News [SA] Advanced Micro Devices EPS beats by $0.01, beats on revenue

[Hitman928]

Advanced Micro Devices EPS beats by $0.01, beats on revenue (NASDAQ:AMD)

Advanced Micro Devices (NASDAQ:AMD): Q4 Non-GAAP EPS of $0.32 beats by $0.01; GAAP EPS of $0.15 misses by $0.11.Revenue of $2.13B (+50.0% Y/Y) beats by...

seekingalpha.com

Edit: initial article had the wrong EPS number and it ended up being a beat by 1 cent on EPS and beat on revenue by $50M.

They also gave lower than expected revenue guidance for Q1 but higher full year 2020 guidance. I am at meetings all day but will try to update with more info later today if others don't do it before then.

 

[RetroZombie]

For all we know Intel might by reminding them of the last time AMD was in a superior position and that they have a "bombshell" coming.

They don't even need that, they just say:

Intel: You guys need to upgrade your servers because of the hardware cpu vulnerabilities as soon as possible.
IT guys: But the new ones also have the vulnerabilities.
Intel: But the new cpus are much faster, so you lose less performance.
IT guys: Never thought about that, thank you for the advice. Buying the new stuff right away.

Example of many of the IT morons that I know, that Mark probably was talking about. Amd is not even an option and it will never be.

 

[HutchinsonJC]

@moinmoin

Smaller companies being more likely to be swamped with security issues? If you're swamped with security issues I'd dare say your company is failing, already.

IT is largely about preempting security issues: FireWall, user permissions/access, to some extent anti-virus, and one of the most important is probably your staff's training in the dos and don'ts of emails and other computer-use habbits, etc.

In the vein of preempting, AMD sounds like a winner. They just haven't had the same kind of vulnerability announcements as Intel has had of recent. They're also cheaper.

 

[UsandThem]

AMD was pretty much irrelevant for a decade, so it will take some time for them to change many people's perception.

While I understand why the priced their X570 motherboards and new CPUs the way they did, I still think it would have been smarter to continue to price them against Intel's lineup like they did with with the first two generations. Then they could have slowly increased the MSRP over a period of time, and not like they did when the X570 motherboards were announced. I think the high pricing caught many people (non HEDT users) who have bought AMD products before off guard (Just a personal opinion as an armchair CEO).

Now that the pricing has fallen a little bit (especially with their motherboards), I imagine that will help their market share increase (at the expense of their gross margin).

 

[BigDaveX]

At the risk of getting myself flamed to a crisp, I don't think Rome is really the game-changer that the original Opteron was. One of the big reasons why that chip made such big inroads into the server market was because people were running up against the limitations of the x86 architecture's 4GB memory limit, and up until the Woodcrest Xeons arrived on the scene, the only Intel alternatives had either no 64-bit support or 64-bit support that bordered on being broken and useless, with absolutely abysmal performance compared to Opteron and utterly ridiculous power consumption (and even then it took until the Nehalem Xeons for Intel to gain any decisive long-term advantage over AMD in the server market).

By contrast, Rome's main strength is in offering high core counts at performance as good as or a bit better than Skylake/Cascade Lake, at lower power usage. Which is certainly nothing to sniff at, but for anything up to about 16 cores, Intel's options are still generally good enough if you can live with the higher power consumption and up-front costs. And "good enough" is all the excuse server admins needs to keep with a brand that's been continuously leading the server market for roughly 11 years now.

(The Zen 3-derived Epycs might be that game changer mind you, but we're still a year or so off from them)

 

[moinmoin]

@BigDaveX I agree, nothing will touch Opteron as the game changer for the reasons you mentioned.

But like with Ryzen the big difference with Epyc is the continued execution with its quick steady cadence of new launches. We can expect Zen 3 based Milan to launch within this year. Intel appears to fall back even with their comparably minor refreshes. Thus the difference accumulates further for the time being, so the longer this is allowed to continue the more glaring the advantage of a switch is.

Smaller companies being more likely to be swamped with security issues? If you're swamped with security issues I'd dare say your company is failing, already.

IT is largely about preempting security issues: FireWall, user permissions/access, to some extent anti-virus, and one of the most important is probably your staff's training in the dos and don'ts of emails and other computer-use habbits, etc.

You'd be surprised. Many non-tech companies simply don't grasp the importance of IT, security or otherwise.

Also anti-virus is not a security improvement but a huge additional attack vector.

