Another AMD leaked slide shows upcoming BARTS is based on Radeon 5xxx architecture...

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

happy medium

Lifer
Jun 8, 2003
14,387
480
126
The link you posted mentions: "That Barts will be offering good performance for a mid-range circuit is pretty obvious, but exactly how good is not as certain, but more information should appear very soon."

Therefore, we have no idea how much faster 6870 will be compared to 5850 or GTX460 . That article didn't mention anything about 8% hehe.


Quote from article.
"According to other sources to NordicHardware AMD will start shipping reference cards to manufacturers and partners this week, for testing and evaluation. We can confirm that several manufacturers have tested the card and performance is said to be on par with Radeon HD 5850, but we will have to return with more details on the performance of the new mid-range cards.?

The 5850 is 8 percent faster, so if this card is "on par" with a 5850, I assume it's also about 8 percent faster. Fair enough?
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
How low can NV go with their gtx460 pricing to counter that? I'd have to say $200 for a 1gb gddr5 ~340mm2 card, packaged and shipped, is a slight loss.

How much did you say Nvidia pays per wafer vs ATI? Unless you can answer that, you have no idea how much ,or even if the gtx 460 is losing money.

Barts is much smaller and can be priced however ATI wants to compete.

I didn't see any official chip size for the Bart. Please link me.

you get a 5850 with enhanced dx11

You mean direct x 11 performance to finally to the level of Nvidia.

You were doing real well until you started to spit out rumors ,guesses, and opinions as fact.:)
 

Skurge

Diamond Member
Aug 17, 2009
5,195
1
71
Quote from article.
"According to other sources to NordicHardware AMD will start shipping reference cards to manufacturers and partners this week, for testing and evaluation. We can confirm that several manufacturers have tested the card and performance is said to be on par with Radeon HD 5850, but we will have to return with more details on the performance of the new mid-range cards.?

The 5850 is 8 percent faster, so if this card is "on par" with a 5850, I assume it's also about 8 percent faster. Fair enough?

On par with a 5850 with beta or whatever drivers they are using and without finalized clocks. Or this is an underestimate from AMD again.
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
On par with a 5850 with beta or whatever drivers they are using and without finalized clocks. Or this is an underestimate from AMD again.

Agreed ,the gtx 460 just recieved its second driver, it will improve also just like the gtx 470 caught up with the 5870.

Skurge got to bed, what time is it where your at?
 

Skurge

Diamond Member
Aug 17, 2009
5,195
1
71

edplayer

Platinum Member
Sep 13, 2002
2,186
0
0
Another thread states the mid-range 6870 (A.K.A 6770) will only be as fast as a 5850. So thats only 8 percent faster then a gtx 460. It better launch under 200$ to be competative vs the gtx 460. The gtx460 will be far below the 200$ mark by then.


If this card is about 8% faster than the GTX 460 1GB, why should it launch at under $200?

I understand your point about it being competitive with the 460 but I believe that 460 prices were recently adjusted and the 1GB model is at $220. I think AMD could do well with pricing near the 1GB 460 including going over it. I'll also add that I think the 1GB 460 will be at $199 when the new AMD card is out. Maybe lower if the AMD card starts at less than $199 but I don't see a reason to sell it "far below" that price before the card is even released.
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
If this card is about 8% faster than the GTX 460 1GB, why should it launch at under $200?

I was assuming the gtx 460 prices would fall in the next 6 weeks. Seems logical.
Amazon had a Asus gtx 460 1gb for 175$ shipped no rebates with a 1 to 2 months delivery time. Thats a good hint of whats to come.:)
 

Barfo

Lifer
Jan 4, 2005
27,554
212
106
I don't really care if Barts is a rebadged Radeon 9800 pro on 40 nm as long as it has the rumored performance at the rumored price.
 

dangerman1337

Senior member
Sep 16, 2010
333
5
81
With the uncertaintiy with the 6***, i am the only one thinking of the possibility that the cancelled TSMC 32nm that it was orginally going to be based on is a lie and it will be based on 32nm :p?
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
With the uncertaintiy with the 6***, i am the only one thinking of the possibility that the cancelled TSMC 32nm that it was orginally going to be based on is a lie and it will be based on 32nm :p?


