Radeon HD 6000 is a minor improvement

[Sam_oslo]

What is so hard? Now that 32 bulk is cancelled on tsmc and gf, 28 is next node. There is two possible things for amd to do. Improve on existing node (that they are doing) or wait for 28. How can you even suggest that just waiting would be any good vs improving on current node?

I'm not suggestion that they should Waite, I've said a shrink is always a welcome move. Because it offers less power usage, heat and more performance, etc. .

The hard part would be to make a good enough tweak worthy for calling it next gen.

Your talking like Nvidia does not have the same manufacturing delays (due to TSMC) like AMD has.

Both of them are in the same mess, delayed 28nm and cancelled 32nm. So both will have a delay in releasing a new uber-high-end single GPU. Both are waiting for a successful 28nm from TSMC.

As said, If ATI is going to settle with just tweak for next gen, then the result will boil down to the flexibility of their existing architecture. The same is true for nVidia.

It is too early to speculate on the performance of such a move, but probably they can't make a big improvement as expected for a new generation. But GF104 has proven the flexibility of Fermi and it can give nVidia a chance to take on, actually. It will be interesting to see what happens next.

Do I have to repeat myself for clearing what people put in moth, or could we all please talk about our own ideas than pithing words in other mouth, please?

[Florinmocanu]

no, i disagree. a future GF104 with full blocks will probably be around 5870 levels, at aproximately the same die size.

I cannot see Nvidia having more flexibility with GF 104 than AMD with Cypress, there in the same ball park.

How they tweak the architecture will be the key issue, that's where we may see who has the upper hand.

[Sam_oslo]

If I understand you right, you are saying the same as me, and you agreed on the "mess" in your last post too , but whatever you call it agreement or diagrammed, lets agree that flexibility of their existing architecture will play a big role in next gen.

I personally mean Fermi has shown a good flexibility with GTX 460, and a full blown 384 GF104 is expected shortly (according to rumors). This means Fermi is flexible and we can expect further improvements, in my opinion. I see we disagree on this part, but that's OK, nobody knows for sure and can say anything for sure about the future.

[Florinmocanu]

well, ofc GF 104 has room for improvements.

That's why i'm saying how AMD/Nvidia manage to tweak their arch's will determine the outcome.

[Boissez]

If I understand you right, you are saying the same as me, and you agreed on the "mess" in your last post too , but whatever you call it agreement or diagrammed, lets agree that flexibility of their existing architecture will play a big role in next gen.

I personally mean Fermi has shown a good flexibility with GTX 460, and a full blown 384 GF104 is expected shortly (according to rumors). This means Fermi is flexible and we can expect further improvements, in my opinion. I see we disagree on this part, but that's OK, nobody knows for sure and can say anything for sure about the future.

ATI rolled out a full line-up from lowend to enthusiast in just 3 months - by that yardstick ATI's current arch is more than flexible enough.

If ATI manages to up the HD6870 to 1920SP with slightly higher clocks it should be more than capable enough to give a full 512 fermi a run for its money. A HD6770 with 960SP should also be able to nip at the heels of the GTX460 while being much cheaper.

[-Boris-]

exactly

Would Nvidia be able to catch up ? i wonder

It's unlikely, considering how far behind nVidia is on performance/mm².

At roughly the same die-size of ATis performance 5870s nVidia makes a mainstream card.

HD 6000 will most likely increase performance per mm² a bit more. Or else it's a failure.

[Manicdan]

so the 4000s came out a long time ago, since they weve only seen addition of dx11 and eyefinity. what does ati plan for the next card? theyve had a really long time to think about it, and i hope they used that time properly.

[Wesker]

Here we go with the "mess" again .Oh well.

It is a mess.

ATI was betting on a smooth transition to 32nm, in order to offer a refresh of the Evergreen family (Northern Islands). ATI is always the first to make a jump to a new node, and the cancellation of the 32nm TSMC node means that ATI is now behind schedule. Typically, Nvidia's process node plans are no where near as aggressive as ATI's, so the 32nm fiasco hasn't affected Nvidia to the same extent.

[Particle]

This news is not news of course. They've been telling us this for months. Due to process availability, we knew to expect a tweaked core and vastly improved support logic. As of a little while ago we were told it might come with six clusters instead of five (1920 SPs vs 1600). That's it.

I don't see the cause for scandal. It'll be just like they've been telling us. I'd expect it to be like the Radeon HD 3000 series was.

[cegras]

Either way, ATI releasing new core tweaks in SI can give it valuable experience on how these new parts will perform as part of NI. In the end they gain experience for the transition.

[informal]

I think they will just up the stream processor count in NI and keep the same SIMD unit design from SI.With 28nm they can go really high on the stream processor count(2x the Cypress is reasonable assumption).

