[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[Silverforce11]

I don't see HBM making it down into 3rd tier products for two reasons:

1. Low yields, its been public knowledge.
2. Expensive.

These factors don't contribute to making lower entry-level gaming SKUs price competitive, selling a 370X with HBM at ~$200 would unlikely be profitable with #1 & #2.

If 370X has HBM, the entire stack will have be priced upwards.

380X, $499? Quite expensive for a mid-range product. While it may have the performance to justify its price, the high price of the stack is going to reduce sales. It's well beyond the pricing sweetspot.

Here's the kicker: If the 370X with HBM is ~780 performance, that makes it slower than the 970. How much would AMD have to price that SKU for to have good margins? $249 - $299? In what world would gamers buy a 370X at those prices over a 970 at $300?

 

[maddie]

I think we would all be guessing at this point - but to me it would make more sense for the 370X to be lined up with the 970, the 380 competing with the 980 and the 390X somewhere between the 980 and the Titan X. Who knows how it all actually ends up playing out though.

If we accept your guess, then what will the 380X and the 390 be marketed against?

 

[maddie]

I don't see HBM making it down into 3rd tier products for two reasons:

1. Low yields, its been public knowledge.
2. Expensive.

These factors don't contribute to making lower entry-level gaming SKUs price competitive, selling a 370X with HBM at ~$200 would unlikely be profitable with #1 & #2.

If 370X has HBM, the entire stack will have be priced upwards.

380X, $499? Quite expensive for a mid-range product. While it may have the performance to justify its price, the high price of the stack is going to reduce sales. It's well beyond the pricing sweetspot.

Here's the kicker: If the 370X with HBM is ~780 performance, that makes it slower than the 970. How much would AMD have to price that SKU for to have good margins? $249 - $299? In what world would gamers buy a 370X at those prices over a 970 at $300?

I don't understand the reasoning here.
370X slower that 970 and cheaper.
380 faster than 970 and more expensive.
Why assume $300 is a magic number?

 

[raghu78]

I don't see HBM making it down into 3rd tier products for two reasons:

1. Low yields, its been public knowledge.
2. Expensive.

These factors don't contribute to making lower entry-level gaming SKUs price competitive, selling a 370X with HBM at ~$200 would unlikely be profitable with #1 & #2.

If 370X has HBM, the entire stack will have be priced upwards.

380X, $499? Quite expensive for a mid-range product. While it may have the performance to justify its price, the high price of the stack is going to reduce sales. It's well beyond the pricing sweetspot.

Here's the kicker: If the 370X with HBM is ~780 performance, that makes it slower than the 970. How much would AMD have to price that SKU for to have good margins? $249 - $299? In what world would gamers buy a 370X at those prices over a 970 at $300?

I think people might have to wait and see if the AMD naming convention has changed this generation due to binning and yields. AMD is manufacturing GPUs on a completely new production process (2.5D stacking) for the first time in high volume. Its expected that yields are going to be difficult so it would not surprise me to see more than two SKUs using the same chip.

Normally we expect R9 390 / R9 390X to be made from the high end chip and R9 380 / R9 380X to be made from the mid range chip and R9 370X / r9 370 to be made from the entry level chip. But things might be different this time. I am keen to see AMD's pricing as they are playing catchup with Nvidia after being so late this gen. I am also keen to see how competitive AMD is in terms of perf/sq mm and perf/watt against Nvidia's Maxwell GPUs. I am also interested in how good the OC headroom is for all the R9 3xx GPUs. For their own good and for the GPU market AMD needs to reply strongly with an extremely competitive GPU stack. :thumbsup:

btw if any GPU below R9 380 sports HBM I would be pleasantly surprised.

 

[3DVagabond]

Considering the price of graphics cards doubled from Nov 2010 (gtx 580) to Feb 2013 (Titan), 27 months, they should be able to afford HBM at current pricings.

 

[JDG1980]

I don't see HBM making it down into 3rd tier products for two reasons:

1. Low yields, its been public knowledge.
2. Expensive.

There is no "public knowledge" about HBM yields - just rumors, speculation, and a handful of alleged leaks. I did a Google search on HBM yields and came up with nothing solid. Here is a message board thread where a few people indicate sources told them yields are good. Take that with a grain of salt, of course - but you should also reserve that salt for the people insisting yields will be very low. The truth is, we don't know for sure. No one outside of the affected companies does. The same is true of expense; we have no idea how much it will cost.

These factors don't contribute to making lower entry-level gaming SKUs price competitive, selling a 370X with HBM at ~$200 would unlikely be profitable with #1 & #2.

A $200 price point for a R9 370X is unrealistic. The R9 270X had an MSRP of $199 when it was released, but that was because it was a straight rebrand of a chip that was over 18 months old at the time. When the 7870 was first released, it was $350.

If 370X has HBM, the entire stack will have be priced upwards.

