[UPDATE 3 - Sapphire's Reference Costs $649 USD] AMD Reveals R9 Nano Benchmarks Ahead of Launch

[adam_S4PH]

I hope its well priced  :wub:

[patrickjp93]

if nvidia got hbm everyone would be applauding them for being an industry leader even if they have yield problems. and if you look at the gpu line up amd has nvidia beat in almost every match except the fury x and gtx 980 ti. the r9 390 beats the 970 r9 380 beats 960 but nooo amd cant compete with nvidia even though they offer better choices

The R9 390/X trade blows with the 970/80. They do not outright beat at all. Further, Nvidia doesn't screw up new techs. Nvidia would not have released a 4GB flagship. It would have been 6, guaranteed. They'd have made a bigger interposer for a wider card and put 6 chips on it. Or they'd have waited for Hynix or Samsung to come out with HBM2 to get rid of the BS limitation.

[Tomsen]

The R9 390/X trade blows with the 970/80. They do not outright beat at all. Further, Nvidia doesn't screw up new techs. Nvidia would not have released a 4GB flagship. It would have been 6, guaranteed. They'd have made a bigger interposer for a wider card and put 6 chips on it. Or they'd have waited for Hynix or Samsung to come out with HBM2 to get rid of the BS limitation.

NVidia have quite the history with terrible memory/fab transition..

// From phone

[Ridska]

If it's from AMD then these should be taken with a grain of salt. As we saw their benchmarks had the Fury X crushing the 980 ti, but in reality the 980 ti beats it

Ture , but the Fury X achives lower temps than the 980 Ti .

[gilang01]

any ideas of the price point ?

Hehehehe NDA m8, can't say a thing till the thing's out

[othertomperson]

if nvidia got hbm everyone would be applauding them for being an industry leader even if they have yield problems. and if you look at the gpu line up amd has nvidia beat in almost every match except the fury x and gtx 980 ti. the r9 390 beats the 970 r9 380 beats 960 but nooo amd cant compete with nvidia even though they offer better choices

 

Nvidia have HBM same as AMD, but they realised that the yields are too piss-poor for it to be commercially viable to use it. If this were a special edition kind of flagship that cost a £1000 and was more of a mascot than a product they needed to sell well then HBM might have been a good gimmick for it.

 

You also have to question why, with yields as abysmal as they clearly are, they thought it was a good idea to split HBM across three tiers of product and limit the availability and max vram of each so drastically -- especially when the only marketing they have done successfully is to convince everyone that you need an obscene amount of vram to do anything.

 

Also you're wrong about AMD beating Nvidia at every price point -- at every price point the two are basically the same. This is deliberate. It would have been easy for AMD to release the 380 as full Tonga and beat the 960, but they chose to rebadge the 285 and match it instead.

 

 

Ture , but the Fury X achives lower temps than the 980 Ti .

 

Only because it's watercooled. Its TDP is still huge.

[Tomsen]

Nvidia have HBM same as AMD, but they realised that the yields are too piss-poor for it to be commercially viable to use it. If this were a special edition kind of flagship that cost a £1000 and was more of a mascot than a product they needed to sell well then HBM might have been a good gimmick for it.

 

You also have to question why, with yields as abysmal as they clearly are, they thought it was a good idea to split HBM across three tiers of product and limit the availability and max vram of each so drastically -- especially when the only marketing they have done successfully is to convince everyone that you need an obscene amount of vram to do anything.

Nvidia have been working on HBM, that is true.

Because you know that is the reason?

No, not really. They would wait, exactly as they are doing now.

 

Maybe AMDs R9 Fury dont have a GDDR5 memory controller? Making them HBM exclusive products.

Maybe someone is stacking them up.. 

[patrickjp93]

NVidia have quite the history with terrible memory/fab transition..