I think AMD has been really smart w/40nm. With what they learned from the 4770 they were able to design Evergreen almost flawlessly for the process. A bit low on yields at first, but having fully functioning chips right off the bat was never a problem. Possibly, with what they learned from Evergreen, they had enough knowledge to do 32nm NI on 40nm and pull it off. nVidia seems to still be struggling mightily w/40nm. No, as in zero, fully functioning chips still! People have been waiting for 485, then 475. Now they're waiting for 455?

nVidia might have to go into AMD cpu vs Intel cpu mode and just sell their cards for whatever the market will allow, just to keep treading water until they figure this out. Hopefully, whatever they have planned for 28nm is really good. If they are just planning a Fermi shrink I don't think it'll hold up at all to AMD who will likely have a whole new generation card for it.
 

dangerman1337

Senior member
Sep 16, 2010
333
5
81
I'm not sure if nvidia can compete with AMD on 40nm if that rumor about the new arch is bad enough, Fermi missed its chance being very powerful at the size it is at or being a lot more effeicent but it seems most of the people at nvidia have a hard on for shaders (CUDA cores) instead of anything else.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
My point being is that the 6XXX cards were designed and finalized probably before AMD knew much of anything about the 460 at all, therefore saying the 6XXX series is a reaction to the 460 is probably not true.

This part I agree with... however, it would seem that TSMC is having trouble getting to its next process shrink after 40 nm.
So the theory that 6xxx was originally planned on a smaller node, and the 6xxx we will now be getting is instead something completely different, closer to a 5xxx-refresh (due to being restrained on the larger 40 nm process)... I think that has some validity to it.

nVidia is probably going through a similar process right now... both companies probably drew up their roadmap years ago, assuming TSMC would have a 28 nm process (or something to that effect) ready about now, and had to change their plans to stay on 40 nm longer.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
I am not making this up. The code name Fermi refers to an architecture. If GT100 is the first generation(chip) based on this architecture, then gt104 is the second generation(chip) based on this architecture. While the word "Generation" is arguable, but the fact stays. GF100 was planned to be released before the end of 2009 and GF104 at the 2nd quarter of 2010. GF100 was 6 month off the mark but not GF104.

Fermi refers to the current generation/family of architectures.
The 'F' in GF100/GF104/GF106 refers to 'Fermi'. So they are all part of the Fermi family, even though at a micro-architectural level they are not entirely equivalent.

The previous generation/family was called 'Tesla', that's what the T in GT2xx was for (the code names that is, not the product names).
Anandtech mentioned that:
http://www.anandtech.com/show/2549
The GT stands for "Graphics Tesla" and this is the second generation Graphics Tesla architecture, the first being the G80.

Prior to that, nVidia used a different naming scheme.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
You are confused. GF104 to GF100 is what Juniper is to Cypress. Both companies create different chips based on the same architecture to target different markets (i.e., high-end and mainstream and low end).

That is not correct.
AMD and nVidia have different strategies.
AMD takes a single architecture and scales it up and down verbatim across their entire line (with the exception of IGPs in some respects).
Juniper pretty much IS 'half a Cypress'.

GF104 however is NOT 'half a GF100'.
They don't share the exact same architecture. GF104 has a superscalar instruction scheduler, and a different balance of execution units per core.
If you read Anandtech's launch review on GTX460, you'll see the differences explained in detail:
http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king

This also explains why AMD releases their product line top-to-bottom at pretty much the same time, and nVidia has a gap of a few months between GF100 and GF104/106.
 

GaiaHunter

Diamond Member
Jul 13, 2008
3,628
158
106
Again, I didn't say "460 is making a hole in AMD's bank", but there exists an indication of a declining rate of sales on 5xxx series in the Dx11 segment while 460 is the sharpest one with a 2.86% growth.

There is no indication of a declining rate of sales (at least not in the steam data and not for the last few months) for the 5xxx series in the DX11 segment. There is a decline of market share which is not the same.

Since more DX11 cards started to appear the 5800 series has been on a decline of DX11 market share and the 5700 series went on a decline shortly after their launch too.
 
Feb 19, 2009
10,457
10
76
How much did you say Nvidia pays per wafer vs ATI? Unless you can answer that, you have no idea how much ,or even if the gtx 460 is losing money.

You were doing real well until you started to spit out rumors ,guesses, and opinions as fact.:)

That was just a commentary actually, no facts, just my speculation that a if they drop the 1gb gtx460 (obviously crippled to increase yields and help with the tight margins, no full fermi yet) to $175 as you have suggested, its not a money maker.

