More R350 Info

[Dave Baumann]

His question was "Does anyone know..." - he wasn't asking what bus size it was, but whether anyone knows the bus size!

 

[Sabastian]

His question was "Does anyone know..." - he wasn't asking what bus size it was, but whether anyone knows the bus size!

Dammit Dave thats bloody well cruel.. lol. On that note though I think I am going to bed earlier tonight.

 

[WaltC]

I think DDR II is wrong, anyway

I do, too. In fact, I've never seen much sense in it for R350. nVidia felt it had to have it for the obvious reasons, though. With Intel now pushing for DDR 400 I think 1 got a big shot in the arm, anyway. I'm more interested in something like GDDR3 for R400 which I think will be DDRII minus the latency penalties. Which would be nice...who knows, by the time DDRII arrives for the mainstream it may well be GDDR3...

 

[elroy]

Taken from HardOCP:

Blue Knights Spying:
Our super-top-ultra-secret contacts as Best Buy have this to say:


...the presells for the r350 will be arriving around 03/03/03...

Just remember, you did not hear that here.

8)

 

[Socos]

Oh sure... pick on the new guy.. I'll make sure I am more picky with my wording from now on..

I understand the value part part.. And that a 256 bus is more expensive, however, if ATI were to put a 256 bus on that RV350 it would seriously rockorz!! [/quote]

 

[RussSchultz]

Oh sure... pick on the new guy.. I'll make sure I am more picky with my wording from now on..

I understand the value part part.. And that a 256 bus is more expensive, however, if ATI were to put a 256 bus on that RV350 it would seriously rockorz!!

[/quote]

Why would it r0x0rs? the RV350 supposedly has half the pipes (which makes it cheaper). At that point, a 256 pin wide bus would be overkill, as its core likely couldn't make use of it.

 

[noko]

Maybe Samsung made alot of DDRII modules for the expected usage by you know who and kept producing them for over 5 months with the most earnest empty promisses from a prospective buyer that ever was. Now that all their ware houses are full of DDRII accumulating dust, rodents and bugs, Samsung realizes that if they don't get rid of the stash the cash will dissappear in a dash.

ATI and their gang of wealthy partners then gets the whole stock and barrel for a mere pittance. Of course the would be buyer will claim that DDRII isn't fast enough and held back utimately their superior design and once and for all GDRIII was indeed needed.

 

[Ostsol]

Sounds like the RV350 will be what the Radeon 9500 was meant to be. Half the pipes, half the bus, and much cheaper. The difference will be that the RV350 is designed that way, rather than just crippled. As such, perhaps it won't suffer from the same problems as the Radeon 9500 -- such as HSR not working.

 

[mboeller]

After a year of browsing this forum I finally wanted to post something bad enough to register... Here goes.

Does anyone know whether the RV350 will be 128 bit bus or 256 bit bus?

Good question, but I think it will be on a 128bit bus simply because it is supposed to be a value part. A 256bit bus would require a more expensive PCB and I think ATi has enough varieties of the 256bit bus to go around at this point. I'd wager on a 128bit bus on a value part..... best guess.

Sure; the PCB would be more expensive, but the RAM would be far cheaper, cause instead of using 400MHz DDR2-RAM they would be able to use 200MHz DDR-RAM. So the package could be (far) cheaper.

 

[MuFu]

Does anyone know whether the RV350 will be 128 bit bus or 256 bit bus?

128-bit.

MuFu.

 

[Arun]

Thanks, MuFu! Your info is always more precise than Wavey's
BTW, continuing in our "value part memory bus width and speed queries" episode, do you know what's the bus width & memory speed of the NV36?
Not like it's very interesting right now since we don't even know the NV36's pipe number or anything... Ah well...


Uttar

 

[MuFu]

Thanks, MuFu! Your info is always more precise than Wavey's

Errrrm... no.

As for NV36 - no idea. My best guess would be mildly bastardised NV30/NV35 core + NV34 memory controller (DDR-I). Maybe they've gone back to the drawing board now that RV350 is out of the bag.

Nice not being under NDA like some people.

MuFu.

 

[Arun]

Well, at least you give a clear answer.
Wavey's "Yes" wasn't quite that informative...

Oh, and why would it use the NV34 memory controller? Why not the NV31 memory controller? Or wait, what does the NV36 replaces? The NV31, or the NV34?


Uttar

 

[WaltC]

Thanks, MuFu! Your info is always more precise than Wavey's

Errrrm... no.

As for NV36 - no idea. My best guess would be mildly bastardised NV30/NV35 core + NV34 memory controller (DDR-I). Maybe they've gone back to the drawing board now that RV350 is out of the bag.

