The AMD Execution Thread [2007 - 2017]
- Thread starter overclocked_enthusiasm
- Start date
-
- Tags
- amd
- Status
Yeah they could really do with Kaveri being a big step forward. This is typical of AMD though, they almost never get all segments performing at the same time and that's why they never have blowout quarters.
This time PC's and normal graphics were poor, next time it'll be servers, after that the consoles won't be great for 6 months etc etc.
This time PC's and normal graphics were poor, next time it'll be servers, after that the consoles won't be great for 6 months etc etc.
Sales should continue to rise until the 3rd or 4th year before dropping off a bit, going on past history. Obviously they are quite seasonal, more so than PC's I believe, so they won't shift an awful lot of units during the first half of next year. A lot depends on China lifting the ban as well.
overclocked_enthusiasm
Regular
Can I come back in to play since I am the OP? I just bought back into AMD last week since selling in 2005, and bought about 10,000 shares. I like the console business, the diversification away from PCs, the exclusivity/potential of the APU and the new Radeon products (although the stock coolers SUCK). Can somebody bring me up to speed on why the APU business is so potentially lucrative and will it migrate from consoles into laptops, lower end PCs and does the APU stand a chance in the high end enthusiast or server market?
It's still a long term plan. They need to have the software in play for it to work and are adding the hardware aspects a piece at a time.
I personally believe Mantle gives them their best chance of breaking out in the short-mid term. Consoles are mostly a known quantity and will pay the bills for the next 5 years.
There is a real worry of AMD going down still though. CPU's got hit hard last quarter and will probably get hit hard again in this one. They need some kind of tablet presence and soon. I feel they've backed the wrong horse in Microsoft, however that might give them some breathing space in classic PC's for a while.
Servers are likely to be all about SeaMicro's Microservers - ARM-based chips coming next year. That could end up being a really good aquisition as classic servers are all but dead.
I personally believe Mantle gives them their best chance of breaking out in the short-mid term. Consoles are mostly a known quantity and will pay the bills for the next 5 years.
There is a real worry of AMD going down still though. CPU's got hit hard last quarter and will probably get hit hard again in this one. They need some kind of tablet presence and soon. I feel they've backed the wrong horse in Microsoft, however that might give them some breathing space in classic PC's for a while.
Servers are likely to be all about SeaMicro's Microservers - ARM-based chips coming next year. That could end up being a really good aquisition as classic servers are all but dead.
Not sure that was such a good idea mate. AMD seems to take massive "one time" charges every quarter that eat up any profits they might have generated. Also, the console win looks great on the surface (and is definitely not a bad thing), but the margins are terrible. In fact, AMD's margins in general are terrible, and that is not looking to improve any time soon.
My prediction is they will be squeezed into almost complete irrelevance by ARM partners on one end and Intel on the other. However for a few reasons they may never die the final death.
My prediction is they will be squeezed into almost complete irrelevance by ARM partners on one end and Intel on the other. However for a few reasons they may never die the final death.
overclocked_enthusiasm
Regular
I am definitely hanging my hat on the APU adoption as I understand all too well the issues in the CPU market with a technology deficit and the challenges of dealing with an ever uber aggressive Nvidia. I am playing more of a technical bounce on the chart up through the 200 dma and then 50 dma...if so a tidy profit. However, I am looking for reasons to think more long term.
What is the general consensus on the following:
1. CPU business - perennial second fiddle to Intel because of Intel fab/R&D power?
2. Server business - same as above?
3. Discrete GPU business - is it neck and neck with Nvidia or is Nvidia considered superior in regards to efficiency and drivers? The 290 seems hot and drivers seem to be meh from the reviews.
4. Management - are they trustworthy and meeting their eps, revenue, and launch targets? Last time I was dealing with Ruiz, Orton, and Richard...the three stooges.
5. ARM - with a license to make ARM products, does AMD stand a chance or will the lion's share of business go to others?
6. Die size - is AMD still behind Intel and ever behind Nvidia? Do they have any fabs left or is it all sub'd out to Charter and others following through on the "asset light" strat from 2006?
Thanks! I am doing my due diligence on the web but would welcome any comments.
1. I don't think AMD can reasonably hope to catch up, but I would expect the gap to narrow somewhat, simply because of the law of diminishing returns. Haswell is a damn good architecture and Intel will have to fight hard for every bit of performance they want to add on top of it. Meanwhile, Piledriver is rather poor and there should be a good bit of relatively low-hanging fruit.
2. The same rationale applies, although AMD has a couple of interesting cards to play. The first one is dense ARM servers, which apparently are getting a good bit of interest from customers, and the second one is HSA enabled APUs, which could prove very interesting to the HPC crowd. However, AMD will have to make sure the software side of things is up to the task, and that's not going to be easy.
