AMD GDC2016 Thread

[ShintaiDK]

On their graphic, Vega has a slightly higher performance per watt compared to Polaris. Perhaps that's simply because it's HBM?

HBM is one reason. But bigger chips tends to be more efficient. If you exclude the 750/750TI and 270X that seems to go against the stream. Then 980 and 980TI got higher performance/watt than 950 and 960 for example. Its a bit more messy on the AMD front due to much more random uarchs in the SKU lines. But 290/290X beats 280X and 285 as well.

And remember its an unmarked chart. It could be for anything. Compute, 4K res, VR etc.

 

[ShintaiDK]

Raja said they will bring HBM 2 when it will make sense to be used in mainstream. That only tells us that HBM 2 as HBM 1 needs more volume in order to drop the price and initial cost investment.

HBM was made to replace GDDR-5, if they cannot bring the cost close to the same level between the two, then HBM wouldn’t be in production and used in an actual product by now.

HBM doesn't need to beat GDDR5 in price before it makes sense. And I doubt HBM will ever in its life get near GDDR5 in cost. GDDR5X is just another GDDR5 that will postpone HBM in anything but flagship for a step longer. But I doubt they can do it again. And then you slowly see HBM moving down in SKUs.

You also use SSD even tho it got lower GB/$ than a HD for the obvious reason.

 

[maddie]

It's not just the HBM chips.

IIRC, the Interposer process adds ~$30 to the PCB cost (read this somewhere legit awhile back). TSV I do not know the prices, but that process itself was the source for Fiji's low volume as the company that did it only went into volume production a month after Fury X was released.

Published papers say 1$ / 100mm^2 for the interposer a while ago.

http://electroiq.com/blog/2012/12/lifting-the-veil-on-silicon-interposer-pricing/

Sesh Ramaswami, managing director at Applied Materials, showed a cost analysis which resulted in 300mm interposer wafer costs of $500-$650 / wafer. His cost analysis showed the major cost contributors are damascene processing (22%), front pad and backside bumping (20%), and TSV creation (14%).
Ramaswami noted that the dual damascene costs have been optimized for front-end processing, so there is little chance of cost reduction there; whereas cost of backside bump could be lowered by replacing polymer dielectric with oxide, and the cost of TSV formation can be addressed by increasing etch rate, ECD (plating) rate, and increasing PVD step coverage.
Since one can produce ~286 200mm2 die on a 300mm wafer, at $575 (his midpoint cost) per wafer, this results in a $2 200mm2 silicon interposer.

 

[maddie]

HBM is one reason. But bigger chips tends to be more efficient. If you exclude the 750/750TI and 270X that seems to go against the stream. Then 980 and 980TI got higher performance/watt than 950 and 960 for example. Its a bit more messy on the AMD front due to much more random uarchs in the SKU lines. But 290/290X beats 280X and 285 as well.

And remember its an unmarked chart. It could be for anything. Compute, 4K res, VR etc.

That's fantastic.

Make a statement and then throw out the actual real world data that contradicts what you claimed.

True to form.

 

[NTMBK]

HBM is one reason. But bigger chips tends to be more efficient. If you exclude the 750/750TI and 270X that seems to go against the stream. Then 980 and 980TI got higher performance/watt than 950 and 960 for example. Its a bit more messy on the AMD front due to much more random uarchs in the SKU lines. But 290/290X beats 280X and 285 as well.

And remember its an unmarked chart. It could be for anything. Compute, 4K res, VR etc.

Oh yeah, the 480 was way more efficient than the 460!

 

[maddie]

I believe the initial difficulty with HBM was the assembly of the memory stack. Aligning the die accurately and ensuring the complete stack was functional. Maybe they have mastered the process and are able to mass-assemble by now. With the volumes that a midrange product will need, a dedicated line could be used.

 

[Silverforce11]

Published papers say 1$ / 100mm^2 for the interposer a while ago.

http://electroiq.com/blog/2012/12/lifting-the-veil-on-silicon-interposer-pricing/

That's the raw material cost for the actual interposer, but the production of getting it to work with a chip and such add to the cost. While $30 may not be much, we have to factor in TSV production, getting those HBM stacks linked to the interposer and able to communicate with the GPU properly.. well, they didn't mention specific cost.

Overall, when AMD has publicly discussed HBM and HBM2, it's always an issue of cost.

