AMD to launch new flagship Radeon HD 7970 desktop graphics card on December 22.

[mattdallastx]

It's simple; no one's challenging you on your purchases or priorities so don't imply that folks who don't share them are lazy or undisciplined.

Actually you implied that.

 

[ddarko]

don't imply that folks who don't share them are lazy or undisciplined.

Actually you implied that.

Hmm...

$150 is "a little extra" If you have a job and the self control to save the extra amount.

Your post speaks for itself. Back to topic.

 

[thilanliyan]

and asked if that applied to when she picked out her wedding dress. Needless to say I got what I wanted.

Lol, I gotta remember that one.

 

[3DVagabond]

Radeon HD 7970 close to HD 6990

http://www.fudzilla.com/graphics/item/25259-radeon-hd-7970-close-to-hd-6990

Now it appears Fud thinks its around as fast as a 6990.


Im still standing by my own logic of (based on rumors of cards spec's):

35% more shaders + 13% higher clocks = +53% gains
+50% more memory bandwidth = +50% gains

= card thats over 50% faster than last gen card (the 6970) = ~35% faster than 580.

That's just typical rumor-media repeating everything they hear and making up the rest. If it's right, they reported it first. If it's wrong, then it's don't shoot the messenger. Since they don't name the source, then it doesn't have to be true. They could be doing the same as you are and extrapolating performance figures from leaked specs.

I'm curious to see how the new arch is going to effect performance. Some people are saying because it's better at gpgpu that it won't do as well with rendering games. Fermi doesn't seem to have any issues rendering though. So, I'm not to sure why it's being put out as a truism.

The VLIW arch had seemed to have reached a wall. More shaders @ higher clocks weren't bringing the performance return you'd expect. Fermi seems to scale better. Hopefully GCN will be more like Fermi in that respect.

 

[AtenRa]

I'm curious to see how the new arch is going to effect performance. Some people are saying because it's better at gpgpu that it won't do as well with rendering games. Fermi doesn't seem to have any issues rendering though. So, I'm not to sure why it's being put out as a truism.

The VLIW arch had seemed to have reached a wall. More shaders @ higher clocks weren't bringing the performance return you'd expect. Fermi seems to scale better. Hopefully GCN will be more like Fermi in that respect.

VLIW architecture is more efficient in performance per die and performance per watt for gaming. It lacks performance efficiency in GPGPU, DX-11 compute (Tessellation) and OpenCl.

Enlarging the Front End takes away die size that could be used to install more SPs, that is the reason some of us saying performance and performance per watt scaling will be lower at the same die size(except Tessellation).

The foremost reason they changed it (My understanding) was because of the FSA.

 

[SHAQ]

"AMD will supposedly introduce the new card on Thursday December 22, though you won’t see it stocked in stores until early January."

Sounds like next year to me.

 

[lavaheadache]

Lol, I gotta remember that one.

logic works with some women as long as you can apply it to their line of thinking =)

 

[Arkadrel]

Never really liked the VLIW idea,.. think its a good thing their going with a approch more akin to nvidias.

If it helps them get better GPGPU/OpenCL and tessellation at the cost of slightly bigger chips, slightly more power use, im fine with that.

...based on AMD’s own internal research at the time of the Cayman launch the average shader program was utilizing only 3.4 out of 5 Radeon cores....

From:
http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/2

VLIW wastes 1.6 out of 5 on avg. (~32% waste)

Those games Radeon cards do bad in? probably games where VLIW is very wastefull.
Graphic core next, will fix that. Performance will go up drastically in those cases.

GCN = shaders that dont waste anything (more consistant gameing / better min fps?) = great.

I dont care if you can squeeze in more when you do VLIW, if it just ends up wasteing ~33% of it.

 

[Ajay]

"AMD will supposedly introduce the new card on Thursday December 22, though you won’t see it stocked in stores until early January."

Sounds like next year to me.

On the plus side, then a review is just around the corner

 

[IlllI]

what happened to that 'leaked info' that they would be using xdr memory or something in their top cards?

