AMD Q3 2017 earnings

[The Stilt]

Pinnacle Ridge and Vega refresh are expected on 12nm.

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

 

[guskline]

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

Agreed Stilt.

Ryzen/TR/Epyc where the reall $$ comes from.

 

[raghu78]

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

Mark Papermaster has confirmed Ryzen and Vega refreshes on 12LP.

http://www.tomshardware.com/news/amd-ryzen-vega-12nm-lp-2018,35502.html

"We followed up with Papermaster in person and confirmed directly that the company will transition both Vega GPUs and the Ryzen line of processors to the 12nm LP process."

Vega might still have hope if AMD get its complete feature set working with the December Crimson driver release. Then if 12LP can reduce its power draw and bring a die shrink then AMD will be better positioned to atleast compete with GV104. AMD would have done very well if Vega on 12LP can stay close behind GV104. Then with Navi and multi die on TSMC 7nm they can once gain try and compete at the highest end.

 

[Yotsugi]

TSMC 7nm

>Doubt

I can't quite see any point in throwing even more money at a non-salvageable design.

It's more than salvageable. Needs software.

 

[SlowSpyder]

Great news for AMD, desktop CPU's are much more interesting with two healthy companies.

 

[Cloudfire777]

I cant believe how many biased stock news sites there is that write not just one, but several articles a day about the same thing, presenting a skewed and untrue representation of yesterdays result.

TheStreet for example. My god, they are obviously doing everything they can to push price down for whatever reason. Seberal of those sites clearly have an agenda. No way. I refuse to belive humans are that thick.

Take a look:
https://finance.yahoo.com/quote/amd?ltr=1

 

[dawheat]

I cant believe how many biased stock news sites there is that write not just one, but several articles a day about the same thing, presenting a skewed and untrue representation of yesterdays result.

TheStreet for example. My god, they are obviously doing everything they can to push price down for whatever reason. Seberal of those sites clearly have an agenda. No way. I refuse to belive humans are that thick.

Take a look:
https://finance.yahoo.com/quote/amd?ltr=1

In fairness a ton of upside and future growth was factored into AMD's stock rise over the last year. It's been a productive stock to buy-sell this year and I'll look to get back in when it drops between $11-12 again.

 

[Glo.]

In all of the talk about Vega on 12 nm process, everybody appears to be forgetting most important part. Vega 10 is not the only GPU in Vega lineup.

There will be smaller Vega GPUs.

 

[HurleyBird]

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

Assuming that the problem is (mostly) on the hardware side of things, it might be possible that there is some tragic flaw that could greatly improve the situation when fixed. R600 -> Terrascale comes to mind.

Or Vega 20 has enough sunk cost at this point coupled with AMD needing a new product with good double precision throughput.

 

[frozentundra123456]

In fairness a ton of upside and future growth was factored into AMD's stock rise over the last year. It's been a productive stock to buy-sell this year and I'll look to get back in when it drops between $11-12 again.

Yes, very valid point. After all, they finally made a profit, but using the most favorable value, earnings were only 0.10 per share, and guidance for next quarter is for lower earnings. And the stock has made a huge run already.

 

[turtile]

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

I'd assume that the amount of revenue they will bring in from the professional side will more than cover the cost of moving to 12nm. Higher clocks or lower power will definitely help Vega a lot.

 

[Qwertilot]

When it'll be up against Volta? Unless there's a miracle repair job available it'd be more or less a waste of money trying.

 

[Jan Olšan]

Guys (ladies?), a company has to keep selling what it has in store until it has the next thing. That is why they didn't stop making CPUs in 2011 to patiently wait 6 years for Zen. If you want to keep existing, you have to sell stuff. "Guys, this thing is less good than we thought it woudc be, let's just drop it immediately and leave the whole market to competition," is not a very sound idea.

 

[Cloudfire777]

In fairness a ton of upside and future growth was factored into AMD's stock rise over the last year. It's been a productive stock to buy-sell this year and I'll look to get back in when it drops between $11-12 again.

