The AMD Execution Thread [2020]

A 25-minute video can't possibly be exhaustive about that topic.

 

Variable Function Units here we come ...

 

The only way this would make any sense is if AMD pays entirely in stock. They simply should not be spending any cash. They've just recently become debt-free or close to it IIRC.

 

There are reasons why for big companies it may be better to have debt than not to have debt IIRC (something about taxes at least)

 

I mean Xilinx is the market leader and profitable on its own. Surely buying them makes sense anyway.

That said, I do not see any obvious synergies between the two companies. Maybe for semi custom system designs? And patents, of course

 

Yes, the interest on debt is subtracted from earnings for the purposes of tax calculation. So some amount of debt is generally good (also instills a certain fiscal discipline at times). However the last time AMD made a major acquisition, namely ATI, they amount of debt they took on was unsustainable. They resorted to selling certain divisions and even sale and leaseback of their campus to raise cash at that point. I'm sure none of us want to see them do something like that again.

Yes but at a pricetag of $30B, share price near the 52 week high and a P/E of 46, it is not an attractive acquisition purely from a financial point of view. Admittedly semiconductor companies in general have a high P/E, Nvidia at 101 and AMD at an even higher 161, but still, I fail to see major synergies as you say.

 

I thought they had accumulated enough cash to readily pay for a large fraction of their debt that was coming due, but there was still an amount of similar magnitude due at various times in the future. It's been a while since I've looked at the financial presentations, though.

AMD's concepts for mix-and-match IP blocks and chiplet integration have often included a reference to programmable logic IP alongside the CPU and GPU/APD blocks. Various patents also reference the possibility, and at times AMD has provided concepts or patents for using FPGA hardware for various purposes. The counterargument to that recurring theme was that they didn't have that IP, so there was something of an aspirational aspect to those disclosures.

AMD's value-add argument for semi-custom or product growth does depend on the availability of desirable IP, and it has some constraints that make it more difficult to address markets besides the ones it already serves. For HPC, AMD's vision gives a chiplet/interposer topology, and only CPU and GPU IP to populate the slots.
Other elements, like system interconnect or additional types of hardware are elements that AMD often leaves to third parties. For interconnects, I recall Intel purchasing at least part of Cray's interconnect technology, and Nvidia has bought Mellanox. The tech for bringing together server nodes is something AMD's larger competitors are bringing in-house and potentially reducing AMD's value-add for servers and HPC.
Then there's NVidia's attempt at buying ARM, which among other things means AMD's security and system integration layer that it touts as a server value-add could be dependent on a competitor's good will as well.

Having some value-add and IP leverage against ever-expanding competitors is a strategic play, and there's the possibility that Intel's Altera setbacks might someday get past the acquisition blunders and 10nm false starts. AMD's positioning is marginal enough that even if Intel's competing heterogenous initiatives aren't massively successful for Intel, they could still dwarf AMD. AMD's integration play is weakening as its stable of partners is moving into the IP portfolio of the likes of Intel or in-house for companies that would have been customers in times past.

Having a lot of cash on hand, or not having a significant amount of debt can also raise the chance of a leveraged buyout that could then use AMD's cash or unleveraged assets to pay for the buyout.
I recall AMD's offering for ATI was all or mostly cash, and I recall discussion about how well ATI likely played AMD in that regard, given that AMD overextended itself and R600 was the next product out.

Given AMD's stock price and the likelihood of some kind of correction, this might be one of the best times to try for such a purchase. The selective nature of which economic sectors have collapsed has moved money into tech and enabled things like Nvidia's attempt at grabbing ARM (Nvidia's stock price and Softbank's motivation to sell such an asset).

 

Must read for those unaware of Xilinx & the FPGA market:
Pondering That Rumored $30 Billion AMD Acquisition Of Xilinx (nextplatform.com)

 

I haven't looked at them in detail recently either but I vaguely remember that they were almost debt free and now generating sufficient free cash. The Q3 results should be out by the end of the month, will delve into it then.

I feel AMD could address any of those shortcomings with partnerships/licensing, the likes of which they have with Cray for Supercomputers for example. AMD have the key IPs, i.e. GPU and CPU, unlike Nvidia, who's only option to get an entry to the market is to purchase ARM. AMD still does not invest in R&D to the scale of Intel or Nvidia and are probably better off focusing their resources towards the key GPU and CPU IPs. The addressable market is big enough without them having to look to other segments.

True, though LBOs in the semiconductor are a lot more complex due to the nature of the industry demanding significant R&D spend. Aside from Broadcom and Avago, I can't recall many other big ones.

Yes the timing of the ATI purchase was just unfortunate for AMD. They'd have probably paid a third if they'd waited a year. Oh how things have might been if they hadn't completely hamstrung themselves for a few years thanks to that.

Quite possibly, as I mentioned AMD's PE is 161 at the moment so just like Nvidia, it's definitely a good time to be paying in stock.

Good read. Thanks!

 

They've been around 500 million debt mark for some time now, with 1775 million cash / cash equivalent at the end of q2

 

AMD has opted for that model, although it's obligatory rather than optional like it is for the more vertically integrated competition.
This can be more efficient in terms of making certain IP choices possible without taking AMD out of its core competencies and burning limited resources. The question for AMD's value-add is how reliably it can coordinate a competitive bundle of IPs while also projecting certainty to current and future customers that they will be able to provide a consistent mix or roadmap going forward. A perennial threat for putting together a mixed product or promising continuity into future generations is the question of "what if your partner is acquired?" in an increasingly consolidated industry. Coordination and logistics are also a possible source of inefficiency, since even within AMD there is a chronic lag between GPU and CPU IP for its APUs and multiple instance of missing critical shopping windows.

This would be part of a buy-in question as well for any partners into whatever scheme AMD wants to use. AMD's hinted at a vision of extending its dis-integration strategy into creating chiplets that can serve as application specific standard products, which would need a physical standard and interoperability with partners and potential customers. If there is no buy-in, the 3D stacks of AMD and hybrid IP would have AMD chiplets and all the other slots painfully vacant.
HSA would be an example of this, and its what happened to it and the partners within it seem illustrative of the threat of separate parties' interests diverging or individual members failing/being acquired.
That HSA found significantly more activity in its dwindling years in China highlights a problem given the rise in political brinksmanship over IP. Partners could be spontaneously walled off by a trade dispute from AMD at any time.

Now that Nvidia is attempting to purchase ARM, it could be buying into an increasingly credible alternative to x86 in some fields, and a stronghold in other markets that AMD would be even less able to get into.
AMD also has a certain amount of its IP dependent on ARM cores or ARM platform elements like TrustZone, and it might be worth keeping an eye on what licensing or legal games could be played in the future.
Even if the ARM acquisition winds up hurting ARM and Nvidia, it may do so by incentivizing something like RISCV, particularly with markets or customers that have limited ties to AMD products already. The aforementioned trade wars could mean looking for something in-house, in-country, or otherwise separate from an AMD product that could be blocked by one government or the other.

AMD has stated its desire to expand, and its addressable market may have increasing competition. If AMD only optimizes what it already serves, the extent of it that is sufficiently served just by CPU and GPU IP is not a certain growth prospect.

Apple's apparent direction of using in-house CPU and GPU IP for all but perhaps professional products could mark the end of Apple's investment in tweaked version of AMD's GPUs, which have been a consistent part of AMD's GPU portfolio for years. I cannot predict what Apple's ambitions are for the highest tier, but it's a diminished prospect for AMD in terms of graphics or any possible chance for its CPUs.
The consoles are the primary semi-custom market, but that's probably a mature market. Sony may be the most reliant on AMD's hardware, and in-line with what AMD's product offerings permit. That's one customer, and I'm not sure it can expand that much going forward because it's already successful in its market and its offerings are somewhat iffy outside of the console itself. Microsoft's broader platform strategy de-emphasizes semi-custom to some extent and could just as readily empower Intel or Nvidia.
Why Microsoft is doing this, and why it's purchasing swaths of the game developer market, is consolidation and the movement of other large vertically integrated platforms into the streaming and game markets: Google, Amazon, Apple, Facebook, etc.

Streaming or other forms of server-based service could use AMD hardware, just as those vendors could use AMD hardware for other purposes in their instances. However, those services also start to care about what can be optimized in the rest of the system or datacenter, so things like software-defined networking, or AI, or FPGAs. Intel or Nvidia can have value-add there, and those customers themselves are buying AMD's potential partners or moving to carve out parts of AMD's market for themselves.

So AMD may want to have something to offer other than a CPU or GPU only. Also, for various markets, AMD lacks the expertise, ecosystem, or history for any customer to take a chance on them, like automotive or telecom. Xilinx would have the credibility, and value-add--presuming its development pipeline isn't hiding hiccups.

If the concern is the long-term health of the company, that may be so. It's best not to have enough assets and cash on hand that it could pay for an immediate fire-sale.

I think it was claimed that Intel helped create the impression they were looking to acquire ATI for that exact outcome, and a bidding war against a phantom offering was something ATI would have wanted to encourage.
AMD went and did this to ATIC when they created the impression that other parties wanted to buy AMD's fabs.

 

Apparently the 3300x has been nowhere to be found since soon after it launched. The 3100 seems to be readily available.

Is that because it’s more likely to get two partially defective CCXs (3100) vs one perfect and one completely defective CCX (3300)?

 

Presumably you are talking about the US?

Both models have had ready availability for a while now in New Zealand.
Without knowing the situation in Asia, where these budget CPUs would have particular appeal, it's hard to reach a conclusion.

 

Yeah in the US. They are presumably very appealing to anyone regardless of region given the performance delta between a 3300x and the best CPUs isn’t that wide when it comes to games.

 

In germany, both are readily available, but the 3300X is at a 40% premium over 3100X, which makes it quite overpriced. Yes, it can be faster, but price-wise it's already halfway up to a R5 3600.

 

3300x pricing in my country is also silly. 45% premium over 3100. 3100 - 3300x price gap is actually bigger than 3300x - 3600. Might as well buy 3500x (6c/6t 32MB cache). Only around 9usd difference between it an 3300x. 3500 is also available in my country and cheaper than 3300x, but only 16MB of cache. 3300x is probably going to be faster vs 3500 on most task.

 

Supercomputer called LUMI (=SNOW (actually in this case (Large Unified Modern Infrastructure))) will be installed in Finland as part of EuroHPC-project.
It's based on HPE Cray EX and uses Milan Epycs and next gen Radeon Instincts. It'll offer peak performance of 0,55 exaflops or 550 petaflops
https://www.hpe.com/us/en/newsroom/...ch-in-science-and-unlock-economic-growth.html

edit: Fixed Genoa > Milan

 

Great news for AMD, they've really done well with the number of supercomputer wins they've managed to secure. Zen 4 + CDNA2 is shaping up to be quite the combination and Intel seemingly has no real answer until 7nm.

 

Made a mistake there, was supposed to be Milan, not Genoa. It's Zen 3 + CDNA.