 

[Arkadrel]

@IlllI

By now, it seems that rumor is killed.

 

[fourdegrees11]

Personally I would never spend more then $300 on a vid card, so I'm more interested on how the 7870 compares to the 6970. Die shrink OC headroom plus possible architecture adjustments (GCN vs VLIW)?

 

[RussianSensation]

yeah problem is 2 x Unlocked 6950's wont do what some people need..

My point was that GTX590 is no faster than 2 unlocked HD6950s. So why would he recommend GTX590 when you can get similar performance for less $ with 6950s? It seemed strange that he would recommend a $700 graphics card based on his logic that it's only "$150 extra" vs. supposed pricing of HD7970. Well, why don't we take it further and add $150 on top of $700 and for $850 get 2x HD7950s then? You see my point now?

 

[3DVagabond]

what happened to that 'leaked info' that they would be using xdr memory or something in their top cards?

That was dismissed months ago, but some people who refuse to listen kept it alive. The GCN for only the 7900 was also reported as false long ago by AMD themselves. People who like certain sources though only go by what that one source says.

 

[Zargon]

My point was that GTX590 is no faster than 2 unlocked HD6950s. So why would he recommend GTX590 when you can get similar performance for less $ with 6950s? It seemed strange that he would recommend a $700 graphics card based on his logic that it's only "$150 extra" vs. supposed pricing of HD7970. Well, why don't we take it further and add $150 on top of $700 and for $850 get 2x HD7950s then? You see my point now?

yeah the point is I *need* 2x7950's

and I dont want to drop 900 on GPU's this year when I spent 600+ last year (on one machine, let alone the dual 5870's and the 6970 I bought.....)

I hope the rumors are off by about 75-100 bucks but its not likely I'm guessing

 

[Silverforce11]

yeah problem is 2 x Unlocked 6950's wont do what some people need.

will barely run bf3 in tri-1080p with stuff turned up, let alone those lucky SOBs with 30" dells.....

Even CF 6950 spank gtx590 for 2/3 the cost. Add some AA and the gtx590 becomes a slideshow at 3x1080p. So, no, CF 7950s onwards will steam roll.

Whoever gets a gtx590 over next gen AND pay more for it is a moron.

 

[Silverforce11]

Never really liked the VLIW idea,.. think its a good thing their going with a approch more akin to nvidias.

If it helps them get better GPGPU/OpenCL and tessellation at the cost of slightly bigger chips, slightly more power use, im fine with that.

From:
http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/2

VLIW wastes 1.6 out of 5 on avg. (~32% waste)

Those games Radeon cards do bad in? probably games where VLIW is very wastefull.
Graphic core next, will fix that. Performance will go up drastically in those cases.

GCN = shaders that dont waste anything (more consistant gameing / better min fps?) = great.

I dont care if you can squeeze in more when you do VLIW, if it just ends up wasteing ~33% of it.

Which is why in the other post, i suggested that GCN 2048 1D shaders in reality should be compared with VLIW ~1200 effective shaders in Cayman since they aren't fully used. It's not just a tiny increase when you look at it this way. But the problem will be how optimized their drivers are for GCN.

 

[palladium]

$550 is too much I think, as I doubt it will outperform the 580 GTX.

I agree $550 is too much - approaching the GTX580 launch price. However, unless the arch is screwed up (cough cough Bulldozer) it should outperform the GTX 580. I'm also hoping the NZ dollar goes stronger after launch to keep the already ridiculous prices of computer gear in NZ in check.

 

[IlllI]

That was dismissed months ago, but some people who refuse to listen kept it alive. The GCN for only the 7900 was also reported as false long ago by AMD themselves. People who like certain sources though only go by what that one source says.

its hard to know which 'legit' sources to trust

but i'm hoping the gcn one is real.