True, but todays stock value should reflect on future outcome as well.
AMD have barely begun shipping EPYC for datacenters which is where the potential is huge. They havent launched Raven Ridge yet, but will in Q4. They finally posted profits after so many years with red numbers.

How could this not make the stock go higher?
I can tell you why. Investors, the same morons that told people to sell AMD stocks when it was down to $2 (almost everyone had a «sell rating» back then), heard Lisa Su yesterday say that Q4 will go down compared to Q3. That was all it took. These individuals, which have almost none experience with hardware, cant grasp the fact that historically Q4 usually been lower than Q3, so they freaked out. And like any mindless sheep, they do their everyday business by looking at sell and buy walls of stocks, meaning they follow the rest when a sell wall appears. Which short the stock value.

 

[krumme]

Perhaps Lisa want the stock down. Lol.
She sure know what the impact is from the guidance. There could be reasons.
Whatever.
Epyc looks nice from Johan review so its just what it is. Slow uptake.

Amd is still tied to the wsa. Anyone guessing about stock price apparently knows the terms and how it will end 2024? Or what?

 

[ajc9988]

Perhaps Lisa want the stock down. Lol.
She sure know what the impact is from the guidance. There could be reasons.
Whatever.
Epyc looks nice from Johan review so its just what it is. Slow uptake.

Amd is still tied to the wsa. Anyone guessing about stock price apparently knows the terms and how it will end 2024? Or what?

No, this effects her performance stock units for the year! Meanwhile, she doesn't want a 10b-5 suit or a derivative suit for false or misleading statements or omissions of material fact that effect the decision to buy or sell stock. It's called securities regulations.

 

[ajc9988]

Ryzen refreshes on "12nm" are quite obvious, but Vega?
I can't quite see any point in throwing even more money at a non-salvageable design.

Actually, it is. Look how Polaris want salvageable. What happened there? Stroke of luck with miners. Now we have Vega 11, which may be 12nm, then Vega 20 which will likely be 12nm, with the first 7nm product on graphics set to be announced or introduced in July or August next year at SIGGRAPH. Look what Asus AIB product for the strix performance looks like compared to the liquid cooled edition. They will weather this fine, especially adding mining features to the main driver.

So before you go off, and I do respect your work on profiles for Asus, think about what is happening. Volta will only have the halo titan, the 1180/2080, and the 1170/2070 with GDDR6 in March. The refresh for AMD keeps them in the game, which launching Polaris without Vega had them out for awhile. That means it isn't a huge chasm, but it also isn't enough for Nvidia to give us the real premium product at a decent price yet, as the 80 series used to be the mid range.

Finally, remember AMD holds 20-30% of the graphics market, but only around 6% on cpu market. You can make gains with a competitive market easier when you are the smaller guy and any sells double your market presence. To gain when you are already double digits is harder. Not to mention bad decisions coming BEFORE Su took the helm which would have bankrupted the company by next year. So, as a business, this is coming along. Did you look at the increase in R&D? The majority of that is 7nm! The reason everyone skipped 10nm was it was a half node that was having major issues. Everyone has made more success at 7nm, whereas Intel is stuck on 10nm. It should be noted that logic density is about the same at 7nm as Intel's 10nm, meaning the market has caught up to Intel because Intel hit a wall. This also is why Intel is abandoning full nodes! Think about it!

 

[krumme]

No, this effects her performance stock units for the year! Meanwhile, she doesn't want a 10b-5 suit or a derivative suit for false or misleading statements or omissions of material fact that effect the decision to buy or sell stock. It's called securities regulations.

And excactly how does that pan out in the real world do you think?
Look at the trading in stock and bonds. Its derivatives like investment for like what 80%??. Its a advanced casino. Amd stock is part of that because its very volatile and risky.
Guidance is used strategically. Its not misleading. Its just taking care. And the words can be optimistic or conservative.

 

[ajc9988]

And excactly how does that pan out in the real world do you think?
Look at the trading in stock and bonds. Its derivatives like investment for like what 80%??. Its a advanced casino. Amd stock is part of that because its very volatile and risky.
Guidance is used strategically. Its not misleading. Its just taking care. And the words can be optimistic or conservative.