There was a video awhile back where AMD's Robert Hallock was asked specifically about HBM on Polaris, and he basically said no, referring to "cost issues" and the market price sensitivity.

Essentially what they've been saying is, HBM tech is only for premium GPUs due to cost.

 

[maddie]

That's the raw material cost for the actual interposer, but the production of getting it to work with a chip and such add to the cost. While $30 may not be much, we have to factor in TSV production, getting those HBM stacks linked to the interposer and able to communicate with the GPU properly.. well, they didn't mention specific cost.

Overall, when AMD has publicly discussed HBM and HBM2, it's always an issue of cost.

There was a video awhile back where AMD's Robert Hallock was asked specifically about HBM on Polaris, and he basically said no, referring to "cost issues" and the market price sensitivity.

Essentially what they've been saying is, HBM tech is only for premium GPUs due to cost.

OK, I missed that. By any chance you have a link or remember which video? I would like a look to see for myself.

 

[el etro]

Oh yeah, the 480 was way more efficient than the 460! <img src="images/smilies/familiar/wtf.png" border="0" alt="" title="Wtf" smilieid="47" class="inlineimg" />

But 460 was already Fermi2. Or at least was the Fermi2 prototype, including even the transistors change that made all GF chips made from that day to reduce spreaded heat. 465 that was the Fermi1 equivalent sucked just as much.

 

[Headfoot]

The interesting cost scenario for HBM2 is on CPU/APUs. On GPUs you need its RAM pool one way or the other, its a matter of what it costs and how fast it is. With CPUs and APUs you dont include the RAM, thats an assumed extra cost. But if you interposed HBM along with an APU/CPU you increase cost that goes to you while eliminating the RAM cost on the end user's side. This math could work out favorably because the end user sees marked-up, full distribution to consumer prices for RAM where the HBM would be available to AMD @ volume prices straight from the manufacturer. So while the costs for RAM are lower in reality, the marked up end user price may not be so far off from the real cost of HBM. I dont have the numbers so I could be off, but its a more interesting equation than GDDR5 vs HBM straight across

 

[thilanliyan]

Off topic but where i live it goes from max of about 44C in summer to around -8C min in winter, if you count wind chill even colder. Would have to be up there for greatest temp differentials for an area that isn't a desert.

Try -50C to +30C...Edmonton, Canada lol. And yes, I DID experience that -50C, and that was WITHOUT the windchill, it was about 6 yrs ago. With the windchill it was -60C. It was like that 2 days in a row I remember. So cold my palms were freezing through the mitts...it was ridiculous lol

 

[itsmydamnation]

it gets that hot in Canada? The furthest north i have been is Seattle and in the summer it was nowhere near 30 more like mid teens.

 

[MrTeal]

it gets that hot in Canada? The furthest north i have been is Seattle and in the summer it was nowhere near 30 more like mid teens.

Depends where you are, places by the ocean tend to be more moderate. Where I am in the prairies, we usually get a few days to a dozen days or so of +30C or above in the summer, and couple dozen nights with air temperatures below -40C in the winter. Record high was 41C, so nothing crazy like you'd see in the Southern US, but record low of -50C makes for a pretty frosty commute.

 

[maddie]

Depends where you are, places by the ocean tend to be more moderate. Where I am in the prairies, we usually get a few days to a dozen days or so of +30C or above in the summer, and couple dozen nights with air temperatures below -40C in the winter. Record high was 41C, so nothing crazy like you'd see in the Southern US, but record low of -50C makes for a pretty frosty commute.

You say 41C is nothing crazy?

Amazing temp variations in the preceding posts. I live about 10 degrees from the equator and here the annual range averages +20C to +33C. And I still wake to cool [to me at least] misty mornings in the valley.

 

[BlitzWulf]

Off topic:anyone else think that Nvidia will be launching their own game engine today?

Tom Peterson hinted at it in an interview with Ryan Shrout recently.

Steve from Gamers Nexus mentioned during an interview with Crytek about ther new engine that Nvidia had a huge announcement to make today also implying that it had to do with engine development.

Edit: seems I was wrong but from their released "Gameworks roadmap" i stll think it's a possibility for GDC next year.

 

[MrTeal]

You say 41C is nothing crazy?

