Fusion vs. Nvidia High End Cards

[dzoner]

Can Nvidia survive in any segment of the consumer AIB market, even the high end?

A year from now AMD will be prepping it's bulldozer/6000 based fusion chips. The top chips can be expected to have ~5770 class graphics and the ability to seamlessly integrate with AMD's line of discrete boards, with finely gated and accurate power gating within the APU and between and with the APU and and discrete graphics boards. This will provide an ideal computing enviroment in which a computer sips a few watts of power as one surf the web and ramps up in a linear manner to meet power demand and, as 4 and 6 core bulldozer variants are incorporated into fusion chips, have enough headroom to handle high end graphics cards.This will make purchasing a top end fusion chip/system an extremely compelling choice for gamers.

One question is how AMD Fusion chips will interface with Nvidia discrete graphics cards ... to what degree will AMD have an advantage in that seamless linearity. It's certain Nvidia will be at some disadvantage.

Nvidia will enjoy no AIB price/performance advantage over AMD for the forseeable future, even being at a disadvantage to AMD for the next 12 months and perhaps beyond.

For gaming oriented users, the main market for those high end boards, and assuming a four or six core Bulldozer APU is up to the high end AIB/gaming task, going all AMD in their next computer is going to be a very compelling choice over an AMD/Nvidia, Intel/Nvidia. or even Intel/AMD combination.

For Nvidia with drying up revenue streams and shrinking cash reserves it will become a question of return on investment in an enviroment when R&D dollars become ever more scarce and prioritizing resouces critical. When does making consumer AIB's, even high end ones, into the AMD/Intel headwind become financially untenable and designing cards strictly for the profressional and HPC markets without provision for the consumer market necessary.

 

[Daedalus685]

They will be fine provided they find high end buyers for HPC cards.

Otherwise there will be no proverbial cash cow to warrant high R&D costs... but that is just my opinion.. It may be perfectly possible for a company to survive on high end (roughly) cards only. They do have tegra as well.

 

[Rifter]

I dont think fusion is going to be that powerful, just think for a minute the size of heatsink you are going to need to cool a couple of 5970's and a CPU all in the same package, even if they manage to reduce the power 20-30% and retain the same performance thats still going to be over 400 watts to cool, good luck pulling that off on air

 

[Wreckage]

I doubt Fusion will even compete with the slowest NVIDIA card. People are expecting way too much out of it. It's just a replacement for low end computers that have slow CPUs and crap integrated GPUs. So now they will have that slow CPU and crap GPU on one chip.

If anything Fusion is going to get competition from Tegra which is already showing up in Tablets and could easily be put into netbooks, laptops and low end desktops. It's much better suited for smaller devices like that. Not to mention that Sandy Bridge will probably out sell fusion 4 to 1.

 

[dzoner]

I dont think fusion is going to be that powerful, just think for a minute the size of heatsink you are going to need to cool a couple of 5970's and a CPU all in the same package, even if they manage to reduce the power 20-30% and retain the same performance thats still going to be over 400 watts to cool, good luck pulling that off on air

The 5970's referred to discrete cards.

2nd generation bulldozer/6000 based Fusion chips on a 32 nm node can be expected to have 5770+ graphics.

 

[Daedalus685]

I don't think dzoner's point was to expect high end performance from fusion... but that it (and the new intel CPUs) represents a cut off of the low end GPU industry.

For all intents and purposes Nvidia will no longer have any market for a dedicated card that performance slower than the integrated.. which lately has been a huge market for OEM GPUs. The question is how small can the GPU industry become before it is not profitable to compete in it if the R&D dollars don't translate to another market.

Wreckage.. the point is not if fusion will out sell sandy bridge (obviously Intel will sell droves more).. but that both will now institute a hard cut off in the GPU market.

 

[Lonyo]

I doubt Fusion will even compete with the slowest NVIDIA card. People are expecting way too much out of it. It's just a replacement for low end computers that have slow CPUs and crap integrated GPUs. So now they will have that slow CPU and crap GPU on one chip.

If anything Fusion is going to get competition from Tegra which is already showing up in Tablets and could easily be put into netbooks, laptops and low end desktops. It's much better suited for smaller devices like that. Not to mention that Sandy Bridge will probably out sell fusion 4 to 1.

Well at the moment based on AT results, Sandy Bridge graphics are faster than the HD5450 from AMD, and the HD5450 is faster than the GF210 from NV.
Assuming it has the rumoured 320~400SPs that I read somewhere, that would put its potential power on par with the HD5550/5650 from a GPU core POV, but obviously with less memory bandwidth. That would make it maybe GT220 level or above, depending on how the memory bandwidth plays out.