3. It's pretty much neck and neck. If you want to really dissect things, GK110 is apparently a little more efficient in perf/W (although I think we might want to revisit that notion when the 780 Ti and custom 290X are out) but Hawaii wins in perf/mm², especially at (very) high definitions, thanks to its 64 ROPs.
But none of this really matters much, because GPUs tend to change quite a bit from generation to generation. In other words, the 20nm generation could really go either way, as could the 14nm one, etc. That said, AMD has two (closely related) good things going for it in the graphics department: GCN-based consoles and Mantle. You might add APUs, which help with market share, even though that's not necessarily a very gaming-oriented part of the market—but then again, this could change with memory stacking in the nearish future.
4. So far, they seem fairly reliable, but it's hard to tell at this point, it's still too soon.
5. Depends on what you mean by "business". AMD has a real shot in the ARM server market, which might well become very large, but it's not clear yet, and for the moment it's quite small.
As for handheld devices, I don't think that's happening. Every company in this business is finding it difficult to compete with Qualcomm, notably because of the latter's integrated modem. As far as I'm aware, AMD has no modem IP whatsoever, let alone integrated. And these days, even in tablets, integrated LTE seems to matter quite a bit. One way they could penetrate this market is with some sort of cross-licensing agreement with Qualcomm (a company that is a founding member of the AMD-led HSA foundation), which would allow AMD to license QC's modem, while letting QC use GCN modules in tablets chips, for examples. But I'm far from convinced that this would be worth it for Qualcomm, as Adreno seems to perform just fine.
6. I'm not sure what you're asking exactly. AMD doesn't have any fabs anymore, they use GlobalFoundries and TSMC. Compared to Intel, that's definitely a disadvantage, but NVIDIA is in the same situation. If anything, AMD has the advantage of being able to use GlolbalFoundries, which NVIDIA is unwilling to do for obvious strategic reasons, and probably a few technical ones too. Then again, AMD has to use GloFo for a given number of wafers per year, which is sometimes a (costly) problem—but probably less so now that contracts have been renegotiated and console chips are ramping up.
It does seem to me that process shrinks are bringing smaller and smaller gains with each generation, so AMD's process handicap might become less important over time.
As for homerdog's point on console chips: yes, the margins are low (mid-teens) but for this type of product, gross margins almost completely translate into operating margins, so that's not a big problem (as evidenced by AMD's return to profitability last quarter with the console ramp). Besides, margins will improve over time, especially when the chips are shrunk to 20nm (early 2015 would be my guess). Further, AMD has mentioned a few more semi-custom SoCs in the pipeline. Valve is a likely candidate for this, as Nintendo or even Apple might be.
All in all, I think AMD has a lot of potential but remains a very risky investment. If I were you, I'd wait for two important pieces of information before making a decision on going long:
a) Just how much performance does Mantle bring? The answer to this should be available within a week.
b) How good is Kaveri? This answer to this should be available within a couple of months, give or take a few weeks. Leaks might happen earlier.
overclocked_enthusiasm
Regular
Great response Alexko!!
I spent 3-4 hours last night researching and think I have a more complete understanding.
1. Even though the 32 nm FX-9590 can get close to the 22 nm i7 in benchmarks it is a full die shrink behind and suffers as a result. While, as you state, die shrink is perhaps less important as Moore's Law wanes it still holds certain advantages. Traditional x86 will likely always favor Intel due to their fab muscle and HUGE R&D budget. Migration from x86 to APUs are a must...
2. ARM does seem to be AMD's best chance in servers. While they can still compete in the x86 space with Intel they face the same fab/R&D issues as I mention above. With the ARM architecture, their server business will come down to how popular Seattle is and if it can scale as well as they think. Migration from x86 to ARM is a must...
3. In the discrete GPU space, it seems like AMD is trying to brute force their way past Nvidia with super hot chips with crappy coolers even though both are at 28nm. Why not a bit more efficient and elegant cooler if you are going to resort to using a brute force approach where excessive heat and cooling will be an issue? Nvidia seems to be more together in the driver space and have far fewer issues especially in regards to frame pacing. I also like how the headroom in Nvidias's GPUs are throttled UP under load where the AMD approach is to start at the MAX headroom and then throttle BACK under load. Seems a bit disingenuous to me on AMD's part and is leading to the legitimate concerns about golden press samples of the 290 family vs the retail boards that are not measuring up. Advantage to Nvidia right now unless AMD gets the heat, cooler, and driver issue sorted out. I also love the idea of Mantle...
4. I am going to read the last 2 CC transcripts and see how they measure up this weekend. I'll report back on any scary items I find.
5. See 2
6. My understanding is they sold all their fabs and they transitioned away from Global Foundries to TSMC. Does TSMC make x86, ARM, GPU, and APU silicon for AMD or do they still use Global Foundries or Charter?
