The AMD Execution Thread [2007 - 2017]

[Kaotik]

Random defects aren't related to GPU revision. TrueAudio isn't official part of HD 7790's feature set, so AMD could sell Bonaire GPUs with defective TrueAudio block. Even single- or double-digit numbers of such GPUs on market would prevent AMD from enabling this functionality later.

True, but there's nothing pointing there would be any out there. Of course everything is possible, but I think someone from AMD (Dave?) did hint that it would come to 7790 too via driver update. Not 100% sure obviously though, it was quite some time ago.
Also, AMD has enabled previously features that came with new cards on older hardware, too, if the hardware supports it.

 

[overclocked_enthusiasm]

I think there may be some significant upside surprises to AMD's Q4. Here are the tea leaves I am reading:

1. PC sales decline less than expected per IDC/Gartner. Maybe down 7% instead of down AMD's forecasted down 10%? On a 85 million unit per quarter market, based on AMD's current market share, every point the market doesn't decline is an additional 128,000 units for AMD. Lets take the midpoint and say down 7 vs down 10 or 384,000 extra units. Not sure of asp...maybe $100? So I'll say $38.4 million revenue.

2. Console sales. I think Q4 was modeled on 4-5 million units (Xbox/PS4 combined). The run rate I am seeing is closer to 5-6+ million units. I think there will be an extra 1 million units shipped. On an asp of $80, that is $80m revenue

3. GPU sales? No earthly idea so I will leave it out. If somebody feels comfortable modeling the potential increase from digital currency mining and the effect that will have on clearing the channels, gross margins, and the resultant increase in unit sales...have at it!

Q3 AMD reported $1.46 billion in revenue and net income of $48 million (.06 eps). They called for an increase of 5% revenue for Q4 (+/- 3%) so up 2-8%. I am going to pick the midpoint of 5%

$1.533 billion (up 5%)
OpEx flat at $450 million
CapEx down sequentially but I will model flat
GM essentially flat at 35%

Here is where it gets sketchy...its tough to assume that they can flow through 100% of GMs from the unexpected incremental revenue to net income...so I will come up with a few modifications to account for the extra volume of products and the man hours etc. to handle them.

1. Let's assume they will earn $48 million on the first $1.46 billion like last quarter since OpEx and GMs are essentially flat with CapEx actually perhaps declining.

2. Let's assume the extra incremental revenue I mention is on top of the +5% I model, so we get $1.65 billion for TOTAL Q4.

3. Here is the kicker...how do we model this extra $190 million in revenues since AMD has said flat OpEx, flat or down CapEx, semi-custom royalties declining as model switches to sales revenue/net income based, and GMs down 1.

$48 million net income on $1.46 billion revenue (using Q3 mix)
$12 million net income on $80 million (incremental semi-custom sales 15% flow through to net)
$9.6 million net income on $38.4 million (incremental PC sales 25% flow through to net)
$18.25 million on $73 million in revenue (up 5% per guidance and shot in the dark 25% flow through to net)

I come up with $1.65 billion and net income of $87.85 million or .11-.12 cents eps on 764 million diluted shares. This assumes a high flow through rate on the incremental revenue if they can stick to their forecast of flat OpEx, flat CapEx, flat inventory, and essentially flat GMs.

HIGH estimate is $1.58 billion and .08 cents...now throw in these weird reports on AMD GPUs flying off the shelves for mining digital currency.

I think my $1.65 billion and .11-.12 is conservative...feel free to shoot holes in my math.

 

[Alexko]

Financial wizardry is not one of my skills, but it seems to me that your $100 ASP estimate is very optimistic. Most of AMD's sales will be made up of low-end/mainstream APUs, including a whole lot of very cheap Kabini/Temash stuff. I have no idea what the real figure is (there may be some official data or IDC/Gartner estimate somewhere) but something like $60~80 wouldn't surprise me.

 

[eastmen]

still wish they had something in the high end for us.

 

[overclocked_enthusiasm]

Ok, this digital currency mining thing has me wondering. At Newegg currently:

290 5/7 sold out
290x 9/10 sold out
280x 12/15 sold out
270x 9/22 sold out
270 5/7 sold out

Something has to be going on here. The historical GPU segment sales at AMD are in the $350 million range before royalties for Xbox One/PS4. Xbox 360 sold about 12 million units per year so lets say 3 million per quarter at $40 asp? I am trying to model discrete desktop GPU sales ONLY.

QUARTERLY (Prior Xbox One & PS4)
Total Revenue $350 million
Console Sales $48 million (Xbox 360)
Professional Graphics $32 million (10% of GPU sales???)
PC Graphics $270 million (90% of GPU sales???)

So the BIG question, assuming my numbers are in the ballpark, what % of that $270 million PC graphics business is discrete desktop? Anybody have any ideas...I can't find anything. AMD has 35% of the Desktop discrete market but I can't find what I need...total units. I'll assume a few things here:

DISCRETE DESKTOP
PC Graphics total $270 million (Professional, integrated, discrete desktop, discrete mobile)
Discrete Desktop Only $135 million (assuming 50% total revs due to high ASPs on discrete desktop parts)
GMs 40% or higher

I am going to model a $40 million incremental sales in discrete desktop GPU sales for Q4 due to digital currency mining. So to adjust my Q4 model:

$48 million net income on $1.46 billion revenue (using Q3 mix)
$12 million net income on $80 million revenue (1M incremental semi-custom sales 15% flow through to net)
$6.72 million net income on $26.88 million revenue (3% incremental PC sales 25% flow through to net) Lowered ASP to $70
$18.25 million net income on $73 million in revenue (up 5% per guidance and shot in the dark 25% flow through to net)
$10 million net income on $40 million in revenue (digital currency increase in GPU sales at conservative 25% flow through to net)

So my guess for Q4 is $1.68 billion and $95 million net income or .12-.13 eps...way higher than consensus ($1.54 and .06). If OpEx increases my eps is too high...period.

My BIG assumption is that digital currency mining, +1m console chips, and a 3% point improvement in PC market were NOT in their "up 5% +/-3%" guidance for Q4. I also assume OpEx is flat at $450 million per AMD's guidance.

Please feel free to correct my math, correct my assumptions on ASPs, or correct my assumptions to isolate discrete desktop revenue/margins. Some big rumblings about "customized" dense ARM server possibilities from the big boys of the world...Google, Facebook etc.. I wonder what company could handle both x86, ARM-64, and brings years and years of overall server experience? Hmmmm

 

[Alexko]

The GPU in the Xbox 360 is not discrete, it's integrated into a chip that is manufactured directly for Microsoft. AMD only gets licensing revenue from this, and I would be extremely surprised if it were anywhere near $40. My guess would be somewhere around $10, and quite possibly lower than that.

As for desktop units, this might help: http://jonpeddie.com/publications/add-in-board-report/

In Q3 2013, 35.7% of 14.5 million units, i.e. ~5.44 million units.

 

[overclocked_enthusiasm]

The GPU in the Xbox 360 is not discrete, it's integrated into a chip that is manufactured directly for Microsoft. AMD only gets licensing revenue from this, and I would be extremely surprised if it were anywhere near $40. My guess would be somewhere around $10, and quite possibly lower than that.

As for desktop units, this might help: http://jonpeddie.com/publications/add-in-board-report/

In Q3 2013, 35.7% of 14.5 million units, i.e. ~5.44 million units.

Lol...I read that Peddie report twice this morning and missed it. The data is just too co-mingled to come up with a reasonable base for just discrete desktop...and it would be a total guess on this digital currency mining impact anyways. I think my eps estimates on the various sources of incremental revenue may be rosy on further review. I'll stick to my $1.65 billion and hedge my eps a bit to .10-.13...still both well above consensus.

I think the 2014 WSA is due anytime which may shed some additional light on how things stand in certain areas. Quote from Q3 CC 10/17/13:

"we are in discussion to figure out the pricing and the wafer volumes for 2014, and I expect those to close within the next 30 to 60 days.

 

[Wynix]

Lol...I read that Peddie report twice this morning and missed it. The data is just too co-mingled to come up with a reasonable base for just discrete desktop...and it would be a total guess on this digital currency mining impact anyways. I think my eps estimates on the various sources of incremental revenue may be rosy on further review. I'll stick to my $1.65 billion and hedge my eps a bit to .10-.13...still both well above consensus.

I think the 2014 WSA is due anytime which may shed some additional light on how things stand in certain areas. Quote from Q3 CC 10/17/13:

"we are in discussion to figure out the pricing and the wafer volumes for 2014, and I expect those to close within the next 30 to 60 days.

I think that you are correct, this is going to be a strong quarter for AMD; People are way too focused on their desktop x86 parts.

When will we get Q4 data btw?

 

[gkar1]

AMD and Hynix announce joint development of HBM memory stacks

AMD and Hynix announce joint development of HBM memory stacks
http://electroiq.com/blog/2013/12/amd-and-hynix-announce-joint-development-of-hbm-memory-stacks/

3DIC memory, and therefore all of 2.5/3D technology, took one step closer to full commercialization last week with the HBM joint development announcement from AMD and Hynix at the RTI 3D ASIP meeting in Burlingame CA.

