The AMD Execution Thread [2007 - 2017]

[3dilettante]

The limitation would be the speed of the GMI link, so roughly 100GB/s based on some of the leaded docs a while back.

It takes 4 GMI links over an MCM substrate to give a 100 GB/s number. Unclear from that is whether AMD is giving bidirectional bandwidth for its IO interface. If it's unidirectionally 25 GB/s per link, that's at least a gain from what AMD managed in 2008.

Was there an agreement they had to use GF exclusively? If they can't produce Polaris fast enough to meet demand, meeting wafer requirements shouldn't be a huge concern. Article doesn't mention chip pricing either, so it's possible some of the costs are offset by cheaper production if there were yield/performance issues.

At least historically, when the term "certain other" products get used in a WSA exemption--like the last time AMD paid cash to avoid GF, it is a product that contains x86. Perhaps this was firmed up to include GPUs with Polaris, but the handling of Polaris is almost like its best use now is the pretext to goad this amendment.

I don't know enough about corporate finance to know AMD's Polaris and Zen hype raising the stock price allowed them to set the stock warrant at just under $6, given where it was back when AMD internally probably decided to do this.

This new "better" agreement just shows how GF have AMD by the balls, pretty ridiculous that the original deal was even legal.
The sooner AMD break free from GF's grasp the better.

By original deal, did you mean the prior amended WSA, or the deal where AMD sold its failing fabs at an inflated price to ATIC?
It probably would have helped if the CEO responsible for that deal didn't write a book bragging on how he duped Abu Dhabi into overpaying.
They weren't dumb enough to take on massive investment and money sink and then let their one customer that set a multi-year course that would not be changed without billions of spending to walk away without strings attached. They probably don't feel like doing AMD a solid after all that.

 

[Alexko]

More one-time-charges. This time for $335 million

along with stock delusion

Not good for AMD investors.

http://techreport.com/news/30600/amd-takes-a-335m-one-time-charge-for-more-sourcing-flexibility

I assume you meant stock dilution, but delusion kind of works too.

 

[Kaotik]

It's possible that Greenland is unrelated to Vega and is "Big Polaris" that was rumored to be removed from the desktop roadmap, possibly due to how close it was going to be to a potential Vega launch.

It would make sense to use Polaris over Vega as it's something that is a known quantity now versus having to implement both new CPU cores AND new GPU cores simultaneously.

Regards,
SB

That's not really a possibility, Polaris is GFX IP lvl 8.x, Greenland (just like Vegas) is 9.x

 

[lanek]

[.........................

 

[lanek]

I assume you meant stock dilution, but delusion kind of works too.

Im sometimes realy ask me , if hes not working for an Nvidia investors ... i think have see only him posting in this thread. It dont mean that is not right ... but i think he specially only post in this thread. That dont mean that hes wrong .

Its funny that every time we see something wrong with AMD we get this type of infromation, from him never anything good from him.. absolutely never ...

 

[A1xLLcqAgt0qc2RyMz0y]

Im sometimes realy ask me , if hes not working for an Nvidia investors ... i think have see only him posting in this thread. It dont mean that is not right ... but i think he specially only post in this thread. That dont mean that hes wrong .

Its funny that every time we see something wrong with AMD we get this type of infromation, from him never anything good from him.. absolutely never ...

I am an individual investor and have owned AMD, Intel and Nvidia (and other tech companies) in the past. AMD in the far far past, Nvidia and Intel recently. Right now I have no investment in any of them. I got tired of the lies that previous AMD CEO's told that cost me when I owned AMD. Since the time I exited AMD, AMD has shown that the lies continue (bulldozer), along with their negative net worth, constant losses, billions in debt and now stock dilution they continue to be a company I would not put one nickel in. That doesn't mean you can't make gains (up or down) because the stock moves up with hype and down when reality sets in. But an investment it is not it is more like a casino play, put your money in - you may win but more than likely you will lose.

I see you think I work for Nvidia Investors (I don't). That said I have in the past invested in Nvidia and like any investment I own/follow I do post my thoughts on them.

Unlike AMD Nvidia is a company the exceeds expectations, makes profits, expands markets, is cash positive, pays a dividend and has been a good investment for me. That said I sold all my shares when the stock exploded above $50 and at $62+ dollars I feel it is now way over bought especially with the PE above 50 and the good possibility that Intel will not renew the licensing agreement. (See I can post a negative about Nvidia (stock)).

As for posting here this is the thread titled "amd-execution-gloom" and what AMD has just done (stock dilution) fits.

 

[pTmdfx]

It takes 4 GMI links over an MCM substrate to give a 100 GB/s number. Unclear from that is whether AMD is giving bidirectional bandwidth for its IO interface. If it's unidirectionally 25 GB/s per link, that's at least a gain from what AMD managed in 2008.

It seems it is more likely bidirectional SerDes, since all other buses in the alleged diagram were rated in the same way (DDR4/HBM). This would put it at 6.25 GT/s, assuming the link width is still 16-bit. While it seems low, it is possible that the rate is lowered for the concurrency available (allegedly four links), or perhaps power/latency?

If it is really at 6.25 GT/s, I would at least expect those for inter-CPU, be it on-package or inter socket, to run at a faster rate. Say 8-10 GT/s.