Short term reason to buy/hold
`Mantle (provided the Catalyst team can keep pace)
`Kaveri (specifically with HSA utilized)
`Consoles
Long term reason to buy/hold
`ARM based Seattle chips (provided it ships in 2014)
`APU adoptions (provided the APU makes up for the loss of x86 and the PC slowdown)
I'm in for a few quarters at this point...
I spent 3-4 hours last night researching and think I have a more complete understanding.
1. Even though the 32 nm FX-9590 can get close to the 22 nm i7 in benchmarks it is a full die shrink behind and suffers as a result. While, as you state, die shrink is perhaps less important as Moore's Law wanes it still holds certain advantages. Traditional x86 will likely always favor Intel due to their fab muscle and HUGE R&D budget. Migration from x86 to APUs are a must...
2. ARM does seem to be AMD's best chance in servers. While they can still compete in the x86 space with Intel they face the same fab/R&D issues as I mention above. With the ARM architecture, their server business will come down to how popular Seattle is and if it can scale as well as they think. Migration from x86 to ARM is a must...
3. In the discrete GPU space, it seems like AMD is trying to brute force their way past Nvidia with super hot chips with crappy coolers even though both are at 28nm. Why not a bit more efficient and elegant cooler if you are going to resort to using a brute force approach where excessive heat and cooling will be an issue? Nvidia seems to be more together in the driver space and have far fewer issues especially in regards to frame pacing. I also like how the headroom in Nvidias's GPUs are throttled UP under load where the AMD approach is to start at the MAX headroom and then throttle BACK under load. Seems a bit disingenuous to me on AMD's part and is leading to the legitimate concerns about golden press samples of the 290 family vs the retail boards that are not measuring up. Advantage to Nvidia right now unless AMD gets the heat, cooler, and driver issue sorted out. I also love the idea of Mantle...
4. I am going to read the last 2 CC transcripts and see how they measure up this weekend. I'll report back on any scary items I find.
5. See 2
6. My understanding is they sold all their fabs and they transitioned away from Global Foundries to TSMC. Does TSMC make x86, ARM, GPU, and APU silicon for AMD or do they still use Global Foundries or Charter?
Short term reason to buy/hold
`Mantle (provided the Catalyst team can keep pace)
`Kaveri (specifically with HSA utilized)
`Consoles
Long term reason to buy/hold
`ARM based Seattle chips (provided it ships in 2014)
`APU adoptions (provided the APU makes up for the loss of x86 and the PC slowdown)
I'm in for a few quarters at this point...
Most people would say that AMD has Nvidia on the back foot. Yes they screwed up with the cooler but the 290 and 290X are going to regain market share for sure, and the Mantle effect is a big one.
They are still nowhere near Intel and might never be again. We are all waiting on Kaveri to see if this is it, their final chance against Intel.
Globalfoundries will fab the big cores ie Kaveri and FX including the server versions of each, and that's it. Everything else at TSMC for now.
Everything except the classic PC market looks pretty good, but sadly that's what is paying the wages still.
They are still nowhere near Intel and might never be again. We are all waiting on Kaveri to see if this is it, their final chance against Intel.
Globalfoundries will fab the big cores ie Kaveri and FX including the server versions of each, and that's it. Everything else at TSMC for now.
Everything except the classic PC market looks pretty good, but sadly that's what is paying the wages still.
overclocked_enthusiasm
Regular
Are they still using Globalfoundries because of contractual obligations from the sale of the fabs? It will be curious to see how this plays out at 28nm and Kaveri. Why not use TSMC for your flagship if there has been issues in the past?
My concern with the 290 series is it appears far less elegant and finished than the 780 series from a cooling, driver, and headroom perspective. If it turns out that the press samples of the 290 were as golden as they seem, and the retail versions are all throttling back far below their stated clocks under load, then AMD can be accused of disingenuous shenanigans in order to reclaim the "lead". While there is no doubt the 290 series offers a good performance/price relationship compared to Nvidia, the excessive heat, poor reference cooling, and quality variance showing up in retail boards is troubling.
Hopefully the partner boards with superior cooling solutions will show off Hawaii's true potential with a well thought out cooling solution that will mitigate the throttling back of clocks under load. As long as the Catalyst team solves the frame pacing issues, Mantle takes off, and drivers are stable and mature...I think AMD has a very real shot to grab back some market share from Nvidia.
Can't wait to see Kaveri in action and discussed next week...
The have a wafer agreement with Globalfoundries, something like $1 billion per year they need to buy. It used to be higher but they keep renegotiating a new one. If you google "AMD WSA" you can get the full details.
Even if the WSA didn't exist, TSMC can't offer AMD enough wafers for the big cores - they would need something like an extra 50,000 per month. They are basically stuck with Globalfoundries for the next 10 years or so.