Not correcting prior guidance is a lie through omission. You have safe harbors for forward looking statements, but here they needed to revise estimates as that would likely be a material omission, especially in light of beating this quarter's estimate considerably. Context is important.

 

[krumme]

Not correcting prior guidance is a lie through omission. You have safe harbors for forward looking statements, but here they needed to revise estimates as that would likely be a material omission, especially in light of beating this quarter's estimate considerably. Context is important.

Obviously there is only a certain room for maneuverability if you want to avoid jail. Practice just shows its darn big bordering on non existing limitations.
Now i dont take Lisa as a crook. As far from as they come. And as i wrote i think the results is legit. Perhaps they are "sandbagging" a bit though even it its not legal
Now i dont think so and i cant find reason atm they should bend it a bit but hey with mubadala ownership wsa and perhaps some cheaper loan perhaps some creativity would unfold.
But yeaa highly unlikely.

 

[Elixer]

Mark Papermaster has confirmed Ryzen and Vega refreshes on 12LP.

http://www.tomshardware.com/news/amd-ryzen-vega-12nm-lp-2018,35502.html

"We followed up with Papermaster in person and confirmed directly that the company will transition both Vega GPUs and the Ryzen line of processors to the 12nm LP process."

Vega might still have hope if AMD get its complete feature set working with the December Crimson driver release. Then if 12LP can reduce its power draw and bring a die shrink then AMD will be better positioned to atleast compete with GV104. AMD would have done very well if Vega on 12LP can stay close behind GV104. Then with Navi and multi die on TSMC 7nm they can once gain try and compete at the highest end.

I am assuming they must shrink Vega for their APU line since Zen will get the shrink, but, I still don't see it helping with desktop cards much, unless they rework things that are currently disabled (broken?) in Vega.

 

[The Stilt]

Actually, it is. Look how Polaris want salvageable. What happened there? Stroke of luck with miners. Now we have Vega 11, which may be 12nm, then Vega 20 which will likely be 12nm, with the first 7nm product on graphics set to be announced or introduced in July or August next year at SIGGRAPH. Look what Asus AIB product for the strix performance looks like compared to the liquid cooled edition. They will weather this fine, especially adding mining features to the main driver.

So before you go off, and I do respect your work on profiles for Asus, think about what is happening. Volta will only have the halo titan, the 1180/2080, and the 1170/2070 with GDDR6 in March. The refresh for AMD keeps them in the game, which launching Polaris without Vega had them out for awhile. That means it isn't a huge chasm, but it also isn't enough for Nvidia to give us the real premium product at a decent price yet, as the 80 series used to be the mid range.

Finally, remember AMD holds 20-30% of the graphics market, but only around 6% on cpu market. You can make gains with a competitive market easier when you are the smaller guy and any sells double your market presence. To gain when you are already double digits is harder. Not to mention bad decisions coming BEFORE Su took the helm which would have bankrupted the company by next year. So, as a business, this is coming along. Did you look at the increase in R&D? The majority of that is 7nm! The reason everyone skipped 10nm was it was a half node that was having major issues. Everyone has made more success at 7nm, whereas Intel is stuck on 10nm. It should be noted that logic density is about the same at 7nm as Intel's 10nm, meaning the market has caught up to Intel because Intel hit a wall. This also is why Intel is abandoning full nodes! Think about it!

I hear what you're saying, but I do think that Vega was already too far gone at arrival.
On average Pascals (GT1030 - GTX 1080 Ti) are 47.5% ahead in performance per watt at 1080 & 1440P resolutions.
GlobalFoundries advertizes 12nm LP with 10% higher performance at ISO power, compared to 14nm LPP. So regardless if that 10% improvement would materialize in lower power consumption at same performance or in higher performance at the same power, 10% improvement simply isn't enough to make a significant difference. IMO refreshing Vega with 12nm LP process would only make sense if AMD can amortize the cost of doing so on one of their customers. If the new silicon revision on 12nm LP would bring additional significant performance improvements in form of errata fixes and tweaks, then that's a completely another story obviously.