 

[ArchAngel777]

I agree $550 is too much - approaching the GTX580 launch price. However, unless the arch is screwed up (cough cough Bulldozer) it should outperform the GTX 580. I'm also hoping the NZ dollar goes stronger after launch to keep the already ridiculous prices of computer gear in NZ in check.

Yeah, lets hope the Bulldozer team didn't design these new Graphics Cores, else we end up with something even worse than VLIW... Yikes. Also sad to see ROPs are still at 32? Dissapointing as I had thought it was ROP limited, or perhaps I am wrong on that count.

 

[Olikan]

Personally I would never spend more then $300 on a vid card, so I'm more interested on how the 7870 compares to the 6970. Die shrink OC headroom plus possible architecture adjustments (GCN vs VLIW)?

actually it's.... MIMD vs VLIW
but i'm curious too

http://en.wikipedia.org/wiki/MIMD
vs
http://en.wikipedia.org/wiki/Very_long_instruction_word

 

[Arkadrel]

Also sad to see ROPs are still at 32? Dissapointing as I had thought it was ROP limited, or perhaps I am wrong on that count.

^ this goes hand in hand with this below:

its hard to know which 'legit' sources to trust

because:

Tomshardware and linking donanimhaber, as a source and saying:

• 4.50 billion transistors, die-area of ~380 mm², built on TSMC 28 nm process
• Advanced GCN 1D architecture
• 2048 1D processing cores
• 128 TMUs, 48 ROPs
• 384-bit wide GDDR5 memory interface, clock slightly below 1 GHz, target bandwidth of 240~264 GB/s
• 6pin + 8pin power connector required
• PCI Express Gen 3.0
• DirectX 11.1 support

However theyre probably wrong and its still 32 ROPs:

reason? this pic posted by Silver:

In this pic, atleast you see 32 ROPs and 128 TMUs.

 

[ArchAngel777]

^ this goes hand in hand with this below:

because:

Tomshardware and linking donanimhaber, as a source and saying:

• 4.50 billion transistors, die-area of ~380 mm², built on TSMC 28 nm process
• Advanced GCN 1D architecture
• 2048 1D processing cores
• 128 TMUs, 48 ROPs
• 384-bit wide GDDR5 memory interface, clock slightly below 1 GHz, target bandwidth of 240~264 GB/s
• 6pin + 8pin power connector required
• PCI Express Gen 3.0
• DirectX 11.1 support

However theyre probably wrong and its still 32 ROPs:

reason? this pic posted by Silver:

In this pic, atleast you see 32 ROPs and 128 TMUs.

Unless AMD design team knows some trick nVidia doesn't, I have to believe these new cores are crap compared to a fermi core. My guess is Fermi can complete 2 times the work per core than one of the GC. I hope I wrong, though. This would probably make sense since Fermi 512 @ 1600Mhz = 1024 @ 800Mhz.. Just throwing numbers out there. Meaning, I think a Fermi core might be capable of performing nearly twice the work. I hope I am wrong... I probably am wrong. All these leaks! I can't wait to see the release....

 

[Vesku]

Could the ROPs have more per clock output than previous gen cards? Seems odd that they would have a big jump in memory bandwidth but not increase the ROPs. Although it could be yet another compromise, bandwidth is a big GPGPU thing but I don't think ROPs are.

 

[bunnyfubbles]

^ this goes hand in hand with this below:

because:

Tomshardware and linking donanimhaber, as a source and saying:

• 4.50 billion transistors, die-area of ~380 mm², built on TSMC 28 nm process
• Advanced GCN 1D architecture
• 2048 1D processing cores
• 128 TMUs, 48 ROPs
• 384-bit wide GDDR5 memory interface, clock slightly below 1 GHz, target bandwidth of 240~264 GB/s
• 6pin + 8pin power connector required
• PCI Express Gen 3.0
• DirectX 11.1 support

However theyre probably wrong and its still 32 ROPs:

reason? this pic posted by Silver:

In this pic, atleast you see 32 ROPs and 128 TMUs.

any explanation as to how they fit 4 x 8 (32) ROPs with a 6 channel memory bus?