380X, $499? Quite expensive for a mid-range product. While it may have the performance to justify its price, the high price of the stack is going to reduce sales. It's well beyond the pricing sweetspot.

Of course the stack will be priced upwards - whether all the cards have HBM or not. That's the whole point of spending R&D dollars on a refresh. The only way the stack won't be priced upwards is if the pessimistic rumors are true and it's full of unmodified rebrands.

Here's the kicker: If the 370X with HBM is ~780 performance, that makes it slower than the 970. How much would AMD have to price that SKU for to have good margins? $249 - $299? In what world would gamers buy a 370X at those prices over a 970 at $300?

If we're talking about 780-class performance at 140W, the retail price would have to be $249 at most. Keep in mind that AMD currently has the R9 290 at that price point - a large (438 sq. mm.) chip with a very wide external memory bus that needs an expensive PCB. The high TDP also means high-quality (and high-cost) VRMs. Tonga is only 359 sq. mm. and can probably be improved with more optimization. A hypothetical HBM-enabled version would be able to use a far simpler PCB than Hawaii, with less expensive external components.

 

[gamervivek]

370 was 64% faster in Manhattan Offscreen leaked benchmarks than the 270, which would put it right next to 780. Though the test was in OpenGL,

http://lensfire.blogspot.in/2015/01/amd-radeon-r9-370-benchmarks-and.html

 

[lyssword]

amazing if true about 370x

 

[Silverforce11]

I don't understand the reasoning here.
370X slower that 970 and cheaper.
380 faster than 970 and more expensive.
Why assume $300 is a magic number?

You don't understand because you haven't paid attention to the recent market.

The R290X is FASTER and cheaper than 970. With prices regularly hitting $280 and sometimes lower.

The R290 is on-par and without the gimped vram situation and is a lot cheaper, with prices around $200-$240 mark.

AMD has to price it so low just to clear inventory because gamers don't want to buy it when they can get the 970.

If the 370X end up slower than the 970 at a similar price, you can forget about AMD making many sales of such a product. It's an instant KO, DOA. It needs to be around $200 if its slower else the market will reject it.

I have my doubts they can be very profitable selling a HBM part for that low.

 

[Silverforce11]

If we're talking about 780-class performance at 140W, the retail price would have to be $249 at most. Keep in mind that AMD currently has the R9 290 at that price point - a large (438 sq. mm.) chip with a very wide external memory bus that needs an expensive PCB. The high TDP also means high-quality (and high-cost) VRMs. Tonga is only 359 sq. mm. and can probably be improved with more optimization. A hypothetical HBM-enabled version would be able to use a far simpler PCB than Hawaii, with less expensive external components.

That's my point, 780 class performance cannot go for much, definitely I agree $249 is max. At this price, its competitor is the 970, with a few of them hitting the $300 mark.

You could also say its competing against the 960 at $200, and at 140W, if it has 4GB HBM, it would be a clear winner. But I think the majority would just pay extra and go with a 970, as they have been doing when the R290/X is around that price mark. The NV tax is strong.

 

[3DVagabond]

The 290(x) is perceived as old and less desirable. Then there's the efficiency factor that nVidia marketing has hammered the internet with and the perception of it being hot and loud. The 370X wouldn't suffer from any of those drawbacks. If it has the roughly same performance as the 970 and is cool, quiet, efficient, and offers (a true) 4GB of high performance HBM RAM there's no reason to think it would have to be cheaper.

 

[gamervivek]

If 370X is a bit slower than the 970, but consumes less power and also has the full fat 4GB memory then AMD could charge higher for it considering the 290X vs. 970 debate.

I doubt that happening though.

 

[maddie]

You don't understand because you haven't paid attention to the recent market.

The R290X is FASTER and cheaper than 970. With prices regularly hitting $280 and sometimes lower.

The R290 is on-par and without the gimped vram situation and is a lot cheaper, with prices around $200-$240 mark.

AMD has to price it so low just to clear inventory because gamers don't want to buy it when they can get the 970.

If the 370X end up slower than the 970 at a similar price, you can forget about AMD making many sales of such a product. It's an instant KO, DOA. It needs to be around $200 if its slower else the market will reject it.

I have my doubts they can be very profitable selling a HBM part for that low.

Still not making any sense.
You are totally disregarding power consumption in your comparison of the R290 vs 970 situation. A lot of the buying public have come to believe the low power mantra. The 290 aren't selling because people see them as power hogs, true or not.

If the latest rumor is true then these cards are at least as power efficient as Maxwell, changing the dynamic.

 

[Azix]

Considering the price of graphics cards doubled from Nov 2010 (gtx 580) to Feb 2013 (Titan), 27 months, they should be able to afford HBM at current pricings.