Nice not being under NDA like some people.

MuFu.

I would be surprised to learn they didn't take nv30 back to the drawing board last September. Of course that was before they had anything finalized or had any idea about yields, so maybe it took them a few more months to understand the need for a redesign. I'd always wondered how you could set up an accurate sim for an unproven manufacturing process--I guess I have my answer...

 

[Hellbinder]

Sounds like the RV350 will be what the Radeon 9500 was meant to be. Half the pipes, half the bus, and much cheaper. The difference will be that the RV350 is designed that way, rather than just crippled. As such, perhaps it won't suffer from the same problems as the Radeon 9500 -- such as HSR not working.

Huh????

 

[Nagorak]

Sure; the PCB would be more expensive, but the RAM would be far cheaper, cause instead of using 400MHz DDR2-RAM they would be able to use 200MHz DDR-RAM. So the package could be (far) cheaper.

I imagine that after a certain point the difference in ram costs is negligible. This is one "trick" the PC industry likes to use. For example, you can buy 100 GB hard drive online for $100, logic would tell you that a 20 GB should cost about $25 or $30 (giving some leeway due to the cost of the drive housing and less efficiency of having smaller platters crammed in). But the truth is the cheapest you can find a 20 GB is $51, and a 30 GB only costs $53!!! (based off of a quick pricewatch check)

So basically, after reaching a certain price point products either are not sold for any less or just drop off the map completely. Granted, this is in the retail space, but I imagine the same holds true for hardware developers like ATI. (though perhaps to a lesser extent since the profit margins aren't magnified all the way from production-wholesale-retail).

 

[Nagorak]

]

Sounds like the RV350 will be what the Radeon 9500 was meant to be. Half the pipes, half the bus, and much cheaper. The difference will be that the RV350 is designed that way, rather than just crippled. As such, perhaps it won't suffer from the same problems as the Radeon 9500 -- such as HSR not working.

Huh????

Part of the reason the R9500 np is so much slower than the pro isn't just because of 4 pipes, it's also because Hierarchical-Z is disabled in it. Performance on the np is pretty abysmal, really, even for things that should be more bandwidth limited (AA, etc).

 

[Socos]

Why would it r0x0rs? the RV350 supposedly has half the pipes (which makes it cheaper). At that point, a 256 pin wide bus would be overkill, as its core likely couldn't make use of it.

Well maybe you have a point there... And maybe thats why Nivida didn't go 256 bit either????


Thanks for answering the question MuFu

 

[arjan de lumens]

I imagine that after a certain point the difference in ram costs is negligible. This is one "trick" the PC industry likes to use. For example, you can buy 100 GB hard drive online for $100, logic would tell you that a 20 GB should cost about $25 or $30 (giving some leeway due to the cost of the drive housing and less efficiency of having smaller platters crammed in). But the truth is the cheapest you can find a 20 GB is $51, and a 30 GB only costs $53!!! (based off of a quick pricewatch check)

So basically, after reaching a certain price point products either are not sold for any less or just drop off the map completely. Granted, this is in the retail space, but I imagine the same holds true for hardware developers like ATI. (though perhaps to a lesser extent since the profit margins aren't magnified all the way from production-wholesale-retail).

With hard disks, once you get down to a single platter and 2M cache, there is essentially nothing left that the hard disk manufacturer can do to cut production costs any further - in a mature product line, the cost of a platter is pretty much independent of its data carrying capacity, and with a per-platter capacity of 40-60 GBytes, that's pretty much why you see hard disk prices flatten out around that point.

Similarly, for graphics cards, you have manufacturing costs for GPU, RAM, chip packaging and circuit board, which flatten out once you get near-100% yield on all the chips - beyond a certain point (~$50?) , the only way to cut costs further is to integrate the GPU with something else.

 

[Sage]

ok considering what ATi managed to do with R300 i am going to laugh my ass off if they pull a 400+ core speed for the R350. as for the mem i'll bet it supports DDRII just like the R300 did but probably not originally planned to be used. remember nVidia was rushing to get DDR-II out while ATi was working on the better GDDR3. most likely ATi knew that by the time GDDR3 was finished they would be ready to put it on the R400 and with a 256-bit bus they wouldnt need beyond DDR-I for the R3x0 series. now then it may be, as someone else suggested, that they will choose to use DDR-II but, not because they need it, but because they can get it cheap considering they are the only ones that can actually use massive ammounts of it. If they do go with DDR-II then i'd expect them to push it to higher clock speeds to overcome the latency issue. it's possible that they will use DDR-I for the 9800 and then switch to DDR-II for the 256MB 9900 to save money. either way, i really doubt anyone will need to worry about it because the thing will be so damned fast....