AMD: R8xx Speculation

How soon will Nvidia respond with GT300 to upcoming ATI-RV870 lineup GPUs

  • Within 1 or 2 weeks

    Votes: 1 0.6%
  • Within a month

    Votes: 5 3.2%
  • Within couple months

    Votes: 28 18.1%
  • Very late this year

    Votes: 52 33.5%
  • Not until next year

    Votes: 69 44.5%

  • Total voters
    155
  • Poll closed .
I have to complain about your image hosting.. it was ultra-slow.. but great numbers there! (seen my pm yet)
 
I guess thats why the future is Fusion and not this card? :cool:

i agree, but fusion will be the lowlowlow end of the graphic, even if it launch in a year, it's strange that at the moment they think that 80sp it's good enought, and for a more complex chip they think that you need at least 400
 
It will come out soon (sooner than FurbyTM)

it will have something between 800 and 1440 SP's and will be price between HD5770 and 5850 ;)

My guess is 1120 SP's and 56TMU's on clocks a bit lower than 5850 priced at least $50 over HD5770. It will (probably) also retain the benefit of a 256 bit bus since it's basically a cut HD5850 (same board et al.)

Huh, looks like someone has posted their guesstimation on Wikipedia.

http://en.wikipedia.org/wiki/Compar...ocessing_Units#Evergreen_.28HD_5xxx.29_series

Those would be solid spec's for where it might sit in the ATI lineup.

I'm thinking the memory speed is kind of low for GDDR5 but maybe ATI is trying to keep it a balanced part. Also might there be a possibility it would only need one 6-pin PCI Express power connector? Or does the 5770 already come close to pushing the limit there?

Having the best possible performing part within that limit might be a minor accomplishment but it would be a neat one imo.

And it might help at the brick and mortar stores.

I see now that the GTS 250 manages to need only one 6-pin connector. And that's using something like 145 or 150 watts, total board power consumption.

As the 5850 is being reported to use 151 watts it seems fairly certain the 5830 will use less than a GTS 250.

But as a likely selling point of the 5830 is that it will overclock well we may well see two 6-pin connectors being either on the reference board or on custom layouts.

Just thinking out loud here. The 5830 might suit me nearly as well as a 5850 given my cpu and resolution limitations.

I've got a Q6600 @ 2.40 GHz and a monitor at 1920 x 1080.
 
Fudzilla talking about Cypress refresh:
http://www.fudzilla.com/content/view/17274/1/

Since it appears rv8xx doesn't quite scale as well as it should from juniper to cypress (that is, 2xjuniper is often quite a bit faster, even though it doesn't appear to be because of tri setup limitations), I'm wondering is that something AMD could fix with a refresh chip (so if internal bandwidth is the problem, double that up or something)? Though that fudzilla article doesn't really say anything...
 
Well we already know that 1ghz is possible on current 5870s and that some companys will bring out an overclocked verison with these speeds.

Perhaps a respin and 6 months of improvements on the 40nm will let them release a card at 1.2 ghz or so. That would be a sizable bump in performance. Not a fermi killer but certianly something more competive. I doubt they will have another gpu with more shader units or rops unless they are currently disabled.
 
Fudzilla talking about Cypress refresh:
http://www.fudzilla.com/content/view/17274/1/

I'm sure there'll eventually be a thread devoted to this at some point but where exactly do you guys suspect AMD will go with their next architecture? Nobody expected Nvidia to tackle geometry head on, are there any other elephants in the room that AMD might want to take on?

With Cypress we've already seen that shaders, rops and texture units are providing rapidly diminishing returns yet we have folks claiming that bandwidth isn't a problem and geometry throughput isn't a problem. So what is it then?
 
I'm sure there'll eventually be a thread devoted to this at some point but where exactly do you guys suspect AMD will go with their next architecture? Nobody expected Nvidia to tackle geometry head on, are there any other elephants in the room that AMD might want to take on?

With Cypress we've already seen that shaders, rops and texture units are providing rapidly diminishing returns yet we have folks claiming that bandwidth isn't a problem and geometry throughput isn't a problem. So what is it then?

I think we will need plenty of shader units , rops and texture units if they want eyeinfinty to catch on. And I mean really those dropping $400-$2000 on just video cards each gpu generation can easily afford three $300-600 monitors to go eyeinfinty.
 
With Cypress we've already seen that shaders, rops and texture units are providing rapidly diminishing returns yet we have folks claiming that bandwidth isn't a problem and geometry throughput isn't a problem. So what is it then?

Partially video drivers?
 
I think the refresh of RV870 - aka "RV890" will be based on more mature 40nm tech, BiG "IF" it will be released in March 2010, that puts 6 months apart from RV870. Simply ATI had time to optimize the core to hit 950MHz+ for the GPU with same power consumption as RV870 @ 850MHz. And as well - faster rev. memory from samsung GDDR5 (6.0Gbps).
Really we are looking here 10-15% percent improvement, still - not enough to challenge on paper GF100 "GT300".

Unless ATI improves some how the core "GPU-design" to run more efficiently to hit 20% percent improvement for the refresh GPU.
 
You really think they will only get 100mhz out of the chip in 6 months ? They are already selling chips at 870mhz . So its not even a true 100mhz upgrade. I don't see what they would get out of that.

launch day cards were already getting close to that or exceeding it

http://www.legitreviews.com/article/1090/1/
http://www.guru3d.com/article/radeon-hd-5870-review-test/26
http://www.hardwareoverclock.com/Radeon_HD5850_und_H5870_Overclocking_Guide.htm


On various forums giving it more juice will get you to around 1.1-1.2ghz depending on if you go 1.2 or 1.3 on the voltage. For many the stock cooler with the fan turned up some keeps the card nice and cool. Though I've seen a guy at 1.3ghz but he had insane water cooling . Wish a site would use some of the new methods for overclocking other than just the ccc. Would be interesting to see what performance is like at 1ghz and 1.1ghz to see what amd would need to really do to the card.

The good news for me personaly is the 5850 seems to clock up to 1ghz also. So if that drops to $200 when fermi comes out i might scoop it up.

So I can easily see a respin with a newer cooler clocking at 1ghz or over.

Though if they have a brand new gpu design coming out in the fall they may not want to push things to far with the 58x0 series anyway . After all kyle at hardocp is saying late march at the earliest and we don't even know if thats both high end cards. Ati can easily compete by droping the price for a few months before the next generation is ready.
 
Oh just a FYI to you people, stop calling Cypress RV870 because Dave will eat you! If you want to talk about a refresh part you should talk in agricultural terms!

So back on topic:

If we're talking about Cypress here, are we talking about improvements in the growth medium (wafers), genetic stock (architecture) or some other factor like better grading of the end product to make better use of the high grade stock?
 
Oh just a FYI to you people, stop calling Cypress RV870 because Dave will eat you! If you want to talk about a refresh part you should talk in agricultural terms!

So back on topic:

If we're talking about Cypress here, are we talking about improvements in the growth medium (wafers), genetic stock (architecture) or some other factor like better grading of the end product to make better use of the high grade stock?


Well I'm not in the know.I would think the process will be much more mature than it was when the final clocks were settled on for the cypress family back in the summer. I'm thinking an april or so release for a 5890 card. So you figure you got a good 7 or 8 months from when the cypress clocks were finalized. Ati might have also done another spin to fix any bugs that were wrong in the original cypress. So 7-8 months of process maturity should amount to not only better yields but the chips them selves testing at higher clock speeds or reduced heat production and less power requirements to run them. Or perhaps all three.

Its like anything else the more you do it the better you become at it.
 
Back
Top