6870 VS GTX460 Die sizes and price/performance

[psoomah]

http://www.computerbase.de/bildstrecke/31272/31/
http://www.nordichardware.com/news/7...-revealed.html

hd 6870 - 255mm sq.
gtx 460 - 366mm sq.

That's a massive differential for ~ the same performance.

and for the heck of it:

gts 450 - ~240 mm sq. ... http://www.guruht.com/2010/08/geforc...e-picture.html

A 1GB 460 is selling for around $220, a 1GB GTS450 for around $130.

Barts 6870 chips are pin compatible with Cypress 5850 chips, both chips are rated at 151w - http://www.computerbase.de/bildstrecke/31272/31/ - so a manufacturer could, quite literally and simply, drop a 6870 chip into a 5850 board ... or a 6850 chip into a 5830 board ... and sell it, which is apparently just what a few are doing for their 'entry level' 6850 and 6870 cards. Zero R&D costs, zero engineering and layout costs ... nothing but net ... profits.

With the 6870, AMD is producing a card die size competitive with a gts 450 and performance competitive with a gtx 470.

Nvidia is hopelessly outgunned.

 

[Tsavo]

The pin compatibility is pretty slick, which reminds me of when Micron tried to institute a socket standard for GPUs so they could be swapped out like a CPU. It was next to impossible then and still is today, but I'd love to be able to change just the GPU.

 

[happy medium]

DO you know the die size of the 5850 and 5870?
It would be fun to see if they can get = performance from a smaller die on the same 40nm process at the same wattage then the 58xx series.

WE also have to remember the gtx460 is not running at 900 core and is not full enabled yet.
A fully enabled gtx460 (gtx475) at 850/900 core, I'm guessing would beat the pants off a 6870.

If they could do it why couldn't Nvidia do it?

 

[psoomah]

DO you know the die size of the 5850 and 5870?
It would be fun to see if they an get = performance from a smaller die on the same 40nm process at the same wattage then the 58xx series.

WE also have to remember the gtx460 is not running at 900 core and is not full enabled yet.
A fully enabled gtx460 (gtx475) at 850/900 core, I'm guessing would beat the pants off a 6870.

If they could do it why couldn't Nvidia do it?

5850 - 334mm vs 6870 - 255mm.

http://www.computerbase.de/bildstrecke/31272/31/

AMD claims 68xx has a 35% performance per mm increase from 58xx.

http://www.computerbase.de/bildstrecke/31272/31/

Benchmarks - 460 vs 6850 & 6870

http://www.computerbase.de/bildstrecke/31272/23/
http://www.computerbase.de/bildstrecke/31272/25/

Overclocking the 460 = overclocking the 6850/70.

I'm guessing you'd be guessing wrong.

Nvidia can't do it because the GF104 is derived from the GF100 = very inefficient vs Cypress. Nvidia won't be even close to competitive until they come out with a new architecture (if then), and that isn't happening until 2H 2011 on the 28nm node ... at the earliest.

 

[happy medium]

http://www.computerbase.de/bildstrecke/31272/31/

AMD claims 35% performance per mm increase from 58xx.

http://www.computerbase.de/bildstrecke/31272/31/

Nice thanks,

SO if AMD can get equal performance ,using less transistors, a smaller die, with equal power consumption on the same 40nm process, why couldn't Nvidia do the same with there upcomming gf110 chip?

People seem to think it can't be done on the 40nm process. Looks like AMD did it.

 

[edplayer]

SO if AMD can get equal performance ,using less transistors, a smaller die, with equal power consumption on the same 40nm process, why couldn't Nvidia do the same with there upcomming gf110 chip?

People seem to think it can't be done on the 40nm process. Looks like AMD did it.

I think they can. Actually, not the same but similar. AMD already had more performance/die size with the 5XX0 series and that has now improved. Nvidia should be able to improve their performance/die size substantially but compared to their own previous design.

Not sure if they can match or beat AMD though.