Now, you can speculate on this if you wish.. why do you think NV is dropping prices like mad all of a sudden on their best perf/price GPU, one that is a recommended buy from pretty much all reviewers?
 

Piotrsama

Senior member
Feb 7, 2010
357
0
76
With the uncertaintiy with the 6***, i am the only one thinking of the possibility that the cancelled TSMC 32nm that it was orginally going to be based on is a lie and it will be based on 32nm :p?

TSMC can't lie like that to their share holders.
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
That was just a commentary actually, no facts, just my speculation that a if they drop the 1gb gtx460 (obviously crippled to increase yields and help with the tight margins, no full fermi yet) to $175 as you have suggested, its not a money maker.

Now, you can speculate on this if you wish.. why do you think NV is dropping prices like mad all of a sudden on their best perf/price GPU, one that is a recommended buy from pretty much all reviewers?
I think it's a realistic scenario. No company cuts 15-20% directly out of its margins for laughs, so something is up with the strategy NVIDIA is taking with the GTX 460. I don't know if they're selling it at a loss, I'd doubt it, but I think they're significantly cutting their margins to gain marketshare and at least make some money for their AIB's. Basically, you summed up Barts very well - you have a card that's superior in every metric to anything else within it's market segment and it's coming out in less than a month. The deciding factor will be how much of a premium AMD charges for this technology, and that will depend on their goals.
 

Ares1214

Senior member
Sep 12, 2010
268
0
0
Yeah, but he's just getting started. Give him some time to moult.

Silver just seems to know a lot of stuff, or just be doing the basic math. We already know most of the specs of the 6770, as that one slide was either a fake, or was going a long with the naming scheme of a new 5770 rev being called the 6770, and 6770 being called 6870. The 460 actually took a large drop in price, and now its right back up there. NV is likely trying to make as much as possible before 6xxx, knowing the popularity of the 460. Most of what we know about the 6770 puts it at $180-240, more or less 460 price range, border line 150 watt, just like the 460, and temps at who knows, but amd has seemed to put on a new stock cooler, so might be better. Then, performance puts it at 5850-5870 range, so right around GTX 470. Id consider that a win if you ask me, 460 will have to drop to way below $200 to keep up, and even then AMD can just let the prices on the 5850 and 5870 plummet to keep pushing the 460 down and get them out of their warehouses.
 

Seero

Golden Member
Nov 4, 2009
1,456
0
0
Fermi refers to the current generation/family of architectures.
The 'F' in GF100/GF104/GF106 refers to 'Fermi'. So they are all part of the Fermi family, even though at a micro-architectural level they are not entirely equivalent.

The previous generation/family was called 'Tesla', that's what the T in GT2xx was for (the code names that is, not the product names).
Anandtech mentioned that:
http://www.anandtech.com/show/2549

Prior to that, nVidia used a different naming scheme.
The first paragraph of http://www.anandtech.com/show/2549
The chip is codenamed GT200 and it's the successor to NVIDIA's G80 and G92 families. Why the change in naming? The GT stands for "Graphics Tesla" and this is the second generation Graphics Tesla architecture, the first being the G80.
G80 = the first generation of the GT architecture.
G92 = the second generation of the GT architecture.
Each of them has a family and also belongs to the GT achitecture family.

Fermi architecture
GF100 = the first generation of GF architecture
GF104/106/108 = the second generation of GF architecture

Again, i don't see the constructive reasoning on picking the use of the word "generation", and to defend myself may lead to more nonconstructive argument. I will stop here on this particular matter.
 
Last edited:

Scali

Banned
Dec 3, 2004
2,495
0
0
I think the strongest argument against the use of 'generation' within Tesla/Fermi is that GF100, GF104 and GF106 are not different generations of products that address the same target (price-range) in the marketplace.
They represent different levels of the same basic technology/generation (DX11-class hardware) operating in the same market at roughly the same time (mainstream, high-end, enthusiast).

I think generation is something a bit bigger (and lasts longer than a few months). GT200->GF100 is a generation.
GF100->GF104 is trickling new technology down from the high-end into the mainstream, combined with a few architectural updates.

But as I also said, you are correct in stating that GF100 and GF104 are not the exact same architecture (I corrected RussianSensation on that).
 
Last edited: