• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

ATi 5890 -Cypress XTX-

d_brennen said:
I would have thought the refresh of the 58xx series would have just been a higher clocked/refined process rather than more transistors on the core
Click to expand...

The 4890 is basically an overclocked 4870 but it uses something like 3million more transistors than a 4870 if my memory serves me well.

As said l can't see ATI doing something very different with the 5890 like adding more shaders or going from a 256-bit to a 384-bit bus.

Maybe we will see a higher clocked part with faster gddr5 to provide more memory bandwidth.
 
drunkenmaster said:
Q2 would be entirely outside the possible, likewise 1920 shaders would perform the same at the same clocks on 40nm as 28nm, and theres plenty of clock headroom left on 40nm.
Click to expand...

I wouldn't want to bet on that - tho granted you know better than I do on that front. Theres deffinatly been a lot of moves to bring 28nm forward... tho unless they cut corners I can't see how they could have retail cards out using that inside of 8 months.

The problem with the current design on 40nm is efficiency - they have very high peak figures but as they increase the number of shaders, etc. in the design the less sustained performance they are getting back - as we've already seen doubling the figures over the 4870 isn't that close to double the performance much of the time... 28nm gives them some breathing space to sort that out.
 
It's probably worth noting that rumours involving a non-power-of-2 bus on an AMD card is generally an extremely big hint that it's fake.
 
Lightnix said:
It's probably worth noting that rumours involving a non-power-of-2 bus on an AMD card is generally an extremely big hint that it's fake.
Click to expand...

EDIT: See jokesters post below.

I still think its fake tho - 1920 shaders on a 28nm shrink is wasting space - I'd expect the real number to be above 2000.
 
Last edited:
Rroff said:
EDIT: See jokesters post below.

I still think its fake tho - a 28nm die shrunk cypress core would be more likely to have 2080 or 2240 shaders, 1920 sounds more like what you'd expect with a 32nm core.
Click to expand...

Not impossible, of course not. So unlikely that I'd call it fake? Absolutely. In the past repeated rumours have floated around about AMD cards with non-power-of-2 bus widths, but they've never emerged, and typically the rest of the specifications would be incorrect also (most recently, 5770 with 192-bit bus and 1120 shaders - you can also find older examples like whatever this thing was supposed to be). I'd actually be quite surprised if AMD's memory controller/ROP configuration could even support such a setup at the moment, because there are multiple cards that a 192-bit bus would have really helped (4670!!!), but of course we are talking about a different core.
 
if you can get 1600 shaders on a 40nm core, you can easily be fitting 2400+ on a 32nm core, 28nm opens up the way for well, basically double shader power quite easily. However we're moving towards a new architecture rather than same old same old, though I don't really see base shader power increasing at a different rate at the moment.

As for bus, and things like the 4670 having 192mbit bus, well, it doesn't happen for cost. Which is partially the die size/complexity/signalling but its more the pcb cost. The more traces they more they cost, the thicker they'll need to be, the more expensive and longer they take to build and the more problematic things get.

You can take any card and say, well increase spec X Y or Z by 50% and it would be superb, of course it would, but it would likely kill off the use of a higher end salvaged part, which just loses them money.

If the 5770 was 192mbit bus, or even a 256mbit, not only would it cost more and perform way closer to a 5850, it would make a 5830 very hard to sell, because the difference in performance would be pretty small, yet the price to make for AMD/Sapphire of a 5830 is WAY higher than a 5770. If you can the 5830, then you have a lot of dead silicon that ends up not being sold, which is a waste and simply means the cost of production is spread across less cores, bad for us.

Rroff, theres not a chance in hell of 28nm from TSMC before the 4th quarter, because its TSMC frankly, what a CEO says publically when trying to secure orders for future parts, and whats really going on are different things. THe shell of the new fab is built, the inside is yet to be equipped, they probably won't start test runs for another 2-3 months, if, due to the same entirely stupid design as the 40nm process, the yields are utterly useless then we're months away from anyone wanting to make a real core on it.

From a TSMC perspective I can't see them producing much more than a "4770" type test product before the 4th quarter. Though its harder to call than normal as since Glo Fo started to take shape they've doubled their R&D spending, its very hard to judge when that money will kick in and effect things. Where they'd wait and not rush things at extra expense, they might do it now.

Glo Fo, well, I wouldn't be surprised if from this point on Glo Fo always have a higher yield, higher quality better working process than TSMC.

IN terms of production of graphics cards, the next 18 months will be very interesting to see who can keep up with who, and if either company is significantly behind on their process, will it utterly screw Nvidia or AMD. Thing is AMD can still use TSMC easily, Nvidia will be very reluctant to give a chunk of the profit from every core, to a company that owns 25% of AMD.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom