Trinity vs Ivy Bridge

Really? From the scarce numbers I could find Trinity even on the GPU side was much closer in performance (near parity) with Ivy Bridge at 17W, whereas obviously with the 100W chips it is faster. Love to see some really good performance comparisons though... there seem to be almost no 17W Trinity numbers online.

Sure, it's doing better relative to Intel in that market segment than it is relative to Intel in the desktop segment. Not better than Intel overall, but certainly much closer than they were in the desktop space, IMO.

They can't compete with Intel on all fronts. And they certainly can't compete in the desktop segment currently. Mobile and ultra portable APUs seem to be their best bet to remain competitive with Intel at the moment.

Regards,
SB
 
Intel 9.18.10.3071(15.31) drivers has been released.

AVDudnN.jpg


32bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProdId=3498&DwnldID=22610

64bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProdId=3498&DwnldID=22605

Update: Intel download page says 9.17 but the driver installer says 9.18 is the real version number.

http://www.geeks3d.com/20130403/int...new-opengl-extensions-and-opencl-1-2-support/
 
Last edited by a moderator:
And that driver does not support Sandy Bridge.
Actually, the sentence "Built for HSW, but now available on IVB" sounds to me like "You should be grateful you don't have to buy HSW to receive mainstream support". But that's just IMHO.
They probably don't have the resources to keep doing substantial work on GPUs older than two years. On the other hand, I seem to recall reading somewhere that IVB&HSW have similar architecture, so it should be possible to keep Ivy fully supported (not just bugfixed) as long as HSW is.
 
Yup, the graphics part in Ivy went architectural changes compared to Sandy and given that fashion to abandon the support of older and not so interesting products, it is understandable from Intel.
Let them focus on the new products- perhaps they will be able to extract more performance and better support.
 
Last edited by a moderator:
Yup, the graphics part in Ivy went architectural changes compared to Sandy and given that fashion to abandon the support of older and not so interesting products, it is understandable from Intel.
Let them focus on the new products- perhaps they will be able to extract more performance and better support.
Well they manage unified driver in linux from the very first i965 to haswell.
Already ditching even Sandy Bridge seems quite aggressive. That would be equivalent to AMD only supporting SI and Nvidia only supporting Kepler (Ivy Bridge is not even one year old, not to mention plenty of Sandy Bridge cpus are still sold). Though I guess older chips might still get driver updates, just not in the same package and probably more focused on bug fixes only.
 
The whole notion of "unified drivers" is a bit of a lie anyways. Just because you package the driver update in a monolithic file doesn't mean you're updating the portions that apply to older architectures, let alone even installing the same DLLs.

Don't be fooled, there's very few driver updates that affect parts older than 2 years in any substantial way from any of the major IHVs; read the changelogs. At best moderate/major bugs will be fixed, but there's almost no performance work that happens with older architectures.
 
One important aspect is gaining compatibility with new OS or display model or graphics infrastructure, about once every three years on Windows and, well, about every six monthes on Linux (or less?).

What if the architectures are really close, though?, maybe Ivy and Haswell are almost the same. Geforce 6 and 7 are almost identical, Geforce 8 and 9 (and GTX 2xx) too, Radeon 5000 series have been saved from driver extinction from now (maybe it helps that radeon 6770 is strictly equal to 5770)
 
Radeon 5000 and 6000 also have Llano and Trinity support perhaps keeping their drivers alive. I imagine they want Trinity in particular to look good yet.
 
The whole notion of "unified drivers" is a bit of a lie anyways. Just because you package the driver update in a monolithic file doesn't mean you're updating the portions that apply to older architectures, let alone even installing the same DLLs.

Don't be fooled, there's very few driver updates that affect parts older than 2 years in any substantial way from any of the major IHVs; read the changelogs. At best moderate/major bugs will be fixed, but there's almost no performance work that happens with older architectures.
Yes I agree having unified driver source code would be what's really required. But I guess if you'd have that there would be little reason to ship non-unified packages (but the opposite isn't true).
But well the linux drivers do that. Granted there's definitely some separate code there too of course, but depending on where the changes are definitely all chips benefit.

What if the architectures are really close, though?, maybe Ivy and Haswell are almost the same. Geforce 6 and 7 are almost identical, Geforce 8 and 9 (and GTX 2xx) too, Radeon 5000 series have been saved from driver extinction from now (maybe it helps that radeon 6770 is strictly equal to 5770)
For intel igp, it looks like i965 through Clarkdale (gen4, gen5, including "gen4.5") share most code, as do
Ivy Bridge and Haswell. Sandy Bridge (gen6) seems to be a bit alone there, but still quite large parts are shared (and obviously all high level parts like the glsl compiler are shared in any case).
 
Is Cinebench heavily SIMD limited? In that scenario the A6 would have more oomph than the A8.
Cinebench should be near exclusively simd limited. Though "simd" probably isn't the right term since IIRC it is all scalar ops, but in any case using simd units.
 
Radeon 5000 and 6000 also have Llano and Trinity support perhaps keeping their drivers alive. I imagine they want Trinity in particular to look good yet.

Nitpicking a tiny little bit, I believe both Llano and Trinity are Radeon 6000 - they'd be variants of the Radeon 6870 and 6970, respectively.
Of course the difference between a 5870 and 6870 is small (one doubled cache that relates to the tesselator, and what else)

Yep, and they've even got Richland, which has yet to reach market.
 
Last edited by a moderator:
Beema is Jaguar+ IIRC. Probably something in the vein of the move from Brazos to Brazos 2.0 or something like that, which is supposed to happen next year.
 
Beema is Jaguar+ IIRC. Probably something in the vein of the move from Brazos to Brazos 2.0 or something like that, which is supposed to happen next year.
Well I'd guess it's a bit more than what Brazos 2.0 was (which you'd never have noticed it was something new until they told you...). Brazos 2.0 was a late addition to the roadmaps due to canceling of Krishna/Wichita, but Beema is a regular scheduled chip so I don't think it will end up _that_ similar (even though both cpu and gpu arch will probably remain mostly the same). I guess there's a possibility it might support something like gddr5m even which alone would make it a worthy successor.
 
Last edited by a moderator:
Back
Top