 

[Markfw]

At the risk of getting myself flamed to a crisp, I don't think Rome is really the game-changer that the original Opteron was. One of the big reasons why that chip made such big inroads into the server market was because people were running up against the limitations of the x86 architecture's 4GB memory limit, and up until the Woodcrest Xeons arrived on the scene, the only Intel alternatives had either no 64-bit support or 64-bit support that bordered on being broken and useless, with absolutely abysmal performance compared to Opteron and utterly ridiculous power consumption (and even then it took until the Nehalem Xeons for Intel to gain any decisive long-term advantage over AMD in the server market).

By contrast, Rome's main strength is in offering high core counts at performance as good as or a bit better than Skylake/Cascade Lake, at lower power usage. Which is certainly nothing to sniff at, but for anything up to about 16 cores, Intel's options are still generally good enough if you can live with the higher power consumption and up-front costs. And "good enough" is all the excuse server admins needs to keep with a brand that's been continuously leading the server market for roughly 11 years now.

(The Zen 3-derived Epycs might be that game changer mind you, but we're still a year or so off from them)

I think you underestimate the performance of Rome. While you make some good points on the original Opteron, I am really impressed with my new Rome 7742.

 

[Chicken76]

I guess slow but steady is all they get, with the buying morons at companies buying into Intels advertizing dept. Who cares about performance....

... or security, or TCO...

 

[Zstream]

It’s all about CYA. You will never get fired for buying Intel. It’s just the way it is. Until AMD can implement some special feature into VDI, database, or something else, people have no desire to switch.

Remember, you need UCS chassis’, Nutanix, Juniper, F5, HP to all recommend AMD.

 

[guachi]

I take "semi-custom" to mean "consoles". And consoles crashing in sales isn't really surprising. The upside from the report is they have enough products to overcome weak sales in other areas.

I suspect the new consoles will sell extremely well. It isn't just that they are new and semi-custom revenue will increase it's that the console has Ryzen and Radeon inside - the two areas that had the highest growth for AMD last year. When you are talking individual consumers (and not IT departments), people are choosing AMD.

 

[moinmoin]

I take "semi-custom" to mean "consoles". And consoles crashing in sales isn't really surprising. The upside from the report is they have enough products to overcome weak sales in other areas.

Indeed. The interesting part is that with semi custom and data center AMD has combined both its lowest margin and its highest margin businesses for the purpose of accounting. Seems like a cunning plan to hide the effect of cyclical low margins in semi custom. Once the data center business really gets rolling it should completely dwarf the former (it already successfully cancels it out now).

 

[RetroZombie]

but for anything up to about 16 cores, Intel's options are still generally good enough

Very good point. I would go even lower.
Intel have the Xeon E (Entry level) from quad core to eight core market just for themselves.

They should have already released some Epyc 2000 (or opteron) even if they just soldered them on the motherboards, no need to create a new socket.

Give the line some TDP wall around the 80 Watts, and release just three models (desktop based like intel):

-Ryzen R5 1500X Quad core 16MB L3 = Epyc E2104​

-Ryzen R5 1600 Six core 16MB L3 = Epyc E2106​

-Ryzen R7 1700 Eight core 16MB L3 = Epyc E2108​

With these they would kill all the intel 23 models.

And later:

-Ryzen R7 3700 eight core 32MB L3 = Epyc E2208​

-Ryzen R9 3900 twelve core 32MB L3 = Epyc E2212​

-Ryzen R9 3950 sixteen core 32MB L3 = Epyc E2216​

If they really want to gain market share they need to do this otherwise going just after the datacenter is very limited, there are many offices with local apps that don't trust the cloud or don't have high bandwidth ISP provider and they need all their stuff local.

 

[moinmoin]

@RetroZombie I don't think at the same amount of cores the pressure to act is high enough (for the exact reason @BigDaveX mentioned). Offering the ability to consolidate multiple servers at same TCO, TDP, space etc. is the real selling point right now that wouldn't work at lower amounts of cores with vastly inferior I/O. Thus I don't think AMD will introduce those until the market moves more into their favor and the demand for it is really high, if at all. (Actually they partly exist already, as Embedded Epyc server boards. But those are obviously niche.)

 

[RetroZombie]

@RetroZombie I don't think at the same amount of cores the pressure to act is high enough (for the exact reason @BigDaveX mentioned). Offering the ability to consolidate multiple servers at same TCO, TDP, space etc. is the real selling point right now that wouldn't work at lower amounts of cores with vastly inferior I/O. Thus I don't think AMD will introduce those until the market moves more into their favor and the demand for it is really high, if at all. (Actually they partly exist already, as Embedded Epyc server boards. But those are obviously niche.)

But if you don't have a product to address that market (small server in small company) you will end up with zero cpu sales. And the vastly inferior i/o is not a problem their because out of the only 20 pcie lanes they will use none.
I think they have the market badly addressed, or incompletely addressed.

Just to give an example the 'new' amd ryzen 5 1600AF, it would be preferable if amd released that 'same cpu' with the epyc/opteron value brand than to address something that didn't needed to be addressed, because if you don't buy the ryzen 5 1600 you buy other one, at best they are killing the refurbished market and those youtubers fans that like to recommend obsolete xeon cpus paired with potato chinese motherboards.

Don't get me wrong it's all fine amd winning big data center contracts like this one:
AMD's Epyc 7742 will power new Atos meteorological supercomputer aimed at enhanced weather prediction
But those aren't volume sales, and increase their market share, or revenue, ...

 

[moinmoin]

But if you don't have a product to address that market (small server in small company) you will end up with zero cpu sales. And the vastly inferior i/o is not a problem their because out of the only 20 pcie lanes they will use none.
I think they have the market badly addressed, or incompletely addressed.

Just to give an example the 'new' amd ryzen 5 1600AF, it would be preferable if amd released that 'same cpu' with the epyc/opteron value brand than to address something that didn't needed to be addressed, because if you don't buy the ryzen 5 1600 you buy other one, at best they are killing the refurbished market and those youtubers fans that like to recommend obsolete xeon cpus paired with potato chinese motherboards.

Don't get me wrong it's all fine amd winning big data center contracts like this one:
AMD's Epyc 7742 will power new Atos meteorological supercomputer aimed at enhanced weather prediction
But those aren't volume sales, and increase their market share, or revenue, ...

You can only expand the market if you're already settled in that market as an acceptable choice. And you better start from the top of the possible performance/value range instead offering bottom grade choices as well, as AMD learned the hard way over the years in the mobile market were a lot of products using its chips where optimized for the cheapest choices, if at all. And looking how few server motherboards for Epyc are being offered compared to for Intel chips it's better to ensure all those are for the high margin Epyc chips, not some lower margin entry offerings that wouldn't change the whole picture much even if successful.

But as I mentioned before with the embedded Epyc 3000 line there actually already exists what you are asking for. Like by Supermicro.

 

[RetroZombie]

But as I mentioned before with the embedded Epyc 3000 line there actually already exists what you are asking for. Like by Supermicro.

Was expecting some tower server, not mini itx stuff, but more or less ok.

From what I see:

- the mobile market will be almost fully addressed when renoir releases​

- desktop hedt is fully addressed (they could release eight channel but I think it's not needed, preferable some epyc 5000 series)​

- desktop almost fully addressed (missing more apus, or just do some i/o die with integrated vega)​

- datacenter is fully addressed, (missing 4S systems, they could solve by solder the i/o die on the motherboard to do all the i/o's dies connectivity or by reducing the number of pcie slots on the 4S systems)​

- server partially addressed, missing some price points and a less expensive platform (could even be the TR platform)​

 

[moinmoin]

Was expecting some tower server, not mini itx stuff, but more or less ok.

From what I see:

- the mobile market will be almost fully addressed when renoir releases​

- desktop hedt is fully addressed (they could release eight channel but I think it's not needed, preferable some epyc 5000 series)​

- desktop almost fully addressed (missing more apus, or just do some i/o die with integrated vega)​

- datacenter is fully addressed, (missing 4S systems, they could solve by solder the i/o die on the motherboard to do all the i/o's dies connectivity or by reducing the number of pcie slots on the 4S systems)​

- server partially addressed, missing some price points and a less expensive platform (could even be the TR platform)​

Tyan does server/workstation motherboards for Threadripper. ASRock has a server rack for Ryzen. But aside such I can imagine AMD filling the gaps on their own with Epyc parts over the time as well, it's only a question of when not if.

 

[RetroZombie]

Tyan does server/workstation motherboards for Threadripper. ASRock has a server rack for Ryzen. But aside such I can imagine AMD filling the gaps on their own with Epyc parts over the time as well, it's only a question of when not if.

Thank you for the links, now that's much better, but see it's tyan 'solving' a product 'problem' instead of amd.

On the S8015 what's the missing chip in the board? Do you know what would/should be in there? It even have some holes for some kind of cooler:

 

[RetroZombie]

Tyan does server/workstation motherboards for Threadripper. ASRock has a server rack for Ryzen. But aside such I can imagine AMD filling the gaps on their own with Epyc parts over the time as well, it's only a question of when not if.

Well I have found that the space is actually for the gpu, with integrated AMD Radeon E8860.
How about the motherboard makers start doing some designs like that for the guys who want gpuless ryzens 5/7/9 with 'onboard' graphics.
There's certainly a big market for that!