Even if the WSA didn't exist, TSMC can't offer AMD enough wafers for the big cores - they would need something like an extra 50,000 per month. They are basically stuck with Globalfoundries for the next 10 years or so.
Don't forget that AMD has long term debt due in 2015. Right now is the best time to do a refinancing since interest rates are so low, but I'd wait until that issue clears up before picking up any shares. A shame that they settled w/ Intel for a measly billion and some change because they had little choice.
AMD has no room for error unlike Intel. I'm guessing a lot of cancellations / delays like Kaveri and scaling back on socket plans were efforts to make change more incremental and minimize places where major problems could occur to avoid Phenom like fiascos. They lost out on major revenues during the P3-P4 era and contract deals like Macbook air in the past due to supply constraints, so it's logistical things like that which matter a lot more to the bottom line than raw performance and features.
Read has said that he wouldn't be leaning into PC production so even though they'll address this market and possibly yank some business away from Intel, I'd look to ARM + Radeon SoCs and micro servers as their main growth drivers going forward. Mantle + unified memory addressing between CPU and GPU is also very interesting, and I think this is a great way to leverage their console design wins in other spaces as well; I imagine they will enable a unified memory space on ARM SoCs too. The state of mobile SoC drivers is quite miserable:
https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/
Having something common and low level like Mantle on an SoC could be very attractive to devs familiar with consoles and OEMs looking for more control of their graphics stack.
AMD has no room for error unlike Intel. I'm guessing a lot of cancellations / delays like Kaveri and scaling back on socket plans were efforts to make change more incremental and minimize places where major problems could occur to avoid Phenom like fiascos. They lost out on major revenues during the P3-P4 era and contract deals like Macbook air in the past due to supply constraints, so it's logistical things like that which matter a lot more to the bottom line than raw performance and features.
Read has said that he wouldn't be leaning into PC production so even though they'll address this market and possibly yank some business away from Intel, I'd look to ARM + Radeon SoCs and micro servers as their main growth drivers going forward. Mantle + unified memory addressing between CPU and GPU is also very interesting, and I think this is a great way to leverage their console design wins in other spaces as well; I imagine they will enable a unified memory space on ARM SoCs too. The state of mobile SoC drivers is quite miserable:
https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/
Having something common and low level like Mantle on an SoC could be very attractive to devs familiar with consoles and OEMs looking for more control of their graphics stack.
Last edited by a moderator:
The cooler sucks on the 290(X), and why AMD used such a crappy cooler may remain a mystery forever, but apart from that it's a very fine product. Frame pacing is only problematic in Crossfire, and even then, it's a half-solved problem already.
The golden sample thing was just a hypothesis and is apparently due to some fans running at a lower speed, which AMD says they'll fix with a software update. Plus, it only happens in "quiet" mode. Once again, crappy cooler.
By the way, AMD's PowerTune and NVIDIA's Boost actually work the same way, i.e. they both try to maintain the highest speed possible and throttle down when needed, they're just marketed differently. And perhaps NVIDIA's marketing is wiser.
I too am very eager to see what Kaveri can do, but I think next week's presentation is more of a preview than a launch, even a paper launch, so I doubt we'll get hard performance numbers, except perhaps in a few HSA benchmarks.
overclocked_enthusiasm
Regular
This is mystifying to me as well. I would argue that a $50 increase in the board price to facilitate a sexy cooler would have been the best $50 AMD ever spent...considering the critical timing in the company's history and the elegance of the competition. They still could have left a little head room for the partners and 3rd parties to overclock the cards and/or apply even more intense cooling solutions. The technology high ground is so critically important for AMD to regain their lost mind share and eventually market share.
This retail board variance is really starting to trouble me. It is out of the tech press and now into the mainstream press (Forbes). Aren't the boards put through a simple quality control process that would prevent this sort of thing from happening? Are you surmising that it is a driver issue, a firmware glitch, or (God forbid) a bad board issue? Any way you cut it, how can they let these boards out the door with an issue that was so obvious it was picked up on immediately by the press?
My warm and fuzzy feeling about AMD is turning into a rash. A quick software fix on the boards and a strong showing by Kaveri next week is the only ointment that will work for me. I so want this company to succeed...not endure self inflicted wounds.
overclocked_enthusiasm
Regular
Sometimes I play a purely technical bounce of the charts. RSI was under 30...all major supports were broken (200 DMA, 50 DMA) and all the ANALysts hate it. I am a bottom feeder...I buy when there is blood in the streets. However, sometimes I hold for a longer trade like I did with ATYT from $4 to $20 and with AMD back in the day from $10 to $25ish.
Go back to page 1 of this thread and read a few of my posts...I do extensive research once I consider holding for more than a technical bounce or day trade. I am trying to talk myself into holding a bit longer instead of dumping my recent shares ($3.15 - $3.25 cost average).
- Status
