Samsung Nintendo AMD semicustom ARM-x86 Showdown Scenario

[mrmt]

I wonder how their ARM adventure will go...

I think their ARM venture it's FUBAR'ed. They are too mum with Seattle and they are toning down their expectations from the ARM microserver products, to the point that last quarter the big goodwill loss was caused by a write down of their Seamicro acquisition. Lisa so far hasn't committed to keep anything, quite the opposite, she asked for 6 months before showing a roadmap for their investors, so I think she's planning deep changes on the R&D pipeline.

 

[frozentundra123456]

Granted, AMD has some expertise in servers that Qualcom, Samsung, et al dont have, but I have always wondered why AMD wants to jump into a fight with those well established ARM companies that have money to burn for R and D. Seems like almost more of an uphill fight than trying to compete with Intel in x86.

 

[Insert_Nickname]

Not to mention sole fact that Zen is the ugly sister compared to the K12 ARM part that gets much more focus.

Or Zen could just be a K12 core with an x86 translator slapped on. Just a thought...

Aren't most modern CPUs RISC internally anyway? It doesn't really make sense to develop two different cores. Why not focus on optimizing a single core design, then just add a translator? I mean, it isn't unheard of, Transmeta did something similar years ago.

 

[ShintaiDK]

Or Zen could just be a K12 core with an x86 translator slapped on. Just a thought...

Aren't most modern CPUs RISC internally anyway? It doesn't really make sense to develop two different cores. Why not focus on optimizing a single core design, then just add a translator? I mean, it isn't unheard of, Transmeta did something similar years ago.

And it was so successful for Transmeta.

Then they can just as well reuse an existing core.

 

[mrmt]

Or Zen could just be a K12 core with an x86 translator slapped on. Just a thought...

What if you were AMD CEO in 2012 and you had to cut your R&D by 30% as your first measure. You also know that your foundry partner is in dire straits, with very poor results on the latest node and with their next two nodes delayed big time. As for market forecast, you also know that your growth opportunities don't lie in gaining Intel market share, because your latest big architecture was an unmitigated failure and Intel has been executing like a clock, and because of that your company won't be competitive for the foreseeable future. Then you call your engineering team bosses (or what's left of them) to discuss the roadmap beyond the unmitigated failure. With that environment, what path would you pick:

1) Let your engineer's imagination fly away in order to get the next crazy big thing, the same thinking that gave Bulldozer, a gift that keeps on giving and is poised to implode your company in the next 4 years.

2) Try to develop a solid, low cost chip in order to explore the growth opportunities outside Intel's area of influence and at the same time not push too hard your foundry partner in terms of needed capabilities.

I would be rather surprised if Rory went with something along 1), so 2) it is.

I'm pretty sure the quote about a new ARM licensee for the semicustom business is Mediatek licensing Radeon graphics IP.

If it's Mediatek and a licensing deal it would explain the paltry revenue forecast AMD gave to investors. In any case, the fact that they are chasing *this* type of opportunity doesn't bode well for their future.

 

[jhu]

To be fair, surely you can agree that of anyone who might be an AMD fan it would include AMD (employees) as well?

Can you imagine the scenario in which the AMD fanpeople are more enthusiastic and optimistic than the AMD employees themselves?

Probably when they introduced Bulldozer. Surely the people developing it knew what the performance was going to be like.

 

[geoxile]

Probably when they introduced Bulldozer. Surely the people developing it knew what the performance was going to be like.

The last performance projection for BD uarch prior to launch - that didn't use clever marketing - said the new servers would have 35% higher total performance for 33% more cores at stock. It was pretty accurate.

 

[mrmt]

The last performance projection for BD uarch prior to launch - that didn't use clever marketing - said the new servers would have 35% higher total performance for 33% more cores at stock. It was pretty accurate.

No, it is not.

 

[theeedude]

I don't get why they got into the ARM race. They can barely survive in a duopoly with just one big competitor that expects close to 60% gross margins, leaving them some breathing room to compete on cost, and wants them to stay around for anti-trust reasons in x86. But they want to jump into a tank full of hungry and nimble ARM piranhas? Sounds like desperation. If AMD can't make it in the x86 space, it's done.

 

[Thala]

I don't get why they got into the ARM race. They can barely survive in a duopoly with just one big competitor that expects close to 60% gross margins, leaving them some breathing room to compete on cost, and wants them to stay around for anti-trust reasons in x86. But they want to jump into a tank full of hungry and nimble ARM piranhas? Sounds like desperation. If AMD can't make it in the x86 space, it's done.

It is not that simple. ARM itself currently exclusively target the mobile/low power space. There is hardly anyone making big wide issue cores using AArch64 coupled with a high performance memory subsystem. There certainly is an opportunity.

 

[ShintaiDK]

I don't get why they got into the ARM race. They can barely survive in a duopoly with just one big competitor that expects close to 60% gross margins, leaving them some breathing room to compete on cost, and wants them to stay around for anti-trust reasons in x86. But they want to jump into a tank full of hungry and nimble ARM piranhas? Sounds like desperation. If AMD can't make it in the x86 space, it's done.

Its just another step in AMDs long run of self damaging decisions. ARM already lost 25% of the tablet segment and it may be 50/50 after this year.

Dense servers isnt going ARM, its just more x86.