That's just nvidia insanity. Compare to 780 ti maybe

 

[3DVagabond]

That's just nvidia insanity. Compare to 780 ti maybe

The Titan-X is $1000. It's nothing more than the direct ascendant of the 580. The original Titan's DP and the current Titan's 12GB of memory are just marketing points to justify it. The current top card from nVidia is $1000 and has been so for 2 years now. Trust me, unless they are forced to, it won't go down.

 

[lyssword]

It would be really exciting if there are new midrange cards, high end cards were always out of reach for me

 

[Silverforce11]

Still not making any sense.
You are totally disregarding power consumption in your comparison of the R290 vs 970 situation. A lot of the buying public have come to believe the low power mantra. The 290 aren't selling because people see them as power hogs, true or not.

If the latest rumor is true then these cards are at least as power efficient as Maxwell, changing the dynamic.

It doesn't matter if its as efficient or more, the mantra will switch to something else NV can market. Could be the return of PhysX, amplified by GameWorks or just "NV drivers are better". The NV tax is there to stay, ie. NV GPUs will fetch more for similar performance at any bracket.

AMD cannot price a part close to NV's offerings if its slower. It has to be significantly cheaper if its slower. They only chance they can price it similar, is if its outright faster & as efficient or more.

 

[3DVagabond]

It doesn't matter if its as efficient or more, the mantra will switch to something else NV can market. Could be the return of PhysX, amplified by GameWorks or just "NV drivers are better". The NV tax is there to stay, ie. NV GPUs will fetch more for similar performance at any bracket.

AMD cannot price a part close to NV's offerings if its slower. It has to be significantly cheaper if its slower. They only chance they can price it similar, is if its outright faster & as efficient or more.

What creates the situation you are referring to, and I'm not saying it doesn't exist, is marketing.

AMD needs to match nVidia's marketing. They can start with making sure there aren't any glaring faults with their new product upon release. Then they too need to drive home the strengths of their products. GCN in the consoles. DX12 and Vulkan having Mantle at their root. Continuing with Mantle 2 and adding features/performance that's exclusive to them. Taking advantage of GCN's compute advantage. They need to market themselves as a generation ahead as nVidia won't have HBM until their next generation. Et cetera.

 

[boozzer]

It doesn't matter if its as efficient or more, the mantra will switch to something else NV can market. Could be the return of PhysX, amplified by GameWorks or just "NV drivers are better". The NV tax is there to stay, ie. NV GPUs will fetch more for similar performance at any bracket.

AMD cannot price a part close to NV's offerings if its slower. It has to be significantly cheaper if its slower. They only chance they can price it similar, is if its outright faster & as efficient or more.

come on, don't be a downer. this time around there is no coin mining to rain on amd's parade.

If the 370x performance/watt is true. amd will regain marketshare by the truckload.

nv tax is just marketing. who says amd can't also do a marketing campaign? just springle some greens on the popular review sites, doesn't even need to be all of them.

 

[Vaporizer]

If this is true i will jump on a r9 370x. It will be a complete new series in which you get a decent fps boost and not only power consuption as some competitors do. I assume that a r9 380 will be in the price range of gtx 970 and will be some % faster.

 

[3DVagabond]

If this is true i will jump on a r9 370x. It will be a complete new series in which you get a decent fps boost and not only power consuption as some competitors do. I assume that a r9 380 will be in the price range of gtx 970 and will be some % faster.

And won't have that BS VRAM setup.

 

[monstercameron]

this bodes well, already a design win http://www.pcworld.com/article/2917...other-unannounced-radeons-in-new-systems.html

 

[desprado]

What creates the situation you are referring to, and I'm not saying it doesn't exist, is marketing.

AMD needs to match nVidia's marketing. They can start with making sure there aren't any glaring faults with their new product upon release. Then they too need to drive home the strengths of their products. GCN in the consoles. DX12 and Vulkan having Mantle at their root. Continuing with Mantle 2 and adding features/performance that's exclusive to them. Taking advantage of GCN's compute advantage. They need to market themselves as a generation ahead as nVidia won't have HBM until their next generation. Et cetera.

It is not about marketing.There are many other factors which plays.For example 4 months driver delay and Freesync CF support driver delay.AMD has a very bad consumer support.

See this topic which is another factor.
http://forums.guru3d.com/showthread.php?t=398858

 

[beginner99]

So HBM and 20nm? That's the only way I can see such a drastic power reduction.

I'm not saying it's not possible, it just seems less likely that they went from 250-300W of power usage at the same level of performance to 170W on the same node.

Even this rumor is more than questionable, don't forget that it is the same node but on a different process. And estimates say that that new process is a lot better (based on other parts moving from TSMC 28 nm to GF 28 nm). Therefore the design could be more efficient than maxwell simply due to being made on a different process.

 

[3DVagabond]

this bodes well, already a design win http://www.pcworld.com/article/2917...other-unannounced-radeons-in-new-systems.html

We have no idea what it is though. It could be a rebranded OEM only SKU like the 8000 series was. It could be latest tech w/HBM, but it doesn't say anything.