Post Xbox One Two Scorpio, what should Sony do next? *spawn* (oh, and Nintendo?)

[function]

Interesting post in the AMD thread:

Here you go.

Samsung DRAM catalog. Looks like 8Gb GDDR5 chips clock lower than 4Gb (understandable), but that low power GDDR5 chips (1.35v instead of 1.5v) are already up to 6 gHz, with higher to come.

This means Neo is probably (?) using 1.5v chips, while by next year the 6.67 gHz chips that Scorpio will need might well be available at 1.35v, meaning somewhat lower power for that phat 384-bit bus without needing to go with GDDR5X or HBM.

Which I think is something that Mr Fox was suggesting as a possibility.

 

[BRiT]

Unless Vega is using a different node or a new architecture then power/perf for the same diesize is not going be dramatically different than RX480. Again wishful thinking on your part, not very much evidence

Yes, they will be using different nodes and architecture. Lolaris is using GloFlo while Vega is said to be using the substantially better TSMC.

 

[function]

Yes, they will be using different nodes and architecture. Lolaris is using GloFlo while Vega is said to be using the substantially better TSMC.

I wonder if Polaris is simply pushing too far out of it's comfort zone in an effort to chase the 980. At 911 mHz and under 1v I could see it being a very efficient setup.

 

[AlNom]

What year is this ?

Year of PlayStation?

 

[Jubei]

Yes, they will be using different nodes and architecture. Lolaris is using GloFlo while Vega is said to be using the substantially better TSMC.

Vega is the same architecture as Polaris with minor enhancements. And there is zero evidence that TSMC is better than GloFo (Samsung really)

Apple dual-sourced 16/14nm LPE from both TSMC and Samsung for the A9, its the only real comparison we have and according to their own tests and Anandtechs, the difference is negligable

http://www.anandtech.com/show/9708/analyzing-apple-statement-for-tsmc-and-samsung-a9

I think AMD fans are unfairly blaming GloFo for Polaris simply not being as efficient architecture as Pascal. I mean Maxwell is hitting the same perf/watt at 28nm

 

[Deleted member 11852]

Can we rename the the thread 'Psychotic Fanboy Jizzfest' ?

 

[iroboto]

Can we rename the the thread 'Psychotic Fanboy Jizzfest' ?

The thread isn't Terrible, I think the issue is that people are playing fantasy football which is more a game, and not so much advancing a real discussion about the positions they are in; what we know, what we don't know; what are realistic economics and what are realistic goals. What are set assumptions for all these discussions, what are real drivers to purchasing behaviour and what are not.

And if those cant be defined at least regions should be graphed. Guys aren't working towards a unified answer, they are looking to win the discussion.


Sent from my iPhone using Tapatalk

 

[MrFox]

Can we rename the the thread 'Psychotic Fanboy Jizzfest' ?

But then... how would we differentiate this one from all the other threads?

 

[Deleted member 11852]

The thread isn't Terrible

Your appetite for fanboy jizz is clearly greater than mine.

Nom nom nom.

 

[eastmen]

It was a different architecture? 780 TI = Kepler, 980 = Maxwell

Try again

I don't have to try. Polaris and Vega are different implementations of GCN. With Polaris AMD tried to regain efficiency and power savings . There is no reason to think they wont carry that forward and introduce a newer more efficient GCN with Vega.

That is why I posted about the 780 and 980. Because with the same process NVidia was able to offer better performance with design changes. AMD has 2 Vega chips coming out between now and Scorpio. There is also a highly unlikely chance that AMD can pull Navi forward for MS.

Taking the first 14nm chip amd came out with and saying this is it in terms of power consumption for their tech is naïve

 

[eastmen]

Yes, I have no idea what you were getting at anyway...I was assuming you were talking about it being why PS4 outsold XB0...so what were you talking about?

I was talking about the power difference showing up in games. This time that advantage will be on MS's side creating a halo product for them. Ps4- Neo or PS4 to Scorpio . If Scorpio has the better looking and playing games why would anyone go for Neo ?

 

[function]

Vega is the same architecture as Polaris with minor enhancements. And there is zero evidence that TSMC is better than GloFo (Samsung really)

Apple dual-sourced 16/14nm LPE from both TSMC and Samsung for the A9, its the only real comparison we have and according to their own tests and Anandtechs, the difference is negligable

http://www.anandtech.com/show/9708/analyzing-apple-statement-for-tsmc-and-samsung-a9

I think AMD fans are unfairly blaming GloFo for Polaris simply not being as efficient architecture as Pascal. I mean Maxwell is hitting the same perf/watt at 28nm

I'm sure architecture is an issue, but there seem to be a number of factors converging.

For one thing, AMD seem to have pushed higher than was wise (in clocks and voltage) given the yeilds they wanted. Different 480s come with different vtables out of the box - some rather higher than others. And while some 480s undervolt really well and are reported to shed 10~30 W off typical gaming loads - while simultanesouly boosting more and so scoring higher while generating less fan noise - others have little appetite for running more frugally and consequently boosting fully.