 

[psoomah]

I think they can. Actually, not the same but similar. AMD already had more performance/die size with the 5XX0 series and that has now improved. Nvidia should be able to improve their performance/die size substantially but compared to their own previous design.

Not sure if they can match or beat AMD though.

GF104~GTX 460 represented Nvidia's best shot at optimizing the GF100 architecture to compete against AMD in the $200-300 price range.

It is unlikely they are going to do further major engineering on that architecture when at best they might get kinda close to Barts, several months from now, and just concentrate on Kepler and 28nm, all the more so knowing AMD is hard at work on their 7xxx cards on 28nm.

The 460 IS price/performance competitive with the 5830/50 cards, although at what HAS to be razor thin profit margins.

But that 68xx 35% performance per mm increase over the 58xx cpu is a KILLER. The 460 just has no chance, particularly with that pin-compatibility feature allowing the AIBs to just switch out GPU chips and selling 'entry level' 68xx cards at a low low prices.

 

[Lonyo]

Nice thanks,

SO if AMD can get equal performance ,using less transistors, a smaller die, with equal power consumption on the same 40nm process, why couldn't Nvidia do the same with there upcomming gf110 chip?

People seem to think it can't be done on the 40nm process. Looks like AMD did it.

It's not that it can't be done, it's just that NV haven't managed to really do it with their existing tweak (GF104 vs GF100), and they would have a lot of catching up to do, since they are currently around 25% slower with a similar die size (GTX460 1GB vs HD5870) or use equal power for worse performance (HD5850 vs GTX460 1GB).

Given that AMD have probably just improved by another 20% in performance/mm^2, and possibly similarly in performance/watt, that puts them even further ahead. Mainly focusing on the mid-range)

Also AMD is speculated to have increased the die size on the top end product as another way of increasing performance. NV won't really want to do that, so their only gains can come from increased efficiency (unless they really want an even bigger die than the 530mm^2 GF100...), so AMD have got a 20% efficiency improvement and maybe a 20% improvement just by adding more stuff. That's 40% potentially improvement at the top end. NV don't need to get that full increase because they are already ahead, but they need to get it ideally without increasing die size because they don't have much headroom.

NV can improve, but it's doubtful they can improve as much/enough to come close to matching AMD, especially when they have a chip which isn't streamlined primarily for gaming.

 

[Arkaign]

The dark side of this is that I very much doubt that AMD is going to give us a break on pricing. I think that's exactly why these things are being named as the replacements to the 5850/5870. I would love to see a price war, but they're probably just going to gouge for these things. It will be particularly funny to see if anyone not paying attention buys a 6870 to replace a 5870

 

[SlowSpyder]

Nvidia is hopelessly outgunned.

Now you've done it...

This discussion has been played out a few times already. Between the 4870 and GTX2xx parts. Between the GTX480 and 5870. It looks like it will be another round of the same.

Some like to claim that AMD is going to run Nvidia out with a price war because they get better performance per mm. Some like to claim that Nvidia is in the drivers seat because they use that extra silicon for CUDA and HPC functions. There really isn't a right or wrong, just different. Though I would tend to agree, AMD can probably more easily make money in the video card world. Nvidia is raking in money in the HPC and professional area. So, meh.

 

[psoomah]

There really isn't a right or wrong, just different. Though I would tend to agree, AMD can probably more easily make money in the video card world. Nvidia is raking in money in the HPS and professional area. So, meh.

AMD is now selling more AIB's than Nvidia, by this time next year, it will very possibly have traded places with the Nvidia (70/30) of two years ago.

That is really right for AMD and really wrong for Nvidia.

There's currently an HIS 5830 on Newegg for $169. No reason they couldn't pop a 6950 gpu in that card and sell it at the same price once gpu availability is there.

Not that they WILL, but HIS has been really agresssive on their pricing, consistently hitting that 'lowest price' slot on newegg.

Mmm, i see MSI is selling a 5850 for $254/$194 w/rebate.

Geez.