// From phone

Memory transitions are beautiful. AMD's the one with that problem (GDDR4). Node transitions tend to be problematic, but Nvidia worked directly with TSMC on 16nm FF starting 2 years ago. I doubt it's going to go so poorly this time, especially since Pascal is just the Maxwell design with full DP support and a few more bells and whistles for DX 12 like Asynchronous Shaders.

[Dabombinable]

Memory transitions are beautiful. AMD's the one with that problem (GDDR4). Node transitions tend to be problematic, but Nvidia worked directly with TSMC on 16nm FF starting 2 years ago. I doubt it's going to go so poorly this time, especially since Pascal is just the Maxwell design with full DP support and a few more bells and whistles for DX 12 like Asynchronous Shaders.

But its supposed to have HBM/HBM 2 isn't it? Because that's exactly what Nvidia's been teasing for a while with their images.

[Tomsen]

Memory transitions are beautiful. AMD's the one with that problem (GDDR4). Node transitions tend to be problematic, but Nvidia worked directly with TSMC on 16nm FF starting 2 years ago. I doubt it's going to go so poorly this time, especially since Pascal is just the Maxwell design with full DP support and a few more bells and whistles for DX 12 like Asynchronous Shaders.

AMD had fewer issues than Nvidia, historically.

The transition from GTX200 -> GTX 400 was truly beautiful.

I doubt pascal will be the big DP changer. It is more focused on 'mixed' precision (thinking fp16, fp32).

I think Volta will be more of a DP beast  

[patrickjp93]

AMD had fewer issues than Nvidia, historically.

The transition from GTX200 -> GTX 400 was truly beautiful.

I doubt pascal will be the big DP changer. It is more focused on 'mixed' precision (thinking fp16, fp32).

I think Volta will be more of a DP beast

Nvidia can't afford to not have big DP power in 2016 and early 2017. If they do Intel will walk all over them and bury them in HPC with Knight's Landing.

[Tomsen]

Nvidia can't afford to not have big DP power in 2016 and early 2017. If they do Intel will walk all over them and bury them in HPC with Knight's Landing.

Which is why I think Nvidia is rushing volta out for those markets.

[Prysin]

Nvidia can't afford to not have big DP power in 2016 and early 2017. If they do Intel will walk all over them and bury them in HPC with Knight's Landing.

At the top end spectrum... at the other end of that spectrum is AMD sitting with powerful hardware on a "budget" cost... So trying to fight a budget war with AMD wont be fun or smart for Nvidia for sure...

 

Still, i havent read reviews of the maxwell quadros yet, but it wouldnt surprise me if Nvidia does like AMD, and simply locks the DP with firmware/BIOS.... i mean, if you flash a R9 290X with a Firepro W9100 Bios, you get a 4 or 8GB W9100... its several sites that has guides for doing so and its been done already...

[patrickjp93]

At the top end spectrum... at the other end of that spectrum is AMD sitting with powerful hardware on a "budget" cost... So trying to fight a budget war with AMD wont be fun or smart for Nvidia for sure...

 

Still, i havent read reviews of the maxwell quadros yet, but it wouldnt surprise me if Nvidia does like AMD, and simply locks the DP with firmware/BIOS.... i mean, if you flash a R9 290X with a Firepro W9100 Bios, you get a 4 or 8GB W9100... its several sites that has guides for doing so and its been done already...

Nvidia easily wins against AMD in a budget war in HPC. It has volume and a following large enough to sustain it. AMD is the one that can't win that war.

[patrickjp93]

Which is why I think Nvidia is rushing volta out for those markets.

Except it's not. Volta is Q2 2017, and the first units are going to 2 of the Department of Energy Supercomputers.

[Prysin]

Nvidia easily wins against AMD in a budget war in HPC. It has volume and a following large enough to sustain it. AMD is the one that can't win that war.

If Intel catches on, Nvidia will quickly lose volume, once they lose volume (and thus following), they must compete on price vs performance. A position where atm, AMD has the "sweetspot"...