Obviously that's comparing it to the current low end cards, which might end up being replaced by the time it's out, but certainly the lowest of the low currently should be beaten, and they already look like they will be beaten by Sandy Bridge.

 

[Red Storm]

I doubt Fusion will even compete with the slowest NVIDIA card. People are expecting way too much out of it. It's just a replacement for low end computers that have slow CPUs and crap integrated GPUs. So now they will have that slow CPU and crap GPU on one chip.

Right now, I do agree with this. However, eventually these CPU/GPU chips will match mid level cards in performance, and that is where both companies make their money in regards to video cards. High end will always be a dedicated card.

I personally don't see the majority of people forgoing Windows and/or OSX to go with a tablet/netbook running a mobile-geared OS for their main laptops and desktops, so I do think nVidia will be in trouble down the road.

 

[Janooo]

NV could end up with Tegra as the only major product in the future.

 

[dzoner]

I doubt Fusion will even compete with the slowest NVIDIA card. People are expecting way too much out of it. It's just a replacement for low end computers that have slow CPUs and crap integrated GPUs. So now they will have that slow CPU and crap GPU on one chip.

If anything Fusion is going to get competition from Tegra which is already showing up in Tablets and could easily be put into netbooks, laptops and low end desktops. It's much better suited for smaller devices like that. Not to mention that Sandy Bridge will probably out sell fusion 4 to 1.

While the top end 2nd generation desktop Fusion chips will almost certainly have 5770 or better class graphics, the salient point is not Fusion graphics vs. AIB graphics, it's high end Fusion chips having a cpu element able to keep up with whatever high end graphics boards it is paired with, making Fusion chips competitive across all desktop markets.

 

[Wreckage]

Well at the moment based on AT results, Sandy Bridge graphics are faster than the HD5450 from AMD, and the HD5450 is faster than the GF210 from NV.
Assuming it has the rumoured 320~400SPs that I read somewhere, that would put its potential power on par with the HD5550/5650 from a GPU core POV, but obviously with less memory bandwidth. That would make it maybe GT220 level or above, depending on how the memory bandwidth plays out.

Obviously that's comparing it to the current low end cards, which might end up being replaced by the time it's out, but certainly the lowest of the low currently should be beaten, and they already look like they will be beaten by Sandy Bridge.

That would put it slower than a GT240 which can be had for $40 AR. I'm betting a GT410 (or whatever the next low end NVIDIA chip is), will at least have GT240 performance. So that will mean zero competition from Fusion.

Sandy Bridge is an improvement over current integrated graphics. But one would expect newer chips to be better than older ones. It's still not competing directly with discrete graphics.

 

[Rifter]

That would put it slower than a GT240 which can be had for $40 AR. I'm betting a GT410 (or whatever the next low end NVIDIA chip is), will at least have GT240 performance. So that will mean zero competition from Fusion.

Sandy Bridge is an improvement over current integrated graphics. But one would expect newer chips to be better than older ones. It's still not competing directly with discrete graphics.

I agree with this, Fusion might cut out extreme budget cards(under $40) but the mid and high end will stay discreet.

 

[Genx87]

Sounds like wishful thinking. Honestly. 5770 performance a year from now? That is a 40 dollar card then. High end discrete graphics wont be the territory of Fusion. If it ever comes to that, consoles will have won the war.

 

[ModestGamer]

I don't think dzoner's point was to expect high end performance from fusion... but that it (and the new intel CPUs) represents a cut off of the low end GPU industry.

For all intents and purposes Nvidia will no longer have any market for a dedicated card that performance slower than the integrated.. which lately has been a huge market for OEM GPUs. The question is how small can the GPU industry become before it is not profitable to compete in it if the R&D dollars don't translate to another market.

Wreckage.. the point is not if fusion will out sell sandy bridge (obviously Intel will sell droves more).. but that both will now institute a hard cut off in the GPU market.

Actually on the OEM "walmart computers" side I'd bet that Fusion will outsell Sandybridge. but we shall see.

 

[BenSkywalker]

The top chips can be expected to have at least 5770 class graphics

I'm curious if you could explain where the bandwidth for this is going to come from?

AMD could slap a 5870 on die with their CPU right now and performance would still suck, they could strap 4 GTX 480s on their too with the same result(I'm not knocking ATi chips, that's why I threw the 480s in the comparison too). There is no magical bandwidth fairy that is going to make anything with the massive shortcomings of CPU bandwidth perform decently in the graphics segment compared to discrete. It has never been an issue of getting it all on one die, both AMD and Intel have been capable of doing that for years- what doesn't work is getting anything resembling decent performance when both parts are horribly bandwidth limited.

 

[Daedalus685]

I agree with this, Fusion might cut out extreme budget cards(under $40) but the mid and high end will stay discreet.

But those budget cards are what OEMS like dell buy millions of.. It is not a small profit section of the market.

