AMD: R9xx Speculation

[Mindfury]

http://www.chiphell.com/forum.php?m...=fastpost&random=0.38606609877944853#lastpost

Barts:

 

[trinibwoy]

I'm lost, Barts is the one that's ~ Cypress right?

 

[neliz]

I'm lost, Barts is the one that's ~ Cypress right?

5700 replacement yeah.

 

[Kynes]

PWM chips in the left of the card?

 

[Kaotik]

http://www.chiphell.com/forum.php?m...=fastpost&random=0.38606609877944853#lastpost

Barts:
http://i51.tinypic.com/t89dg1.jpg
http://i51.tinypic.com/xndavp.jpg

Only 1 CrossFire bridge connector?

 

[Jawed]

wrt 2 I think it might not improve with Barts (if it has same number of TMUs than Cypress), only Cayman (e.g. there could be a 2 times cache bandwidth improvement for 1.5 times the TMUs)

Patent documents I linked earlier might suggest two cores' TMUs share a single L1. So while Barts might have the same TMU count as Cypress, the bandwidth per L1 from L2 would be much better.

That still doesn't really answer how effectively two sets of TMUs share an L1, i.e. what trade-offs occur.

 

[Mianca]

Don't those 2x6Pin power connectors really give it the look of an HD 68xx series card?

 

[UniversalTruth]

68** series using such low quality cooling solution. never gonna happen.

 

[Mianca]

68** series using such low quality cooling solution. never gonna happen.

So, you think a 150W+ HD 67xx card seems more probable?

That being said, I wouldn't hold my breath on the cooling solution being used here given that the card doesn't even sport a slot plate

Looking closely, is that the retail-cooler's ouline drawn on the PCB?

 

[Zaphod]

68** series using such low quality cooling solution. never gonna happen.

Looking at the layout, it seems to be obviously designed for a shrouded cooling setup with exhaust. I assume that particular HSF is just for the engineering sample.

 

[Kaotik]

The PCB's of late have those drawings for "shrouded cooling setup" regardless of the actual used cooler, even HD5750 which has the "egg-shaped" cooler as reference has one drawn on the PCB (and no, it doesn't match the shape of the "egg-cooler")

 

[neliz]

what's wrong with just a shroud over this fan? look at the MSI cyclone and nvidia's gtx460 reference cooler (without shroud.)

 

[Zaphod]

The PCB's of late have those drawings for "shrouded cooling setup" regardless of the actual used cooler, even HD5750 which has the "egg-shaped" cooler as reference has one drawn on the PCB (and no, it doesn't match the shape of the "egg-cooler")

I was thinking more in terms of PWM placement and the big block of outputs than the silkscreened outline.

 

[Mianca]

Still, the PCB seems rather big for an HD67xx card ...

Moreover, as for the cooler/shroud design, both the HD5770 PCB (top) and the the rumoured Barts PCB (middle) have "cooler drawings" on them, but the Barts PCB actually seems to have "holes" to fasten one (cf. right hand side).

Also, notice how the ram placing seems very similar to that of the HD5870 PCB (buttom) / green line.

 

[UniversalTruth]

Have some performance details about Barts leaked ot not yet?

 

[Kaotik]

Have some performance details about Barts leaked ot not yet?

No, but unless it outperforms HD5800-series, there's just no way to explain naming it 6800-series.

 

[UniversalTruth]

In fact there is a way. It is "Lack of Competition".

 

[Kaotik]

In fact there is a way. It is "Lack of Competition".

That's not a way to explain the naming if you're losing to your own last gen x8xx-part

 

[no-X]

9800GTX was slower than 8800Ultra (and many times slower than 8800GTX)...

 

[UniversalTruth]

That's what I expect though, 68** slower or in a best case scenario equal in performance to the one-year-old 58**... So I really hope AMD is not using NVidia's poor naming methods. It is a kind of lack of taste or I don't know how to call it.... I hope Dave or whoever from AMD is thinking really seriously on this...