 

[tviceman]

hd 6870 - 255mm sq.
gtx 460 - 366mm sq.

Not that it really matters, but I do believe the gtx460 is 331mm^2.

 

[Arkaign]

Now you've done it...

This discussion has been played out a few times already. Between the 4870 and GTX2xx parts. Between the GTX480 and 5870. It looks like it will be another round of the same.

Some like to claim that AMD is going to run Nvidia out with a price war because they get better performance per mm. Some like to claim that Nvidia is in the drivers seat because they use that extra silicon for CUDA and HPC functions. There really isn't a right or wrong, just different. Though I would tend to agree, AMD can probably more easily make money in the video card world. Nvidia is raking in money in the HPC and professional area. So, meh.

Hah, so damned true. Things have swapped places so many times where someone was outright dominant that it's just funny. Nvidia could still botch things and go away, but I doubt it.

Ti4200 >>>> 8500
9500/9700/9800 >>> Ti/GF-FX
8800GTX >>> HD1x00/2x00

There is usually something resembling parity, as during :

GF3 vs. 8500
GF6 vs. X8XX
GF7 vs. X1XXX
GTX2XX vs. HD4XXX

But yeah, swings happen.

 

[SlowSpyder]

AMD is now selling more AIB's than Nvidia, by this time next year, it will very possibly have traded places with the Nvidia (70/30) of two years ago.

That is really right for AMD and really wrong for Nvidia.

There's currently an HIS 5830 on Newegg for $169. No reason they couldn't pop a 6950 gpu in that card and sell it at the same price once gpu availability is there.

Not that they WILL, but HIS has been really agresssive on their pricing, consistently hitting that 'lowest price' slot on newegg.

Mmm, i see MSI is selling a 5850 for $254/$194 w/rebate.

Geez.

I agree that Nvidia will probably struggle somewhat in the gaming world, especially with APU's on the horizon from Intel and AMD. But, Nvidia appears to make a LOT of money from the professional space and will likely continue to do so for some time. Losing revenue in the gaming world will not help them obviously, but I'm not sure that GeForce is their big money maker anymore anyway.

 

[happy medium]

It's not that it can't be done, it's just that NV haven't managed to really do it with their existing tweak (GF104 vs GF100), and they would have a lot of catching up to do, since they are currently around 25% slower with a similar die size (GTX460 1GB vs HD5870) or use equal power for worse performance (HD5850 vs GTX460 1GB).

Given that AMD have probably just improved by another 20% in performance/mm^2, and possibly similarly in performance/watt, that puts them even further ahead. Mainly focusing on the mid-range)

Also AMD is speculated to have increased the die size on the top end product as another way of increasing performance. NV won't really want to do that, so their only gains can come from increased efficiency (unless they really want an even bigger die than the 530mm^2 GF100...), so AMD have got a 20% efficiency improvement and maybe a 20% improvement just by adding more stuff. That's 40% potentially improvement at the top end. NV don't need to get that full increase because they are already ahead, but they need to get it ideally without increasing die size because they don't have much headroom.

NV can improve, but it's doubtful they can improve as much/enough to come close to matching AMD, especially when they have a chip which isn't streamlined primarily for gaming.

So you also disagree with the "na sayers" stating that Nvidia's will bring out a super hot , super large, 512sp gtx580 with higher clocks to compete with AMD's 6000 series?

I see Nvidia releasing a 384 sp gtx 475 at 850 core to overtake the 6870 for about the same price and leave the gtx460 1gb against the 6850 at a lower price.

Next, when AMD releases there higher end in late Nov or Dec, we should start seeing what the gf110 has to offer.

 

[GaiaHunter]

Not that it really matters, but I do believe the gtx460 is 331mm^2.

The only source I've seen for 331mm^2 was BSN.

I've seen a few measurements photos in places like xtremesystems and b3d that put it at 366mm^2. Wiki puts it at 368mm^2.

No NVIDIA official number though.

 

[OCGuy]

Time to start consolidating these threads now that it is launch week?

