The AMD Execution Thread [2007 - 2017]

[Silent_Buddha]

I can't verify that statement about most cards shipped/sold since q4 of 2014 unless I mix and match JPR and Mercury numbers.

Outside of buying Mercury's report there's no way to verify the numbers. And considering that some of AMD's shareholders (investment firms, banks, etc.) have access to the Mercury reports, you can expect lawsuits if AMD are lying about this.

Regards,
SB

 

[entity279]

We've had a lawsuit about nv's last 512 MB of RAM on the GTX 970, so yeah..

 

[Razor1]

That is a bit different that was something that was concrete, this is much more ephemeral

It would be very hard to sue something like this unless next Q's numbers don't align with what we see these numbers at, for one thing, if nV's 20% shipment drop didn't happen or is restored to what it was before....

 

[firstminion]

here are the share values for 2015 q1 and back.

http://www.mercuryresearch.com/graphics-pr-2015-q2.pdf

AMD is talking out of their ass.

These numbers are arbitrary numbers.

If total shipped of GPU were 10 million, and ya had 28% market share then you have 2.80 million units right?

If now you have 12 million total shipped and 23% market share 2.760 million units?

Get the picture, this isn't margin of error.

Q2 of last quarter was worse than Q2 of this quarter.

The Paradoxically statement, is as a whole discrete market. And from the info you have that you are privy to Q1 of 2015 it wasn't weak not as weak as this recent quarter.

For crying out loud! what 10M? What 12M? Where the hell did you get these numbers?

Dr. Lisa Su isn't the one being arbitrary here. Enough.

 

[Razor1]

MY numbers were arbitrary, there were no numbers to speak of from AMD's side, just showing how the %'s work out when you have different amount of total volume sales, btw, I'm saying this because I'm looking at certain figures.

It is mathematically impossible to have an increase in total shipments with the 2.5% increase in marketshare for discrete factoring into nV's decline in their own share and an overall decline in the desktop discrete shipment units unless what I stated happened. 2.5% is not a large number of units. When you consider the factor of what nV lost on their own side of things and overall discrete amounts.

Eyes raise about these things because remember r290 sold great, miners loved these cards that was why they are hard to get, low stock, high prices, then the marketshare numbers come out, its like yeah that was lie.... NO ONE can sue them for saying such things, because they can turn around and say hey demand dropped which we weren't expecting or what ever.

So many ways to dress pig and keep things, all intangible, but when the shit hits the fan, it will all come out.

Yeah I would poke AMD stock with a stick thats about it.

What AMD has effectively stated, is that the desktop discrete in channel counting all the changes in the over all marketshare numbers (decreases from both sides in the various segments) they gained like 9 points, that isn't what I see at all.

 

[Silent_Buddha]

MY numbers were arbitrary, there were no numbers to speak of from AMD's side, just showing how the %'s work out when you have different amount of total volume sales, btw, I'm saying this because I'm looking at certain figures.

It is mathematically impossible to have an increase in total shipments with the 2.5% increase in marketshare for discrete factoring into nV's decline in their own share and an overall decline in the desktop discrete shipment units unless what I stated happened. 2.5% is not a large number of units. When you consider the factor of what nV lost on their own side of things and overall discrete amounts.

Eyes raise about these things because remember r290 sold great, miners loved these cards that was why they are hard to get, low stock, high prices, then the marketshare numbers come out, its like yeah that was lie.... NO ONE can sue them for saying such things, because they can turn around and say hey demand dropped which we weren't expecting or what ever.

So many ways to dress pig and keep things, all intangible, but when the shit hits the fan, it will all come out.

Yeah I would poke AMD stock with a stick thats about it.

What AMD has effectively stated, is that the desktop discrete in channel counting all the changes in the over all marketshare numbers (decreases from both sides in the various segments) they gained like 9 points, that isn't what I see at all.

Discrete includes mobile and desktop discrete GPUs.

Without knowing the total number of both discrete desktop and discrete mobile numbers there is no way of knowing. It's entirely possible to lose share in one while gaining share in the other. And depending on the numbers have a percentage increase in shipments that is greater than the percentage increase in market share. Heck it's also entirely possible to lose market share in both yet have a percentage increase in shipments.

You are entirely far too focused on the market share numbers when it's the total shipped numbers that is important to this discussion. And you cannot derive shipped numbers from market share numbers without first knowing shipped numbers.

When looking at Mercury reports this is further compounded by the fact that the numbers released to the public aren't the actual numbers. They are based on a 4 quarter volume-weighted average.

Mercury Research is reporting the shares as a four-quarter volume-weighted average to smooth the noise of seasonal inventory cycles and reveal ongoing share trends.

So the market share number in their report isn't even the 2Q 2016 market share. If it wasn't already impossible to derive shipment numbers with what I outlined above, this guarantees that there is no way to do so.

To make that clear. Without purchasing the report you cannot

1. Know the actual market share for any given quarter.
2. Know the shipped numbers for any given quarter.

What we can infer is that actual Q2 market share for AMD discrete desktop was higher than 22.8%. And it had to be significantly higher to outweigh the previous 3 quarters.

That also means that Nvidia's actual shipped market share was significantly lower than 77.2% as it again, has to overcome the volume weighting of the previous 3 quarters. Which makes sense since the report noted that Nvidia's shipments for discrete desktop dropped by 20%.

Why do they do it that way? Well, to provide information while also making it just vague enough that you are encouraged to buy the report if you want any specific information of any use. Plus as they said, it's meant to show trends without the numbers being skewed by seasonal sales. As such, it's useful to see trends, but mostly worthless otherwise. All truly useful numbers are basically behind a paywall.

Regards,
SB

 

[Razor1]

Yes Silent Bhudda, I did work for Credit Suisse, a few years ago, I still have access to some (old) paid reports, (no one will show me the latest one but that was to be expected but I did get total shipments from JPR and Mercury for the quarter) from both of them, fair enough? (that's why I know they needed 9 points to do what they stated, I don't buy it. There is no correlation based on causation these reports are never ment for that, even the full reports all these reports show are trends and that is it.

And no, you don't know whats in the full reports, I do. And what you just stated is not true at all in this case. I'm looking for the trend to fit with what they stated, I can't find it in a single report.

 

[Silent_Buddha]

In which case expect to see news from Mercury, Nvidia, or other major players about AMD fabricating information that isn't contained in the report they cited as it would be childishly easy to prove them wrong.

Especially from Mercury as that calls into question the accuracy of their reports. And NVidia has never been shy about calling them out when their numbers are incorrect, especially when it impacts their own numbers (IE - the reported AMD number means the NVidia numbers are significantly lower).

If we don't see anything, then we can assume that what AMD cited is correct.

I am rather shocked, however, that Mercury's reports don't contain actual shipment data for the quarter as that is required to derive their market share trends. Which numbers also make it childishly easy to calculate actual market share per quarter.

Regards,
SB

 

[Wynix]

When are we going to see AMD make a genuine push for the gaming laptop market? HBM2? ZEN? it's growing every year and AMD is nowhere to be found, is there some issues using an Intel CPU and AMD GPU?

 

[Deleted member 13524]

When are we going to see AMD make a genuine push for the gaming laptop market? HBM2? ZEN? it's growing every year and AMD is nowhere to be found, is there some issues using an Intel CPU and AMD GPU?

The gaming laptop market is growing every year? Are you sure of this?
AFAIK, laptop sales in general are falling really fast because they're being replaced by tablet/laptop hybrids like 2-in-1s and foldables.
Gaming laptops are a rather small niche and the competition seems to be really fierce in there.

Regardless, Zen APUs are coming somewhere in the middle of 2017. There's this HPC beast that's appeared on leaked slides over a year ago, but this is definitely not a laptop chip:

However, an APU with half of everything in here (8 Zen cores + half Greenland + 1 stack 8GB HBM2 + dual-channel DDR4) would be a very decent single chip for a very high-performance gaming laptop.
This of course, assuming Greenland is similar to a Vega chip that will replace Fiji (64 - 96 CUs).

 

[Razor1]

hmm I don't think any version of greenland will come to a laptop, power using, cost, vs performance compared to what a dgpu can offer, it wouldn't make sense.

 

[function]

The gaming laptop market is growing every year? Are you sure of this?
AFAIK, laptop sales in general are falling really fast because they're being replaced by tablet/laptop hybrids like 2-in-1s and foldables.
Gaming laptops are a rather small niche and the competition seems to be really fierce in there.

Regardless, Zen APUs are coming somewhere in the middle of 2017. There's this HPC beast that's appeared on leaked slides over a year ago, but this is definitely not a laptop chip:

However, an APU with half of everything in here (8 Zen cores + half Greenland + 1 stack 8GB HBM2 + dual-channel DDR4) would be a very decent single chip for a very high-performance gaming laptop.
This of course, assuming Greenland is similar to a Vega chip that will replace Fiji (64 - 96 CUs).

I'd like to see that. Even 4GB of HBM would be a good start on a system that wouldn't need to use drivers to swap texture data over PCI-E. If the HPC APUs could use the HBM as a last level cache then everything would just fly, with fewer headaches than for a dGPU.

hmm I don't think any version of greenland will come to a laptop, power using, cost, vs performance compared to what a dgpu can offer, it wouldn't make sense.

Power should be less than for a similar performing CPU and dGPU from AMD. Board area would be much less and the complexity of cooling and power delivery reduced. You could make smaller systems with the same kick.