[Manicdan]

how about the memory? think they will be using 256bit on both SI and NI? not sure things will be that much faster in the next 8 or so months, a 28nm chip should easily be 2x faster than a 5870, but ram might only be 20-30% faster?

[-Boris-]

It is a mess.

ATI was betting on a smooth transition to 32nm, in order to offer a refresh of the Evergreen family (Northern Islands). ATI is always the first to make a jump to a new node, and the cancellation of the 32nm TSMC node means that ATI is now behind schedule. Typically, Nvidia's process node plans are no where near as aggressive as ATI's, so the 32nm fiasco hasn't affected Nvidia to the same extent.

Both GTX 280 and GTX 480 was designed very aggressively and on the very edge of what's possible on their nodes. And without being able to do a 32nm refresh of the GTX 480 like they did on the GTX 280, I think they are clearly the most aggressive ones when it comes to node plans. Since they even produces chips that might work on the next node already at the node before.

ATi had a backup plan, they have both NI and SI, so it's not a mess at all, they had a plan for this all along. nVidia on the other hand is knee deep in since they already have a lineup that could need 32nm or 28nm.

The next move from ATi is here within 5 months, at 40nm, and it's a planned move. What is nVidias next move?

[Chumbucket843]

how about the memory? think they will be using 256bit on both SI and NI? not sure things will be that much faster in the next 8 or so months, a 28nm chip should easily be 2x faster than a 5870, but ram might only be 20-30% faster?

http://en.wikipedia.org/wiki/Memory_wall#Memory_wall

[Eastcoasthandle]

Yeah you ninja edit now to "explain" how a mess is not a failure.Ok,thanks,bye.
AMD didn't even make a public statement about any performance metric of their new GPU and they are somehow messed up.Oh and the "wonder" new GF chip has them worried so much that they somehow,magically,rushed out a non-impressive new GPU (we know next to nothing about).Right.

No, they didn't make any statement about performance, etc. Wow, some people take this sort of thing really personal. At this time we really don't know what was actually done to the 6000 series arch to call it anything. We can only take this news with a grain of salt. No different then any news regarding new unreleased gpus.

[-Sweeper_]

how about the memory? think they will be using 256bit on both SI and NI? not sure things will be that much faster in the next 8 or so months, a 28nm chip should easily be 2x faster than a 5870, but ram might only be 20-30% faster?

cypress is not BW limited, so i wouldnt mind 256-bit bus with high-speed GDDR5 chips

[RaZz!]

i thought this was what we all knew ages ago: southern islands is a tweaked cypress whereas northern islands is a new arch.

[Eastcoasthandle]

I think it's clear that with SI then NI ATI is thinking (from what we know) 2 steps ahead right now. We just have to wait and see what kind of performance they bring.

[herderien]

i would be surprised by ATi releasing a minor core change with a major version change (5xxx to 6xxx) they usualy use de x870 to x890 naming scheme in those cases.
ie 5890 and 5990.

or it would be a "minor" change with support of DX11, thus explaining the 6xxx naming ?

[Calmatory]

cypress is not BW limited, so i wouldnt mind 256-bit bus with high-speed GDDR5 chips

Haha, right. Define "BW limited"?

[-Sweeper_]

Haha, right. Define "BW limited"?

when overclocking your RAM from 1200mhz to 1350mhz makes no difference in performance even at high resolutions (>1080p)

[naokaji]

i would be surprised by ATi releasing a minor core change with a major version change (5xxx to 6xxx)

Maybe they just want to distance the new products from the old ones cause they don't want them to be seen as part of a series that came out like a year ago.

[Eastcoasthandle]

i would be surprised by ATi releasing a minor core change with a major version change (5xxx to 6xxx) they usualy use de x870 to x890 naming scheme in those cases.
ie 5890 and 5990.

or it would be a "minor" change with support of DX11, thus explaining the 6xxx naming ?

Don't forget the 4870 to 4890 with the ring bus (if I recall correctly).

[Mad Pistol]

You guys forget how efficient Cypress is as an architecture. It wouldn't be that hard to wring some more performance out of it by slightly increasing the die size and Stream Processor count. The cards idle low, and their temperatures top out fairly low as well. Not to mention, their cooling solutions, while impressive, are nothing like the one seen on the GTX 480. As soon as ATI employs something a little more extreme, we'll see just how far they can push their current gen architecture.

ATI still has a lot of room to improve before they move to a completely new architecture. The 5850 and 5870 were extremely potent and efficient parts. Just make them faster, add a few more stream processors, increase the efficiency of the core logic, and you've got a card that is again faster than anything Nvidia currently has.

[Mechanical Man]

Don't forget the 4870 to 4890 with the ring bus (if I recall correctly).

Eh no.