AMD: R9xx Speculation

[mczak]

Stock 460 1GB

Using 4xAF though not 16 - not that I think it'll make a difference .
So HD6870 being close to GTX460 1GB would be an improvement (over HD5870) but not by all that much (30% maybe). I guess AMD didn't lie when they said it's not much faster with extreme tesselation .
Without AA the gap between GTX460 and HD6870 should widen there.

 

[neliz]

How about if GTX 580 is GF100 based AND has 128 TMUs I'll paypal you $100? You deserve that much for sticking with the hair-brained story

deal

 

[psolord]

I had seen this slide


but I missed this one.


The first one was a fail imho, due to the 768 vs 1024MB comparison, but this one is uh-oh if true!

PS Borderlands does not have any AA options on my 5850....

 

[mczak]

I had seen this slide


but I missed this one.

I think the interesting thing here is that just by looking at the slides you get the impression the HD6850 is roughly 25% faster than the GTX460 768MB whereas the HD6870 is only 20% faster than the GTX460. Since there is only about a 10% difference between GTX460 768MB and 1GB (unless the former is memory limited) I definitely smell some carefully chosen benchmarks as the difference between HD6850 and HD6870 imho should really be larger than the difference between the two GTX460, and not the other way round...

 

[mboeller]

I had seen this slide



The first one was a fail imho, due to the 768 vs 1024MB comparison, but this one is uh-oh if true!

Why is it a fail? Nvidia decided months ago that the original GTX460 needs only 768MB. Therefore IMHO it is a valid comparison.

Also, as far as I know all the 1GB versions are custom versions of the GTX460. At least during launch it was said that the 768MB version is the "correct" version Nvidia had intended from the beginning. So if it is a fail in your opinion blame Nvidia.

 

[fbuffer]

Why is it a fail? Nvidia decided months ago that the original GTX460 needs only 768MB. Therefore IMHO it is a valid comparison.

Also, as far as I know all the 1GB versions are custom versions of the GTX460. At least during launch it was said that the 768MB version is the "correct" version Nvidia had intended from the beginning. So if it is a fail in your opinion blame Nvidia.

I'm not sure where the "fail" is or belongs.. it's "fail" to compare competing products of one similar price range ($150-225) that have different configurations (768MB v 1024MB) that may expose a weakness.. however it's "A-OK" to compare competing products of one another price range ($250-325, 399 v 499) that have different configurations (1024MB v 1280/1536MB) that may expose a weakness. Ahh ok... just so long as I understand this correctly.

 

[Broken Hope]

No idea how much price gouging is happening on these.

http://img155.imageshack.us/f/6800series.png/

 

[Jawed]

Farewell to ATI: Radeon HD 5000 Graphics Card Family Roundup

The premise of the article seems a bit comical, but the performance summary at the end is interesting, e.g. with HD5870 being 69-90% faster than HD5770, i.e. things seem to have matured nicely:

http://www.xbitlabs.com/articles/video/display/ati-radeon-hd-5000_12.html

The HAWX 2 "benchmark" also makes an appearance, "At this time tessellation cannot be activated on Radeon cards [...]" which is a puzzler. The terrain most definitely needs tessellation

 

[fbuffer]

No idea how much price gouging is happening on these.

http://img155.imageshack.us/f/6800series.png/

hard to tell , particularly given the HD5000 series which seems to be an anomaly (or so we hope) where launch prices where actually lower then as demand increased so did pricing (age old supply v demand). I would hope as a consumer that the HD5000's scenario was a one off however knowing retail as well as I do I have strong suspicions that retailers are foaming at the mouth. I know it's typical for brand new products to have a slightly higher "entrance" fee, such is the price for having the newest new bling thing. With the recent price cuts by nVidia to try and counter the HD6800 it will be interesting if ATI/AMD feels the need to respond. One one side AMD will probably feel comfortable with their set pricing as they are "new" products after all and seem to be competitive and for the most part prices are set well in advance of launches. On the other hand, AMD might feel they can do more (such as lower prices via manufacturer rebates, incentives etc) to gain more capital with the enthusiast community and possibly create greater demand v. nV's newly price cut products. What amazes me still is that ATI/AMD is able to compete rather well utilizing a chip that is half the size at nearly half the transistors.

