Article "AMD vs. Intel CPU Market Share Q4 2019: EPYC and Ryzen Growth Decelerate, Mobile Ryzen Up" - Tom's

[CHADBOGA]

To be fair, back then intel's P4 was complete garbage. It was good in one thing: encoding (but just because of an instruction set advantage). That was also the time where every AMD hater tried to ridicule the FX CPUs, saying that nobody cares about gaming.
Nowadays intel's consumer CPUs have absolutely nothing wrong with them, after the core count catch-up they are actually very good products, they're just poor value compared to AMD's respective offerings (same goes for the 5500XT, it's a damn fine GPU - for $150, not for $200 ffs). Enterprise CPUs have severe security issues, but after mitigations those are just making them to have even poorer value, which seems to be almost entirely offset by their excellent sales team - or should I say sales army?
I'm not completely surprised that things are moving slower than back then, but I'm surprised that sometimes it seems they aren't moving at all

So it appears that the combo of single core advantage and breaking the 4gb barrier that the K8's had, matter/mattered much more than twice the cores and much lower power?

On the server side, Opteron addressed what was becoming a serious issue - the x86 architecture's 4GB memory limit - and Intel didn't have a remotely worthwhile response until Woodcrest, and arguably not even until Nehalem. Rome is good, but unless you need the high core counts it offers, it's not really as overwhelmingly superior as the K8 Opterons were against the mid-2000s Xeons.

On the desktop side, weird as it sounds, I think the fact that such a small percentage of Ryzens have iGPUs is really hurting AMD with OEMs, who aren't going to go to the extra mile of putting together an AMD system with a dGPU when they can save the time and money putting together an Intel system with an iGPU, which will meet the needs of probably 90% of their customers.

I agree that the lack of iGPU hurts AMD in certain segments, but as apparently the server market is both the hardest nut to crack and most lucrative, this is the area where I want to see AMD make gains, so they can become a long term proper rival to Intel.

Hasn't AMD said that they're steadily selling all of the Ryzen parts that they can make? Perhaps there's a few models that are starting to build up a stockpile, but the chiplet-based approach should allow them to make small adjustments to production to some degree to avoid this issue by increasing production to where there's more demand.

We know that they're drastically increasing their wafer purchases from TSMC as that's already been announced. They certainly wouldn't do that if they could satisfy demand with existing production.

Perhaps the simple answer is that they can't gain marketshare any faster than they currently are because they can't sell enough products to do so, at least not in any way where they could have doubled their gains or something that large.

I fear much of the increased wafer purchases they have booked from TSMC will go towards the new consoles coming out in the next 12 months.

It is AMD's own guidance which concerns me the most, as prior to hearing about it, I was fairly sure they would be gaining much more than they seem to think is on the cards.

EPYC's contribution to AMD's revenue is rather lethargic, so far. But, they claim to be on track for 10% marketshare by mid-year, and the road from 10% to 20% can be very quick, if history is any guide. Back in April 2006, the Opteron processor marked its third anniversary with 22% share. Former AMD CEO Hector Ruiz had predicted that Opteron would exit 2004 with 10%, over one and a half years after launch, which didn't happen, and he set a new target for 12% by the end of 2005. They reached 11% in 2005-Q2 already. It then doubled from there in less than a year.

I'll feel a lot more relaxed about AMD's long term prospects if and when they hit 10% server marketshare, but didn't the latest guidance from them, predict that this will be a ways off yet?

 

[itsmydamnation]

It is AMD's own guidance which concerns me the most, as prior to hearing about it, I was fairly sure they would be gaining much more than they seem to think is on the cards.

AMD is quoting 30% yoy rev growth for this year, i dont see anything wrong with that.

I'll feel a lot more relaxed about AMD's long term prospects if and when they hit 10% server marketshare, but didn't the latest guidance from them, predict that this will be a ways off yet?

They expect to this year, but the difference in the numbers is that AMD are saying 10% of the 1P and 2P markets not 10% of total DC CPU market.

 

[CHADBOGA]

AMD is quoting 30% yoy rev growth for this year, i dont see anything wrong with that.

From the thread on here about AMD's last QTR results :

Most of the 30% revenue growth should come from consoles alone:

https://twitter.com/x/status/1222397918839279616

 

[Hitman928]

From the thread on here about AMD's last QTR results :

Also from that thread, he's wrong, at least according to AMD's predictions.

 

[itsmydamnation]

From the thread on here about AMD's last QTR results :

What he has ignored is that AMD also said Q1 and Q2 console rev is almost non existent. AMD have also previously said that they make most margin from console sales later in the cycle which suggests MS/Sony only pay for/buy good dies, not wafers. So both other those lead him to vastly over estimate the performance of semi custom in fy 2020.