 

Sure, a super cheap tesla would be nice, but hey man. Do you really think they could sustain drastically cutting their margins on those chips? For a while sure, for months, or maybe years (depends how good of a grip intel would get on the market and how rapidly intel could evolve their products)??? I think Nvidia would struggle.

 

Like you said, they got volume and following, now. But a company that has more volume and more following is intel. Honestly though, if Nvidia lost revenue in the professional segment, they would have to cut in the RnD of consumer products and or drop products (like shield perhaps) due to lower revenue... At that point they would be fighting "the underdog" and an overwhelmingly well funded new player at the same time... Nobody would benefit from that war, not the way these companies are right now (size vs market share wise)...

[Tomsen]

Except it's not. Volta is Q2 2017, and the first units are going to 2 of the Department of Energy Supercomputers.

Except it is. They are targeting earlier release for those markets.

[patrickjp93]

If Intel catches on, Nvidia will quickly lose volume, once they lose volume (and thus following), they must compete on price vs performance. A position where atm, AMD has the "sweetspot"...

Sure, a super cheap tesla would be nice, but hey man. Do you really think they could sustain drastically cutting their margins on those chips? For a while sure, for months, or maybe years (depends how good of a grip intel would get on the market and how rapidly intel could evolve their products)??? I think Nvidia would struggle.

Like you said, they got volume and following, now. But a company that has more volume and more following is intel. Honestly though, if Nvidia lost revenue in the professional segment, they would have to cut in the RnD of consumer products and or drop products (like shield perhaps) due to lower revenue... At that point they would be fighting "the underdog" and an overwhelmingly well funded new player at the same time... Nobody would benefit from that war, not the way these companies are right now (size vs market share wise)...

In accelerators, Nvidia is king by far. AMD doesn't have the finances. Nvidia does, especially with long-term focus being the goal. It has the volume in consumer graphics too by far.

[patrickjp93]

Except it is. They are targeting earlier release for those markets.

That's not rushing. That's fulfilling contracts. Rushing would be narrowing the release schedule to 8 months after Pascal instead of 14.

[Prysin]

In accelerators, Nvidia is king by far. AMD doesn't have the finances. Nvidia does, especially with long-term focus being the goal. It has the volume in consumer graphics too by far.

It has Volumes in consumer graphics now...

 

Pascal vs whatever AMD has... if AMD has one thing going for them, it is real life experience with HBM.. It is an edge when it comes to the technology. Who knows, with DX12 on the steps, perhaps we are looking at another shift between team Red and team Green.

 

Who knows. Assuming Nvidia will always be on top because "more money" is simply foolish. Bigger companies has made shittier products with bigger investments then what AMD or Nvidia has.

[Tomsen]

That's not rushing. That's fulfilling contracts. Rushing would be narrowing the release schedule to 8 months after Pascal instead of 14.

Fulfilling contracts and fulfilling a market is two different things.

Nvidia can still fulfill their contracts in regards to volta, without fulfilling the professional market.

I still expect earlier release for the professional market.

[Tomsen]

In regards to R9 nano, I think we will see how its turbo can influence the performance.

Might expect it to go as high as +1000Mhz in certain workloads.

[TheMissxu]

In regards to R9 nano, I think we will see how its turbo can influence the performance.

Might expect it to go as high as +1000Mhz in certain workloads.

It's rumored to be a ~800MHz. It's hard to push it to Fury X numbers without the extra power.

[Tomsen]

It's rumored to be a ~800MHz. It's hard to push it to Fury X numbers without the extra power.

I have also stated previously in this thread +800Mhz is the sweetspot for efficiency.

However, if the power- and thermal-budget allows it, hopefully it will run at higher clockspeeds.

[marldorthegreat]

In accelerators, Nvidia is king by far. AMD doesn't have the finances. Nvidia does, especially with long-term focus being the goal. It has the volume in consumer graphics too by far.

But nvidia cant push back too hard. As they will just become a monopoly. Amd can do whatever, I bet they have pascal ready but dont want to release it until AMD greenland comes out