AMD: R9xx Speculation

I think it's reasonable to expect NI to be more bandwidth-efficient than Evergreen. That would tend to indicate an increase in ROPs but not an increase in bandwidth.

The indication is 75% more bandwidth for Barts in comparison with Juniper, which seems to make doubled-ROPs inevitable. At least, if the ROPs are basically the same as Evergreen's.

It seems to me that Barts is aimed to be 30-50% faster than Juniper. Seems awfully wasteful of extra bandwidth and extra ROP capability :???:

---

The roadmap shows 4 product segments, Antilles, Cayman, Barts and ... I presume Turks or Caicos fits in there as Juniper's replacement. Call it Turks for the sake of argument.

I guess that Turks would aim for 20-30% more performance than Juniper. Probably with the same ROPs/Bandwidth configuration.

So the puzzling thing is why would Barts exist with barely more performance? Barts Pro seems likely to be the same performance as Turks XT.
 
It seems to me that Barts is aimed to be 30-50% faster than Juniper. Seems awfully wasteful of extra bandwidth and extra ROP capability :???:


Juniper; HD5770 ~X4500
BartsPro; HD6850 ~X6500 (if the leaks are true)
BartsXT; HD6870 ~X7500 (same disclaimer as above)

7500/4500 => BartsXT = 166% x HD5770
 
I think it's reasonable to expect NI to be more bandwidth-efficient than Evergreen. That would tend to indicate an increase in ROPs but not an increase in bandwidth.

The indication is 75% more bandwidth for Barts in comparison with Juniper, which seems to make doubled-ROPs inevitable. At least, if the ROPs are basically the same as Evergreen's.

It seems to me that Barts is aimed to be 30-50% faster than Juniper. Seems awfully wasteful of extra bandwidth and extra ROP capability :???:

---

The roadmap shows 4 product segments, Antilles, Cayman, Barts and ... I presume Turks or Caicos fits in there as Juniper's replacement. Call it Turks for the sake of argument.

I guess that Turks would aim for 20-30% more performance than Juniper. Probably with the same ROPs/Bandwidth configuration.

So the puzzling thing is why would Barts exist with barely more performance? Barts Pro seems likely to be the same performance as Turks XT.

I think Turks is more likely to end up 0~15% faster than Juniper, but a good bit smaller, and perhaps less power-hungry as well.
 
I'm disappointed that Cayman XT only has 1GB of VRAM.

As for this,
83317390.jpg


I'm looking at that with the y postioning indicating pricing postition, since the gaps are too small for performance. If that is the case Cayman and Antilles will both be more expensive than Cypress and Hemlock. Somewhat expected, but still a little bit of a let down.
 
I'm disappointed that Cayman XT only has 1GB of VRAM.


I'm looking at that with the y postioning indicating pricing postition, since the gaps are too small for performance. If that is the case Cayman and Antilles will both be more expensive than Cypress and Hemlock. Somewhat expected, but still a little bit of a let down.

I would assume a similar pricing as last generation.
seems likely.
 
Cayman's price is probably not set in stone yet, anyway. Between now and late November, the market could change quite a bit.
 
That slide is referencing performance not pricing. Rumored pricing for Barts XT ~$250, and Barts Pro ~$200.

Honestly, if Barts Pro is somewhere between the GTX 460 and HD 5850 as rumors place it (which is to say barely faster than the 460) it won't sell very well at $200.
 
6Gbps = 192GB/s, or about 25% more than Cypress. Performance gain is hopefully in excess of that.

MBA? Manufactured board assembly?

Are those actual speeds or just memory chips :?: Barts XT and PRO on those slides has 5Gbps. Cayman PRO has also 5Gbps on the slides.
That would mean same 160 GB/s for Barts XT.PRO and Cayman PRO. Thats unlikely me thinks.:rolleyes:
 
That slide is referencing performance not pricing. Rumored pricing for Barts XT ~$250, and Barts Pro ~$200.

If you are referring to Redvi's post, do you think Redvi's explanation was wrong?

If Cayman XT was so close to the Cypress XT, what would be the point of R&Ding them and releasing them?

This is the performance slide.



On a sidenote, if these price/performance slides are true, it seems AMD is giving more performance/price in their higher end this time. I mean the 5850 launched at 259$ and the Barts XT seems to be close to that with maybe slightly better performance. So no real gain for the customer.
 
Are those actual speeds or just memory chips :?:
I wondered that, too, but forgot to put it in my earlier posting. I don't see any way to tell, in the end.

Juniper and Redwood are both 128-bit and Juniper appears to be about 70% faster (with about 56% more die space). But that's a doubling of ROPs and cores.

I suppose we could use the same formula for Cayman and Barts: 256-bit bus, doubled ROPs and cores.

What do we think the die increase for Cayman is in comparison with Barts? Barts is anything in the range 220-280mm²? Cayman is in the range 360-420mm²? That's a big range, 29-82%. I can't help seeing 50-60% more area as extremely likely.

Anyway, currently I'm dubious that AMD would put only 16 ROPs in Barts (but it did put only 8 ROPs in Redwood along with the 128-bit bus - though DDR3 muddies things there). But then I was also dubious about there being two 256-bit chips. It seems that AMD wants to have a $150 low-end 256-bit chip and have 3 SKUs above that, while being unwilling to use more than 256-bit.

Well, we don't know that Cayman has only 256-bit.
 
I
If Cayman XT was so close to the Cypress XT, what would be the point of R&Ding them and releasing them?

This is the performance slide.


It's just a slide. A 5% higher position on that slide does not necessarily mean Cayman XT is only 5% faster than a GTX 480. If Cayman is 15% faster than GF100, that would mean it's about 30% faster than Cypress. Considering they're still on the same process, that would be an acceptable performance boost, in my opinion.




On a sidenote, if these price/performance slides are true, it seems AMD is giving more performance/price in their higher end this time. I mean the 5850 launched at 259$ and the Barts XT seems to be close to that with maybe slightly better performance. So no real gain for the customer.
Until we know exact prices and performance we can't tell for sure. The 5850 is ~10-15% faster than the 460-1G. If Barts XT is 10% faster than a 5850, that would amount to a comfortable lead over the 460. A price of 249$ would be ok in that case.

I think they'll wait until christmas shopping season is over before they lower any prices, because production ramp up won't happen over night, and if demand exceeds supply anyway, lowering prices too early would be counter-productive.
 
Back
Top