT* !


Uttar

Hynix recently announced 500Mhz DDR-I didn't they?

 

[Simon Templar]

They Sure did, SirPauly found the press release:

SEOUL, Korea, December 10, 2002 – Hynix Semiconductor Inc. today announced it has developed a 500MHz-128Mb (4Mx32) DDR-I SDRAM for supply to the graphic chipset manufacturing market.

This high speed DDR-I SDRAM product, an industry first, features 500MHz frequency ( 1Gbps) speed, 32bit wide I/O and 144-ball FBGA package type. The new product supports the high speed needs of desktop and work station graphics and switch and router applications.

The new product has the highest speed in the DDR-I product line and is compatible with the DDR-I board. Hynix forecasts the combination of speed, compatibility, and an estimated cost advantage of 10% over DDR-II, will gain quick acceptance for the new product from the graphics and network industries. Farhad Tabrizi, Vice President of Worlwide Marketing, stated "providing the highest performance at the lowest possible price is the key for differentiating the product by end application. DDR-I at 500 MHz does just that."

Discussed in this thread:http://www.beyond3d.com/forum/viewtopic.php?t=4527

 

[mboeller]

Without being on the "inside" it's impossible to know what's more cost effecitve: 183-200 Mhz DDR on a 256 bit bus, or 375-400 Mhz on a 128 bit bus.

Contradicting my first posting in this Thread, maybe the 128bit bus is cheaper, cause you need only halve the amount of RAM-chips.

DDR-SDRAM - chips are only 32bit (64bit) wide and so with an 128bit bus you need 4 chips "to fill the bus", but with an 256bit bus you need 8 chips.
So maybe that's the reason for an 128bit bus.

 

[Dave Baumann]

Not that I wish to speculate on the bus width...

DDR-SDRAM - chips are only 32bit (64bit) wide and so with an 128bit bus you need 4 chips "to fill the bus", but with an 256bit bus you need 8 chips.

You are forgetting the size of the chips. With 4 chips you'd need 32MB modules for a 128MB card, which are probably more expensive than 8 16.

 

[mboeller]

Not that I wish to speculate on the bus width...

DDR-SDRAM - chips are only 32bit (64bit) wide and so with an 128bit bus you need 4 chips "to fill the bus", but with an 256bit bus you need 8 chips.

You are forgetting the size of the chips. With 4 chips you'd need 32MB modules for a 128MB card, which are probably more expensive than 8 16.

Really?

I thought that today the size of the chips does no longer count (so much), cause RAM in itself is rather cheap. I thought the packaging of 8 chips could be far more expensive then the packaging of only 4 chips.

If 8 16MB chips are really cheaper then 4 32MB chips, then an 256bit bus at 200MHz should really be cheaper then an 128bit bus with 400MHz. Good news.

 

[Pete]

I wonder if Fuad is a real person, or just the pseudonym that Inquirer writers use when they are spreading fuad - fear, uncertainty, and doubt. Fuad seems to write most of their "speculative" (fabricated?) articles, most of which turn out to be completely wrong.

Yeah, makes sense considering TR's follow-up. They can hide behind a psuedonym to hide their seemingly flimsy grasp of memory tech.

 

[Dio]

He seemed pretty real when I met him.

 

[Tahir2]

He can't be fake.. no one can type e-mails that are that bad grammatically (he is Bosnian I believe and English is not his strong point) and be a figment of someone's imagination.

 

[Pete]

So his specialty is video cards, or does he do other reporting?

 

[Nagorak]

I wonder if Fuad is a real person, or just the pseudonym that Inquirer writers use when they are spreading fuad - fear, uncertainty, and doubt. Fuad seems to write most of their "speculative" (fabricated?) articles, most of which turn out to be completely wrong.

Compare it to some of the other articles and see if they use the same style (not that there's much style in news articles, especially tabloid style).

 

[Bjorn]

Without being on the "inside" it's impossible to know what's more cost effecitve: 183-200 Mhz DDR on a 256 bit bus, or 375-400 Mhz on a 128 bit bus.

We're talking about a value card here (that's probably going to be available for quite some time) so i guess the question is also, which is going to be cheaper in 6+ months ?

But as you say

Without being on the "inside" .....

 

[MuFu]

RV350 is 4x1, 128-bit DDR-I memory interface.

The bus tech is pretty obvious from that leaked photo.

MuFu.

 

[Joe DeFuria]

What are the clock speeds of the variant parts (pro/non pro), oh wise one?