Already sampling and AMD says it's ready to work with customers. First implementation looks like it'll be for GPUs then networking and HPC.


more: https://www.semiwiki.com/forum/content/3003-amd-goes-3d.html

 

[Deleted member 13524]

It seems obvious to me that APUs would be the chips that would benefit the most out of this, where they need to dramatically increase memory bandwidth ASAP.

 

[Alexko]

They mention GPUs first, which makes sense to me. My (renewed) prediction is that we'll see some sort of HBM first with 20nm discrete GPUs in 2014, then with Carrizo and the successor to Beema/Mullins in early 2015.

 

[Deleted member 13524]

They mention GPUs first, which makes sense to me.

Why? Do you think memory bandwidth is a major concern for today's GPUs?

It seems to me that 7GT/s GDDR5 should be able to cover most of the price budgets in discrete GPUs for a while..

Due to the new consoles, I think memory amount in higher performance GPUs will take a much more needed spin than memory bandwidth.

 

[3dilettante]

Why? Do you think memory bandwidth is a major concern for today's GPUs?

The GPUs could readily take advantage of the bandwidth, and unlike APUs they have a presence in a premium market that might be willing to pay the extra cost as they work the kinks out. The volume demands there are also closer to what HBM is likely to sustain early on.
Another concern is the power consumption of the memory controllers and memory interface.
At the same bandwidth levels, a GPU like Hawaii might be able to shave off a fair amount of power consumption. If it were relative to a chip like Tahiti, the benefit would probably be in the tens of watts, but I haven't done the math for the wider and slower Hawaii bus.

If 20nm does get a high-end GPU before the hop to the hybrid nodes, the lackluster power scaling is going to put a damper on their ability to make it significantly better than the 28nm ones, absent a more significant architectural shift. GCN's power consumption relative to its competitors might hint that rearchitecting a few things wouldn't hurt, but still.
Even if the next big architecture is on a FinFET process, getting the extra power would still be a good thing.

 

[Alexko]

If I'm not mistaken, HBM would also allow for die size savings on the GPU itself.

More to the point, Hawaii's maximum power consumption in the 290X is 206W; the whole card can draw up to 300W. Now I doubt memory accounts for the entire 94W difference, but then again fast GDDR5 also increases the power consumption of the memory interface, as you've pointed out, so there clearly appears to be a very sizable chunk of the card's power budget dedicated to memory.

As for the premium added by HBM, I have no idea how much it actually is, so I don't know how viable it could be in APUs in the near term.


In any case, 7 GT/s GDDR5 is not viable for next-generation GPUs. Again, taking Hawaii as an example, it would roughly double the size of its PHYs, thereby making the die quite large, and the power cost would be huge. AMD and NVIDIA will be forced to do that if they can't get HBM to work (within acceptable parameters) soon enough, but it's a crappy solution.

 

[overclocked_enthusiasm]

http://wccftech.com/amd-athlon-sempron-revived-quad-core-25w-apu-desktops/

Is this noteworthy? It seems more marketing based using the Athlon/Sempron brands than a surprise in "new" products or architecture.

 

[eastmen]

Ah they killed the fx name now it's time to kill the rest of the brands

 

[swaaye]

I've always wondered why Duron was flushed to make way for that goofy Sempron name....

 

[Alexko]

I've always wondered why Duron was flushed to make way for that goofy Sempron name....

Don't try to understand the logic behind a marketing department's decision, it can be very hazardous to your mental health.

That said, Athlon certainly didn't die in 2005, or at least not permanently:
http://products.amd.com/en-us/DesktopCPUResult.aspx

And in the wild:
http://www.newegg.com/Product/Product.aspx?Item=N82E16819113328

 

[overclocked_enthusiasm]

I just got one of those Asus 270x with the improved DirectCU II fan/heatsink for my son for Xmas...that thing is gigantic! Gonna pair it with a i5 4670k on a Asus Z87 Pro mobo with 16g of PC 2400 ram....should be pretty sexy. Nice to see what the partners can do vs the crappy reference cooler.

http://hosted.verticalresponse.com/1025131/bdf3231b3c/522028065/82abc47f8c/

Coupled with the data from Gartner & IDC, I am expecting an upside surprise in AMD's core PC business at least in Q4 and maybe even improve outlook for 2014. Has anyone heard anything about the 2014 WSA? According to the CFO it should have closed between Nov 16 and December 16. There was some loose talk (on the stock boards) about the $200 million fine due in 1Q-2014 from the take or pay provision in the 2012 WSA being rolled into the 2014 WSA and somehow negating the payment from AMD. I find this to be dubious at best...any thoughts?

 

[Kaotik]

Currently it looks like MSI's Twin Frozr IV is "The Cooler" to beat, even when the new and improved DCII is in the mix