With a more aggressive approach to binning the 480 looks like it could have been both faster and cooler. While I'm sure this could be the case for any chip, the size of the variations is suprising IMO.

The fact AMD have been shipping 8GB cards locked to 4GB in the bios is also ... suprising. And potentially quite wasteful. Not to mention that they're using 8 gHz ram which is crazy fast and something that Nvidia decided to steer clear of for whatever reason. Underclocking and undervolting the ram has also been reported to bring drops in power, and the ability to sustain higher levels of boost and therefore rasing overall performance.

The 480 smacks of a product that hasn't been tuned optimally for whatever reason. Hopefully there will be a better binned, lower volted, 7 gHz (1.35v?) GDDR5 Radeon 485 at some point before it's irrelevant.

Regardless of the 480 iissues, and even if Neo is on GF, with a clock of 911 and avoiding 8 gHz GDDR5 I don't seem Sony having any of the issues associated with the 480 launch. At those speeds the machine should be using very manageable levels of power. I wouldn't assume the same for a hypotherical 6TF Scorpio based on the current Polaris 10 configuration.

 

[eastmen]

I'm sure architecture is an issue, but there seem to be a number of factors converging.

For one thing, AMD seem to have pushed higher than was wise (in clocks and voltage) given the yeilds they wanted. Different 480s come with different vtables out of the box - some rather higher than others. And while some 480s undervolt really well and are reported to shed 10~30 W off typical gaming loads - while simultanesouly boosting more and so scoring higher while generating less fan noise - others have little appetite for running more frugally and consequently boosting fully.

With a more aggressive approach to binning the 480 looks like it could have been both faster and cooler. While I'm sure this could be the case for any chip, the size of the variations is suprising IMO.

The fact AMD have been shipping 8GB cards locked to 4GB in the bios is also ... suprising. And potentially quite wasteful. Not to mention that they're using 8 gHz ram which is crazy fast and something that Nvidia decided to steer clear of for whatever reason. Underclocking and undervolting the ram has also been reported to bring drops in power, and the ability to sustain higher levels of boost and therefore rasing overall performance.

The 480 smacks of a product that hasn't been tuned optimally for whatever reason. Hopefully there will be a better binned, lower volted, 7 gHz (1.35v?) GDDR5 Radeon 485 at some point before it's irrelevant.

Regardless of the 480 iissues, and even if Neo is on GF, with a clock of 911 and avoiding 8 gHz GDDR5 I don't seem Sony having any of the issues associated with the 480 launch. At those speeds the machine should be using very manageable levels of power. I wouldn't assume the same for a hypotherical 6TF Scorpio based on the current Polaris 10 configuration.

Also don't forget this is the Stock cooler for the rx 480

I am sure when the partner cards come out with better coolers we will see the true performance capability of the gpu.

 

[Jubei]

I don't have to try. Polaris and Vega are different implementations of GCN. With Polaris AMD tried to regain efficiency and power savings . There is no reason to think they wont carry that forward and introduce a newer more efficient GCN with Vega.

That is why I posted about the 780 and 980. Because with the same process NVidia was able to offer better performance with design changes. AMD has 2 Vega chips coming out between now and Scorpio. There is also a highly unlikely chance that AMD can pull Navi forward for MS.

Taking the first 14nm chip amd came out with and saying this is it in terms of power consumption for their tech is naïve

Honestly at this point i think you must be trolling. How can anyone sit there and compare Kepler vs Maxwell and then try to claim Vega will offer the same effiency improvements over Polaris even though there is not one shred of evidence supporting that.
Look at the AMD roadmap, the 2,5x perf/watt improvement comes with Polaris because its the new uarch, Vega is building off that uarch, there will be some improvements but NOTHING similar to Kepler vs Maxwell

The majority of performance from Vega will be from more CUs, the improvement in power draw will be from HBM2 but it will also cost a fair amount more than RX480

Im done responding to you

 

[function]

Also don't forget this is the Stock cooler for the rx 480

I am sure when the partner cards come out with better coolers we will see the true performance capability of the gpu.

The stock cooler is another thing that makes me think AMD tried to pitch upwards very late and after much of the 480 was already finalised.

That's a pretty toss cooler for a 170W~180W card (that was actually supposed to be a 150W MAX card).

Too wide a bin, 8 gHz memory drawing more power than the card's delivery system was designed for, pushing clocks to chase the 980, overshooting on power and being stuck with a cooler designed for back when they weren't delivering a 180W 150W MAX card.

Partners can sort put the power delivery and the cooler, but unless AMD are prepared to introduce a better bin for their partners then it'll still be a chip out of its ideal window. And it still won't clock much higher due to the limits of the architecture.

Hopefully AMD are holding out the really sweet Plolaris 10 chips for mobile, where they can be paired with more power efficient memory and clocked in the chips ideal range. Desktop Polaris 10 has landed witha bit of thud IMO, due to AMD squeezing it out of its ideal (for the silicon) position.