 

[psoomah]

Not that it really matters, but I do believe the gtx460 is 331mm^2.

My URL showed the die with the heatsink machined off. That's pretty definitive. The heat sink of the GF 104 is far larger than Cypress'

What's your source?

 

[bunnyfubbles]

Nice thanks,

SO if AMD can get equal performance ,using less transistors, a smaller die, with equal power consumption on the same 40nm process, why couldn't Nvidia do the same with there upcomming gf110 chip?

People seem to think it can't be done on the 40nm process. Looks like AMD did it.

You're failing to consider time constraints. Fermi was months late and is a lot less efficient (for games) so that means nVidia has more to do with less time compared to what AMD is doing with Evergreen -> NorthernIslands, which is why it doesn't seem feasible and why we don't have faith even in a GF110 part. Fermi's problems and delays makes it less feasible to keep trying to invest in a losing round, which is why nVidia is full steam ahead on Kepler and Maxwell.

 

[psoomah]

Time to start consolidating these threads now that it is launch week?

Why? In three days it'll be all about actual retail prices and tech site benchmarks and all these threads will just fade away.

 

[Attic]

Good info. Glad to see serious performance improvement at the $200 mark over the past year.

 

[RussianSensation]

Nvidia is hopelessly outgunned.

Once again, NV designs a GPU for 3 tasks: (1) games (2) Quadro line (3) GPGPU computing.

ATI designs GPU for 1 task: games.

Unless this changes, NV will always have a larger GPU relative to AMD for the same performance. This was true during G80, G200/b and GF100 times. Nothing has changed.

Why is it always news when a chip designed specifically for graphics is more lean and more efficient than a chip designed for multiple tasks? NV amortizes its R&D/costs across 3 product lines while AMD amortizes it against 1 product line.

Furthermore, last quarter AMD only had 1 million operating profit despite the smaller die sizes for HD5000 lines. Therefore, it's pretty obvious that other factors are involved such as marketing costs, distribution costs, etc.

From an efficiency perspective, AMD has led for as long as I can remember.

 

[Gloomy]

Why is it always news when a chip designed specifically for graphics is more lean and more efficient than a chip designed for multiple tasks?

Because the people that are surprised paid for a chip specifically for graphics? :awe:

NVIDIA would do well to have a more varied lineup. Joe Schmo does not care nor want to pay for non-graphics silicon. In his mind, he's paying for a graphics card. If this is what's holding NVIDIA's gaming performance back, then well... they should stop, and make two products instead of one.

 

[psoomah]

So you also disagree with the "na sayers" stating that Nvidia's will bring out a super hot , super large, 512sp gtx580 with higher clocks to compete with AMD's 6000 series?

I see Nvidia releasing a 384 sp gtx 475 at 850 core to overtake the 6870 for about the same price and leave the gtx460 1gb against the 6850 at a lower price.

Next, when AMD releases there higher end in late Nov or Dec, we should start seeing what the gf110 has to offer.

There are already pics of factory oc'ed 6870 boards on the web, a few of which are probably in reviewers hands already. After all Asus could just pop a 6870 in a ROG 5870 board and get an immediate 20>30% kick on the reference 6870. Same goes for any of the premium o/c 5870 boards form MSI, Sapphire, Gigabyte etc.

I doubt the 460 has the headroom to match that.

 

[happy medium]

You're failing to consider time constraints. Fermi was months late and is a lot less efficient (for games) so that means nVidia has more to do with less time compared to what AMD is doing with Evergreen -> NorthernIslands, which is why it doesn't seem feasible and why we don't have faith even in a GF110 part. Fermi's problems and delays makes it less feasible to keep trying to invest in a losing round, which is why nVidia is full steam ahead on Kepler and Maxwell.

The gf100 was late. The gf104 was right on time. So why would the gf110 be late? I'm sure it was planned well before the gf100 launch.

The gf100 being late has little or nothing to do with the gf110 launch date.