 

[Hitman928]

What he has ignored is that AMD also said Q1 and Q2 console rev is almost non existent. AMD have also previously said that they make most margin from console sales later in the cycle which suggests MS/Sony only pay for/buy good dies, not wafers. So both other those lead him to vastly over estimate the performance of semi custom in fy 2020.

Lisa Su said in the conference call that they expect about 20% revenue growth from non-console sources, so console revenue in 2020 should only account for about 1/3 of the predicted 30% growth (i.e. console revenue will add 10% to the total revenue in 2020 over 2019). I expect 2021 to have more significant revenue from consoles (but other sources to grow as well).

 

[eek2121]

The reality of the situation is that Intel is beginning to react. AMD was able to quickly return to profitability thanks to having lower prices and better value. However, Intel has begun to drop prices across the board in order to respond. This is a good thing. This means AMD must continue to execute and provide compelling products.

It is important to note that AMD has decent profit margins on everything except console parts. Console parts are typically high volume, low margin parts. This could actually pull down margins while increasing revenue. In order for AMD to offset that, they need to sell more high margin parts. That indicates to me that we'll see quite a few high end and upper-mid range launches this year. Existing products (Zen 2, Radeon 5000 series) will cover the low end. I doubt we'll see much in the way of new low end launches, as that would hurt margins and also cause availability issues due to volume.

I expect we'll only see 2-3 Radeon 6000 parts launching before August. As far as Zen 3 I have no idea what to expect. The move from 17h to 19h for the CPU family could indicate major changes. I don't expect SMT4. AVX-512 will either come this year or next year. Outside of that, who knows? AMD will likely want to finish closing any performance gaps. Also integrating an APU would help the OEM market, as well as those of us that use GPU pass-through in a virtualization scenario. I guess we'll see on the other stuff.

One other thing, there is a slide that shows EPYC launching later this year then it did last year. Some have used this slide to claim that AMD would not launch until Q3/Q4. I would take that with a grain of salt. AMD launched the Ryzen 3000 series before EPYC Rome. It's in AMD's best interest to launch before August due to back to school sales. July is actually really late in the game as it is, because OEMs need time to build/validate/launch systems.

 

[DrMrLordX]

Zen+ also came with a clockspeed boost.

Anyway, I don't think AMD's price increase for the 3000 series over the 2000 series is unwarranted, and if the prices had been the same, AMD would have been throwing money away, as they appear to be selling all the Ryzen 3000's they can make.

Also observe the model names for the Pinnacle Ridge (Zen+) products. The highest-end Pinnacle Ridge was the 2700x, which was roughly analogous to the 1700x from Summit Ridge. The 2700x launched at $329 while the 1700x launched at $399 but quickly came down in price due to lackluster demand. R7 2700 launched at $299, which is exactly the same as the launch price of the R7 1700. So Pinnacle Ridge pricing wasn't really an outlier.

 

[lightmanek]

I see first sings of upcoming Ryzen 4000 series desktop launch at retailers. Here in UK there are multiple Ryzen 3000 processors consistently selling way below MSRP week after week with ever lower price. 3800X below £300 or 3700X for £260 brand new anyone? Maybe 3900X for £429 ?
These 'sale' drops are more frequent lately and with record low prices almost on a weekly basis. Same pattern happend with Ryzen 2000 starting about 5 months before Ryzen 3000 series. I have a suspicion, we will get first info at Computex with launch following shortly after.
Just a speculation on my side.

 

[DownTheSky]

Sales also happening in my country, and all over the place really. Zen 3 surely getting closer.

 

[beginner99]

Anyone having been following these threads and seen my post about the quest to get a threadripper/epyc based system for "HPC", I finally got a quote for a system from IT. To get here it took about half a year. And what I wanted was actually a threadripper workstation that would cost around $7000. But that isn't stantdard compliant so it has to be a server from a big OEM. What I got offered it >100% overspeced but I'm actually shocked how cheap it is given the hardware. 2xepyc 7742 (=128 cores / 256 threads!!!), 256 GB 3200mhz RAM, and a tesla v100 + 4 tb ssd storage (over multiple drives). That would "only" cost $22k. Tells you a lot about list prices (the V100 alone is 10k, 1 7742 around $7500 so cpus + gpu alone is already over finally price if we take list price)

But the real plot twist is now that we don't have budget anymore. obviously coronavirus is a good chance to save costs.

 

[coercitiv]

what I wanted was actually a threadripper workstation that would cost around $7000. But that isn't stantdard compliant so it has to be a server from a big OEM [...] that would "only" cost $22k.

But the real plot twist is now that we don't have budget anymore. obviously coronavirus is a good chance to save costs.

Definitely insist on the $22K standard compliant server. And ask for a second one for WFH.

 

[VirtualLarry]

And ask for a second one for WFH.

 

[misuspita]

And another one for F@H. Even that way, you probably can't outcore Markfw