 

[Anarchist4000]

I think 32bit makes more sense here to line up with memory channels. Maybe the link speed varies by distance/latency? Linking components on chip and operating full speed over PCIE simultaneously seems a bit sketchy to me. They mentioned protocols for these links so negotiating a transmission rate and routing around a network MPI-style seems likely.

 

[pTmdfx]

I think 32bit makes more sense here to line up with memory channels. Maybe the link speed varies by distance/latency? Linking components on chip and operating full speed over PCIE simultaneously seems a bit sketchy to me. They mentioned protocols for these links so negotiating a transmission rate and routing around a network MPI-style seems likely.

I don't see what's sketchy. Modern multi-chip links are always running at a decoupled rate from the on-chip interconnect.

Moreover, they are expected to use MCM instead of 2.5D. In this case I would expect high speed SerDes that lowers the stress in I/O density.

 

[Kaotik]

I don't see what's sketchy. Modern multi-chip links are always running at a decoupled rate from the on-chip interconnect.

Moreover, they are expected to use MCM instead of 2.5D. In this case I would expect high speed SerDes that lowers the stress in I/O density.

2.5D is a requirement for HBM

 

[pTmdfx]

2.5D is a requirement for HBM

I was talking in the context of multi-chip cache coherent links, not the memory. They could very well do an MCM between the entire 2.5D GPU stack (with HBM, of course) and the CPUs. Apparently the alleged diagram depicted this precisely.

http://www.fudzilla.com/news/processors/38402-amd-s-coherent-data-fabric-enables-100-gb-s

 

[Anarchist4000]

I don't see what's sketchy. Modern multi-chip links are always running at a decoupled rate from the on-chip interconnect.

The cache coherent part is where I see the problem. Doing a cached read that ended up all the way across a system bus on a GPU for instance.

 

[3dilettante]

It seems it is more likely bidirectional SerDes, since all other buses in the alleged diagram were rated in the same way (DDR4/HBM). This would put it at 6.25 GT/s, assuming the link width is still 16-bit. While it seems low, it is possible that the rate is lowered for the concurrency available (allegedly four links), or perhaps power/latency?

DDR4 and HBM are labelled with their total bandwidth because the DRAM bus does not statically divide its bandwidth between directions, whereas IO interconnects like HT and PCIe do. Historically, AMD marketing has given aggregate bandwidths for its HT links, since the combined number is higher. In this case, it's not clear that a static split is going to fit the GPU's needs. Bandwidth-wise 6.2 GT/s on a x16 link is exactly where AMD was 8 years ago, and an apparent regression from the PCIe links on existing chips.

 

[A1xLLcqAgt0qc2RyMz0y]

Well I thought that AMD stock dilution would be further down the road.

But with AMD stock sky high AMD is tapping the market now.

http://seekingalpha.com/news/320722...email_rt_mc_readmore&uprof=44&dr=1#email_link

Advanced Micro Devices -5% after hours; $600M common stock, $450M convertible senior note offerings announced

AMD (NYSE:AMD) intends to use the proceeds of both offerings to repay up to $226M of borrowings and to purchase up to $1.02B of senior notes.

Good to see AMD reducing their debt some (around $500-$550 million).

Too bad current investors got hosed.

 

[Wynix]

It's better for everyone that they do it though, good move AMD.

 

[Razor1]

Yeah it drops their debt, something they sorely need.

 

[Anarchist4000]

Doesn't appear to have affected investors all that much like the GF deal. ATM after hours are only showing a 20 cent slide which is well within the fluctuations over the past month. Restructuring their debt likely opens up the stock for investments by larger firms as the debt won't be coming due in the next couple years. That's likely to drive prices far higher than any dilution for current investors. It does only make a difference if someone ultimately purchases the stock issued.

 

[Gubbi]

Doesn't appear to have affected investors all that much like the GF deal.

Why would it affect investors? At least in the short term all that's happened is that they increase their cash holding while expanding stock holders' equity by the same amount. The net worth is the same. When they use this cash to reduce debt, the net worth stays the same, - minus the effect a lighter debt load may have in the future.

If Zen sells like hotcakes and AMD profits goes through the roof, old shareholders get a slightly smaller share of the profit. But if AMD didn't do this they'd be bankrupt and investors hosed.

Cheers

 

[Anarchist4000]

Why would it affect investors? At least in the short term all that's happened is that they increase their cash holding while expanding stock holders' equity by the same amount. The net worth is the same. When they use this cash to reduce debt, the net worth stays the same, - minus the effect a lighter debt load may have in the future.

If Zen sells like hotcakes and AMD profits goes through the roof, old shareholders get a slightly smaller share of the profit. But if AMD didn't do this they'd be bankrupt and investors hosed.

Cheers

Specifically because investors would have been buying for earnings, not equity. But like you mentioned, it was unlikely to have an effect IMHO. The GF deal could as they effectively wrote off a bad investment. It's only a positive if you believe the deal was holding AMD back. Prices would fall if investors believed there was no upside. Writing off 100m plus equity for the GF deal and then hoping you can produce products to possibly show a profit isn't all that attractive.

 

[Razor1]

well take another 6% off this morning, its not good for investors, as it dilutes what shares they have, but its a great step for AMD as it cuts out ~20% of their debt.

This is why the stock price and day to day fluctuations don't show anything. This price drop from last night and this morning, is just reactionary, and not really factually based on what AMD is doing.