The question is not whether high end will always be discrete.. but whether one can make enough money to develop high end cards without the huge volume market segments.

 

[dzoner]

Sounds like wishful thinking. Honestly. 5770 performance a year from now? That is a 40 dollar card then. High end discrete graphics wont be the territory of Fusion. If it ever comes to that, consoles will have won the war.

The point is not Fusion competing with high end graphics cards, it is Fusion providing excellent mainstream computing on it's own WHILE being able to seamlessly integrate with higher end AMD graphics cards when needed.

It's having an extremely economical, yet highly capable low power computer that can instantly morph (with a high end graphics card) into a screaming gaming monster.

 

[Genx87]

The point is not Fusion competing with high end graphics cards, it is Fusion providing excellent mainstream computing on it's own WHILE being able to seamlessly integrate with higher end graphics cards when needed.

It having an extremely economical, yet highly capable low power computer that can instantly morph (with a high end graphics card) into a screaming gaming monster.

I dont see the problem. Honestly, are you suggesting AMD and Intel will cripple PCIE or whatever succeeds it?

So Fusion wont directly compete with Nvidias high end card. Where is the problem? Fusion will honestly be low end. Does Intel integrated crap eat away at 100+ market? I doubt it. You are talking about 5770 performance a year from now. And I seriously doubt they will even achieve that.

 

[dzoner]

Wreckage.. the point is not if fusion will out sell sandy bridge (obviously Intel will sell droves more).. but that both will now institute a hard cut off in the GPU market.

Considering the settlement terms and AMD's APU graphics advantage, I wouldn't bet the farm on that continuing into the future.

 

[dzoner]

So Fusion wont directly compete with Nvidias high end card. Where is the problem? Fusion will honestly be low end. Does Intel integrated crap eat away at 100+ market? I doubt it. You are talking about 5770 performance a year from now. And I seriously doubt they will even achieve that.

AMD high end cards DO directly compete with Nvidia high end cards. If you had a Fusion motherboard and chip and wanted to add a high end graphics board, if buying an AMD board provided substantial advantages over a comparable Nvidia board with that Fusion chip, which would you buy?

 

[Daedalus685]

Considering the settlement terms and AMD's APU graphics advantage, I wouldn't bet the farm on that continuing into the future.

Ok, I can't be sure that Intel will sell more.. but I'm pretty damn sure and If I had a farm I'd put my farm up against your farm that Intel will sell more sandy bridge than AMD does fusion products...

 

[Genx87]

AMD high end cards DO directly compete with Nvidia high end cards. If you had a Fusion motherboard and chip and wanted to add a graphics board, if buying an AMD board provided substantial advantages over a comparable Nvidia board, which would you buy?

Considering AMD hasnt been the choice of gamers for 4 years. What makes you magically believe fusion is going to change their mind? Intel has been smacking AMDs ass since Core 2 duo and they arent looking back.

We have no idea if buying an AMD board would provide an advantage for high end discrete graphics for AMD discrete cards? That is all speculation at this point. AMD would be dumb as shit to cripple their platform with the very crowd who buys high end cards. THink right now if AMD forced people to buy ATI cards to achieve the highest possible performance on their CPU platform? That pathetic 2:1 ratio they are losing to Intel on Steam would be 4:1 within 18 months. Intel CPU + ATI or Nvidia GPU is the gamers choice. I dont see that changing regardless of fusion.

 

[dzoner]

Ok, I can't be sure that Intel will sell more.. but I'm pretty damn sure and If I had a farm I'd put my farm up against your farm that Intel will sell more sandy bridge than AMD does fusion products...

Lol ... I'll grant you they sell more, I'm not betting my farm on that, i was pointing out the 'obviously Intel will sell *droves* more' might not apply in the future.

 

[BenSkywalker]

Considering the settlement terms and AMD's APU graphics advantage, I wouldn't bet the farm on that continuing into the future.

It would be a very, very safe bet. AMD is not capable of producing enough chips to compete with Intel. They could sell every single chip they made for $10K each and it would take them ~ a decade before they would have the ability to produce enough chips to outsell Intel(production time for two new fabs back to back). I'm not saying that AMD can't make a clearly superior product to Intel, we all saw it for quite some time in the Athlon days, but I am saying that they don't have the production capacity to outsell Intel.

 

[Nemesis 1]

The 5970's referred to discrete cards.

2nd generation bulldozer/6000 based Fusion chips on a 32 nm node can be expected to have 5770+ graphics.

Based off of the memory limits I will bet that its doesn't . Your talking fairy dust . But both intel and amd will destroy the lowend market. The second generation fusion will be great. Looking at SB results intels next gen looks promising also , Thats 1 year from sandies release.

Bobcat looks very interesting however.