Amazing temp variations in the preceding posts. I live about 10 degrees from the equator and here the annual range averages +20C to +33C. And I still wake to cool [to me at least] misty mornings in the valley.

Keep in mind, that's a record high. That's still really hot, but most years probably top out low 30s. A few years ago my wife and I drove down to LA, and stopped along the way in Bullhead City, Arizona to visit her Aunt's house. As we started driving down the hills towards the river, you could literally watch the outside temp climb on the car display, several degrees per minute. It topped out at 115F (46C). When you stepped out, it was like getting hit with a hammer. The average daily temperature there is over 110F in July and August, so that's pretty a typical summer day. That's crazy; those people are nuts.

 

[kondziowy]

WTF, is Nvidia opening Gameworks? Maybe they found out AMD will add extreme tesellation in Polaris and there is no point anymore ?
Just Kidding, I guess they just have enough of internet wars. Thank god...
http://www.pcper.com/reviews/Graphics-Cards/Details-about-NVIDIAs-GameWorks-31-Update

Coming soon = after extreme tesellators in AMD are confirmed. Stupid joke ,sry :<

 

[boozzer]

gpu open just served it's purpose.

competition is a very, very nice thing. you could say it is essential.

 

[Paul98]

WTF, is Nvidia opening Gameworks? Maybe they found out AMD will add extreme tesellation in Polaris and there is no point anymore ?
Just Kidding, I guess they just have enough of internet wars. Thank god...
http://www.pcper.com/reviews/Graphics-Cards/Details-about-NVIDIAs-GameWorks-31-Update

Coming soon = after extreme tesellators in AMD are confirmed. Stupid joke ,sry :<

Excellent! Now I am excited, things seem to be looking up for PC gamers.

 

[kondziowy]

So is that it? Can AMD run PhysX on GPU with this now? Just like that after all those years?

 

[AnandThenMan]

So is that it? Can AMD run PhysX on GPU with this now? Just like that after all those years?

Highly doubt it. Just because Nvidia makes the source available doesn't mean anyone can modify it as they see fit. My hunch is this is a smoke screen, NV will use similar phrasing as used by AMD but the reality of Nvidia's proprietary tech won't change at all.

Hope I'm wrong.

 

[Silverforce11]

So is that it? Can AMD run PhysX on GPU with this now? Just like that after all those years?

Those are CPU PhysX only, it's been available last year.

And NV's youtube channel has been trolled a lot lately, everytime they showcase a new game with GameWorks, people automatically call it GimpWorks or 30fps cinematic experience. lol

Maybe finally they are getting a hint.

If they actually gave a damn about gamers, they would encourage more open source, develop cool features and allow developers free access to it and they can modify it at will and share it fully.

That is pushing the gaming industry forward. Not giving them a black box that deprives their ability to optimize for all hardware and hide the source code behind a special $ license.

 

[tential]

Highly doubt it. Just because Nvidia makes the source available doesn't mean anyone can modify it as they see fit. My hunch is this is a smoke screen, NV will use similar phrasing as used by AMD but the reality of Nvidia's proprietary tech won't change at all.

Hope I'm wrong.

It said after a time. They didn't say how long the times will be between the experimental phase and the usable phase for everyone.

 

[R0H1T]

Those are CPU PhysX only, it's been available last year.

And NV's youtube channel has been trolled a lot lately, everytime they showcase a new game with GameWorks, people automatically call it GimpWorks or 30fps cinematic experience. lol

Maybe finally they are getting a hint.

If they actually gave a damn about gamers, they would encourage more open source, develop cool features and allow developers free access to it and they can modify it at will and share it fully.

That is pushing the gaming industry forward. Not giving them a black box that deprives their ability to optimize for all hardware and hide the source code behind a special $ license.

Seems like smoke & mirrors to me, agree with the poster above you that they'll do this just for PR & damage control without changing Gimpworks too much in reality.

 

[Bacon1]

Seems like smoke & mirrors to me, agree with the poster above you that they'll do this just for PR & damage control without changing Gimpworks too much in reality.

Yep, they are already phasing out what they are releasing, HBAO+ is going away for VXAO, even though they don't look much different, VXAO is "better" tech and knocks an extra 20-30% off your performance, and is Maxwell 2nd gen only (980 TI+ only TBH).

Hairworks got trampled by TressFx / PureHair, no reason to keep it hidden now.