 

[racca]

nVidia's gross margin is the same like AMD's. 50%+ of their revenue coming from the geforce business, "ATi" puts only 25% into AMD's revenue stream.
They make enough profit with their geforce business, because their gross margin must be in the range of 30%+.

Let me give you some (not accurate but true enough) numbers:

• GeForce series makes up around 2/3 of Nvidia's revenue, Radeon/Fire is 1/4 of AMD
• NVDA has a gross margin of 35% in 2009 while AMD has 42%.
• NVDA reported ~17% gross margin last Q, while AMD has ~45% for a few Qs now
• AMD claims 90% of DX11 and >50% discrete GPU market share

Considering Quadro has exceptionally high gross margin and makes up 1/3 of Nvidia's revenue. You can safely assume GeFore product line up to GF100-based has very very low margin if not negitive (ie. if Quadro have ~50% gross margin which is of course an understatement, GeForce will have 0%).

Even if GF104/106/108 product have the same gross margins (likely around 35%, I'm guessing) and propotion (~50% of AMD cards shipped last Q was DX11) AMD have, Nvidia would be very lucky to get a 30% margin this Q with GeForce still less than half that of Radeon.

 

[Florin]

Why is it a fail? Nvidia decided months ago that the original GTX460 needs only 768MB. Therefore IMHO it is a valid comparison.

Also, as far as I know all the 1GB versions are custom versions of the GTX460. At least during launch it was said that the 768MB version is the "correct" version Nvidia had intended from the beginning. So if it is a fail in your opinion blame Nvidia.

Wut? The reference circuit board already had the 256-bit version in mind.

 

[psolord]

Why is it a fail? Nvidia decided months ago that the original GTX460 needs only 768MB. Therefore IMHO it is a valid comparison.

Also, as far as I know all the 1GB versions are custom versions of the GTX460. At least during launch it was said that the 768MB version is the "correct" version Nvidia had intended from the beginning. So if it is a fail in your opinion blame Nvidia.

I'm not sure where the "fail" is or belongs.. it's "fail" to compare competing products of one similar price range ($150-225) that have different configurations (768MB v 1024MB) that may expose a weakness.. however it's "A-OK" to compare competing products of one another price range ($250-325, 399 v 499) that have different configurations (1024MB v 1280/1536MB) that may expose a weakness. Ahh ok... just so long as I understand this correctly.

Part of my definition of fail and one of my personal rules when evaluating graphics cards, goes like this:

You are a graduate for the fail award, if you benchmark a graphics card, on settings that are not realistic for the purpose and target audience it was built for. Ie, Uber high resolutions with 8XMSAA, on cards that will end up framebuffer limited. Whenever a framebuffer limit occurs, during a benchmark, this result should be discarded and a note that the card got framebuffer limited, be posted instead.

 

[Florin]

The HAWX 2 "benchmark" also makes an appearance, "At this time tessellation cannot be activated on Radeon cards [...]" which is a puzzler. The terrain most definitely needs tessellation

Oh boy, better duck for incoming damage control spin *runs for cover*

 

[SimBy]

Oh boy, better duck for incoming damage control spin *runs for cover*

You mean Ubisoft deciding to screw 85% of DX11 market?

 

[ZerazaX]

What'll be interesting is when people bench the 6870 vs. the 5850 at the same clocks. It will be interesting to see how much efficiency is actually to be had

 

[Florin]

You mean Ubisoft deciding to screw 85% of DX11 market?

You'd probably do well to avoid attributing to malice that which is adequately explained by stupidity.

 

[Zaphod]

Whenever a framebuffer limit occurs, during a benchmark, this result should be discarded and a note that the card got framebuffer limited, be posted instead.

Yes. Perhaps it could even be standardized across review sites. I'd suggest Framebuffer Amount Is Limiting, or some sort ot initialization thereof.

 

[fbuffer]

You mean Ubisoft deciding to screw 85% of DX11 market?

I thought it was 90% .. ?

 

[Gipsel]

Yes. Perhaps it could even be standardized across review sites. I'd suggest Framebuffer Amount Is Limited, or some sort ot initialization thereof.

 

[SimBy]

I thought it was 90% .. ?

I realize AMD says it's 90% I was using Steam DX11 survey numbers.