 

[Deleted member 13524]

Power should be less than for a similar performing CPU and dGPU from AMD. Board area would be much less and the complexity of cooling and power delivery reduced. You could make smaller systems with the same kick.

Yup.
A perfect gamer APU would be a ~3.2GHz 8-core/16-thread Zen, dual-channel DDR4, 36-40 CU Vega at 1GHz and single stack of 4GB HBM2. Get that into a Mini-ITX board and there would be loads of SFF PCs using it.
Then the laptop version would have lower clocks, like 2.4GHz CPU and 800MHz GPU.

 

[Razor1]

I'd like to see that. Even 4GB of HBM would be a good start on a system that wouldn't need to use drivers to swap texture data over PCI-E. If the HPC APUs could use the HBM as a last level cache then everything would just fly, with fewer headaches than for a dGPU.



Power should be less than for a similar performing CPU and dGPU from AMD. Board area would be much less and the complexity of cooling and power delivery reduced. You could make smaller systems with the same kick.

Not if you want performance out it at a reasonable price. just doesn't work that way, you need both for a lower end and midrange laptop, now if you want to go higher end, cut down Greenland just doesn't fit the bill does it?

Either its going to be more expensive for a midrange to lower end part, or the power savings at the high end just doesn't justify the performance you get out of it over a dGPU.

Hence why HBM hasn't go down the midrange desktop cards, you think that is going to change in a product that is more price sensitive? Yes laptops are more price senstive than discrete cards, specially low end and mid range gaming notebooks. Every large oem has them and its all going to come down to what I stated.

HBM only makes sense at the enthusiast level tier (for now) where the margins are high enough to cover the cost of HBM.

 

[function]

Not if you want performance out it at a reasonable price. just doesn't work that way, you need both for a lower end and midrange laptop, now if you want to go higher end, cut down Greenland just doesn't fit the bill does it?

Either its going to be more expensive for a midrange to lower end part, or the power savings at the high end just doesn't justify the performance you get out of it over a dGPU.

Hence why HBM hasn't go down the midrange desktop cards, you think that is going to change in a product that is more price sensitive? Yes laptops are more price senstive than discrete cards, specially low end and mid range gaming notebooks. Every large oem has them and its all going to come down to what I stated.

HBM only makes sense at the enthusiast level tier (for now) where the margins are high enough to cover the cost of HBM.

The cost that can be supported depends on how competitive the parts are. High end products from Nvidia (for both desktop and laptop) command huge margins despite being completely impractical for low/mid rage products.

Laptops also place interesting demands on size and power that bring additional concerns - and opportunities - to bear. High end gaming laptops and small form factor gaming PCs would be an interesting market for AMD to address.

 

[Razor1]

I can see it work on a high end gaming laptops, they go for 3k and up, sli versions, dont' think there are any xfire versions since 2 gens ago but they were priced around that too I think.

 

[CarstenS]

The cost that can be supported depends on how competitive the parts are. High end products from Nvidia (for both desktop and laptop) command huge margins despite being completely impractical for low/mid rage products.

And that's exactly the problem: With HBM you're taking a great risk, because it does not leave you much leeway in terms of price cuts should the competition have a better or comparative but cheaper product. And you probably throw away quite a bit of money when designing two sets of memory controllers into one and the same chip.

So not going the HBM route at this point is basically the ultimate risk management.

 

[Deleted member 13524]

Performance APUs with HBM would be on >1200€ laptops, obviously.
The much smaller PCB would give way to a larger cooling system and/or larger battery.

As for SFFs, a company like ASRock could make a gamecube-sized gaming machine using an external power brick, SO-DIMMs and a M.2 SSD.

 

[function]

And that's exactly the problem: With HBM you're taking a great risk, because it does not leave you much leeway in terms of price cuts should the competition have a better or comparative but cheaper product. And you probably throw away quite a bit of money when designing two sets of memory controllers into one and the same chip.

So not going the HBM route at this point is basically the ultimate risk management.

I'm assuming that AMD will already have to design two sets of memory controllers into the same chip for the HPC version, and solve any issues relating to CPU <-> GPU communication. Solutions should already have been produced and be working blocks on whichever process they're using for the HPC version. We're not talking about solving any new problems here, or even necessarily adding any new features.

While it's true that there may be additional risk associated with HBM2, it's not like there isn't a risk with a dGPU once you lock into a certain die and memory configuration. The issue is the same IMO - having a competitive product that can command the margins that are worth the investment. There's also a risk to not competing with Intel and Nvidia in the areas where there are actually profits to be made.

 

[3dilettante]

If the more recent leaked slides about AMD's HPC APU are legitimate, AMD's solution to implementing two sets of memory controllers on the same chip is to not do so by using an MCM hosting separate CPU and GPU silicon.