 

[eastmen]

Honestly at this point i think you must be trolling. How can anyone sit there and compare Kepler vs Maxwell and then try to claim Vega will offer the same effiency improvements over Polaris even though there is not one shred of evidence supporting that.
Look at the AMD roadmap, the 2,5x perf/watt improvement comes with Polaris because its the new uarch, Vega is building off that uarch, there will be some improvements but NOTHING similar to Kepler vs Maxwell

The majority of performance from Vega will be from more CUs, the improvement in power draw will be from HBM2 but it will also cost a fair amount more than RX480

Im done responding to you

That's good because you never read what I write and assume way to much.

I never claimed Vega would offer the same level of improvement as Kepler to Maxwell. I used Kepler and Maxwell to refute your claim that improvements can't happen. Because they can.

You admit as much in your post that there will be improvements to the perf/watt . It may not be as big as Kepler vs Maxwell but there will be improvements. There will also be improvements to the 14nm process.

You may be right about where the majority of improvements come from with the new tech but that doesn't mean its the only place.

Vega will be an improvement over Polaris , 14nm will improve . There is a year before launch for all of that.

But AMD is doomed cause of the rx 480 stumbles

 

[MrFox]

If Scorpio has the better looking and playing games why would anyone go for Neo ?

Personal favorite franchises?
Must-have titles being exclusives?
Friends have a PS4?
Upgrading from a PS4?
Lower price?
Earlier launch date?
VR cost of entry?
Already own PSVR and games?
Brand reputation and trust?

Visual differences not really noticeable without a side-by-side comparison?

 

[eastmen]

Personal favorite franchises?
Must-have titles being exclusives?
Friends have a PS4?
Upgrading from a PS4?
Lower price?
Earlier launch date?
VR cost of entry?
Already own PSVR and games?
Brand reputation and trust?

So basicly stay with a ps4 and upgrade to Scorpio. Why would they go from a $350-$400 to another $400 + system ?

The ps4 will let them play with their friends , cost nothing since they own it , already own it so they don't have to worry about Launch date, they own the system so the only cost is the vr helm.

As for reputation and trust , they own a ps4 . Of course you and I have different opinions on Sony.

 

[Jubei]

I'm sure architecture is an issue, but there seem to be a number of factors converging.

For one thing, AMD seem to have pushed higher than was wise (in clocks and voltage) given the yeilds they wanted. Different 480s come with different vtables out of the box - some rather higher than others. And while some 480s undervolt really well and are reported to shed 10~30 W off typical gaming loads - while simultanesouly boosting more and so scoring higher while generating less fan noise - others have little appetite for running more frugally and consequently boosting fully.

With a more aggressive approach to binning the 480 looks like it could have been both faster and cooler. While I'm sure this could be the case for any chip, the size of the variations is suprising IMO.

The fact AMD have been shipping 8GB cards locked to 4GB in the bios is also ... suprising. And potentially quite wasteful. Not to mention that they're using 8 gHz ram which is crazy fast and something that Nvidia decided to steer clear of for whatever reason. Underclocking and undervolting the ram has also been reported to bring drops in power, and the ability to sustain higher levels of boost and therefore rasing overall performance.

The 480 smacks of a product that hasn't been tuned optimally for whatever reason. Hopefully there will be a better binned, lower volted, 7 gHz (1.35v?) GDDR5 Radeon 485 at some point before it's irrelevant.

Regardless of the 480 iissues, and even if Neo is on GF, with a clock of 911 and avoiding 8 gHz GDDR5 I don't seem Sony having any of the issues associated with the 480 launch. At those speeds the machine should be using very manageable levels of power. I wouldn't assume the same for a hypotherical 6TF Scorpio based on the current Polaris 10 configuration.

The only controversy to me is breaking specs. Otherwise most of the issues with RX480 are normal IMO, Samsungs LPP process is likely denser than TSMC wich i ncreases leakage wich leads to heat and increased power since its only 232mm2 and as you mention AMD is trying to push the clocks more than they should. Intel had a similar issue with Haswell, but they had the luxury of selling the best bins as overclockable

So pushing clocks over a small die size with a uarch that is not as efficient as Pascal and you have RX 480. Still a damn good card for 199 but TSMC FF+ wont solve all those things, maybe some of them

And i agree, its likely to be underclocked with slower ram speed if its used in Neo.

 

[Shifty Geezer]

I never claimed Vega would offer the same level of improvement as Kepler to Maxwell. I used Kepler and Maxwell to refute your claim that improvements can't happen

He said they don't happen on the same architecture on the same node. Maxwell improved over Keplar by being a brand new architecture. Is Vega a radical departure from Polaris and able to usher in significantly better performance per watt?

So basicly stay with a ps4 and upgrade to Scorpio. Why would they go from a $350-$400 to another $400 + system .

They can sell the PS4 for $200 (or whatever) and get the Neo cheaper, that'll still play their existing library only better and connect with all their Sony experience, in one box under the TV. Buying a Scorpio will cost more because there's no trade-in option, require a second box and the faf of that (what if your friends are playing COD on PSN and they want you to join in, but you're playing Scorpio at the time and have no idea they're even online?), and provide the PS4 exclusive experience in a lower quality than the person wanting a more powreful console would like.