 

[ajc9988]

I hear what you're saying, but I do think that Vega was already too far gone at arrival.
On average Pascals (GT1030 - GTX 1080 Ti) are 47.5% ahead in performance per watt at 1080 & 1440P resolutions.
GlobalFoundries advertizes 12nm LP with 10% higher performance at ISO power, compared to 14nm LPP. So regardless if that 10% improvement would materialize in lower power consumption at same performance or in higher performance at the same power, 10% improvement simply isn't enough to make a significant difference. IMO refreshing Vega with 12nm LP process would only make sense if AMD can amortize the cost of doing so on one of their customers. If the new silicon revision on 12nm LP would bring additional significant performance improvements in form of errata fixes and tweaks, then that's a completely another story obviously.

In part, that may be why Su took out over during the sabbatical, to make sure implementation on key fixes hit schedule and get a better sense of what is happening over on that side. I agree power consumption is an issue, but less so in the past as multi-GPU dies. If you could still do above two cards for sli or games actually implemented mGPU, I think people would care more. Personally, I've got a 1600W EVGA T2 and am mad at the route mGPU support is going. That being said, I could go either which way. But point is definitely taken on an average consumer with a 650-850W power supply.

Although this may change with the multi-die solutions and to see if the Vega 11 is actually more efficient than the big brother.

Also, I think that the cost savings had to outweigh the switch, unless it gave a chance to get it closer to what was needed for the 7nm shrink. We know the process is going to IBM's for 7nm and 12 is supposedly on Samsung's 14nm, so the cost of tweaks, if it were to fix errata, as you suggested, while playing with other tweaks to get ready for what they already know about 7nm, it almost could fold into 7nm R&D in part. But this is obviously spitballing speculation at that point.

 

[raghu78]

I hear what you're saying, but I do think that Vega was already too far gone at arrival.
On average Pascals (GT1030 - GTX 1080 Ti) are 47.5% ahead in performance per watt at 1080 & 1440P resolutions.
GlobalFoundries advertizes 12nm LP with 10% higher performance at ISO power, compared to 14nm LPP. So regardless if that 10% improvement would materialize in lower power consumption at same performance or in higher performance at the same power, 10% improvement simply isn't enough to make a significant difference. IMO refreshing Vega with 12nm LP process would only make sense if AMD can amortize the cost of doing so on one of their customers. If the new silicon revision on 12nm LP would bring additional significant performance improvements in form of errata fixes and tweaks, then that's a completely another story obviously.

imo Vega in its current form is a product with significant bugs. AMD have not even been able to get HBM2 pseudo channel to work on Vega. Vega regresses in raw and effective bandwidth from even Fury X.

https://forums.anandtech.com/thread...on-oc-uc-beyond-3d-suite-ms-dx11-sdk.2510875/
https://www.reddit.com/r/Amd/comments/6mocjw/vega_fe_has_a_serious_regression_in/

The drivers for Vega have not enabled most of the advanced new features like primitive shader.

https://forum.beyond3d.com/threads/amd-vega-hardware-reviews.60246/page-59#post-1997699

AMD has a lot of work to do to get Vega running upto its full potential. Hopefully they can get Vega's advanced features to work properly with their yearly December Crimson driver release. The next refresh of Vega 12LP should fix hardware bugs and get the memory controller working upto its maximum potential efficiency. HBM2 yields should improve allowing AMD to clock HBM2 at 2 Ghz speeds. AMD's goal with Vega 10 on 12LP is to be competitive with GV104. If they fail to do that they will get thrashed and bleed market share. If they succeed to do that then AMD will be able to stay around 30% GPU market share until Navi arrives in 2019 and AMD can try to compete across the entire range upto the highest end GPUs like the Gx100 series.

 

[Abwx]

GlobalFoundries advertizes 12nm LP with 10% higher performance at ISO power, compared to 14nm LPP. So regardless if that 10% improvement would materialize in lower power consumption at same performance or in higher performance at the same power, 10% improvement simply isn't enough to make a significant difference.

20-25% lower power at same frequency...