And as you say, Intel at least commands a ~60% gross margin on traditional x86 products leaving some headroom for AMD to work within. We already see ARM competitors on 20nm while AMD still fields 28nm. So its not like the node issue will change either. Plus all AMDs experience is with x86. Its a nobody company in the ARM world.

R&D wise AMD will have to battle it out with the lower ranking ARM companies as well for the pennies. Even MediaTek is close to being a bigger company than AMD. And they already field a larger R&D budget.

But again, over 10 years with more or less with one bad decision after the other. Why would it change.

 

[NTMBK]

ARM already lost 25% of the tablet segment and it may be 50/50 after this year.

Intel bought 25% of the tablet segment by giving away SoCs wrapped in money. The mobile division had negative revenue in the last quarter; their rebates were literally larger than the money they got from sales.

 

[mrmt]

I don't get why they got into the ARM race. They can barely survive in a duopoly with just one big competitor that expects close to 60% gross margins, leaving them some breathing room to compete on cost, and wants them to stay around for anti-trust reasons in x86. But they want to jump into a tank full of hungry and nimble ARM piranhas?

I think Andrew Feldman played a pivotal role on this. He shpuld have said "if you develop a cpu with xyz capabilities then you can build a Microserver like abc and will earn a huge pile of cash". AMD should have looked at their engineering resources and reached the conclusion that it would be cheaper to go with ARM, and here they are. It was just a decision process they were ill-equiped to make.

 

[Nothingness]

Intel bought 25% of the tablet segment by giving away SoCs wrapped in money. The mobile division had negative revenue in the last quarter; their rebates were literally larger than the money they got from sales.

Also what percentage of that 25% are Windows tablets which require x86?

 

[Fox5]

Heh, ARM makes no sense for servers.

In the server market, CPU cost is relatively negligible. ISA is possibly negligible, if you have a significant advantage in performance per watt. Software costs of switching ISAs are significant though, so I can't see anyone going ARM "just because". As Nvidia's Tegra X1 offering is probably more compelling as a microserver than anything AMD will put out.

 

[monstercameron]

Heh, ARM makes no sense for servers.

In the server market, CPU cost is relatively negligible. ISA is possibly negligible, if you have a significant advantage in performance per watt. Software costs of switching ISAs are significant though, so I can't see anyone going ARM "just because". As Nvidia's Tegra X1 offering is probably more compelling as a microserver than anything AMD will put out.

how so? seattle might be mia but it had dual 10Gbe and other enterprise level accelerators.

 

[theeedude]

It is not that simple. ARM itself currently exclusively target the mobile/low power space. There is hardly anyone making big wide issue cores using AArch64 coupled with a high performance memory subsystem. There certainly is an opportunity.

There are opportunities to do something, and there are opportunities to make money. Those aren't the same.
If there was an opportunity to make money, someone smarter than AMD would already be doing it. There is no barrier to entry to license ARM cores and design a memory subsystem for it. But others looked at Calxeda et al, and decided it wasn't worth it.

 

[ikachu]

Just a random note from my personal experience... as an AMD fan who became an AMD employee (and then left after a couple of years), most of the people working there never looked at CPU benchmarks or had any real sense of how our chips performed vs Intel.

Most people would get their info on our roadmap from the all hands meetings, and in those the execs always talked about how great our chips were, and Intel was never mentioned (and Nvidia was only mentioned when it was something bad for them). So, if an employee didn't go out and do their own research, they'd probably believe everything was going great (except for the layoffs and lack of bonuses of course...).

Granted, I was working on the BIOS side, so I only interfaced with the chip design people tangentially, so I'm sure they had more of a sense of what was going on, but I think people on this site have a really skewed idea of what goes on at these kind of companies. Most people are only focused on their job and don't have time / motivation to try to keep up with the industry. And, maybe that's a good approach because I probably took AMD's failures too personally and that definitely led to me burning out on the company.

That being said, from the initial info I heard about Zen and K12, I am cautiously optimistic. I think AMD set their sights high... whether they can execute or not I have no idea though.

 

[MiddleOfTheRoad]

I don't get why they got into the ARM race. They can barely survive in a duopoly with just one big competitor that expects close to 60% gross margins, leaving them some breathing room to compete on cost, and wants them to stay around for anti-trust reasons in x86. But they want to jump into a tank full of hungry and nimble ARM piranhas? Sounds like desperation. If AMD can't make it in the x86 space, it's done.

On the contrary -- I think it's the other way around. AMD is looking for little niches in the market where they won't be competing with Intel. It does sense for them to specialize in niches instead of fighting an uphill battle against a vastly larger competitor.

 

[MiddleOfTheRoad]

Heh, ARM makes no sense for servers.

Completely not true.

It makes a ton of sense for data centers if they can get sufficient performance out of these ARM processors. It really depends on how well the actual silicon can perform.

Here is a great article why ARM can potentially be a huge for data centers:
http://www.enterprisetech.com/2014/05/06/arm-servers-now/

 

[ShintaiDK]

Completely not true.

It makes a ton of sense for data centers if they can get sufficient performance out of these ARM processors. It really depends on how well the actual silicon can perform.

Here is a great article why ARM can potentially be a huge for data centers:
http://www.enterprisetech.com/2014/05/06/arm-servers-now/

Potential for how many years?

And this just confirms ARM is as far as ever from that goal:
http://anandtech.com/show/8357/exploring-the-low-end-and-micro-server-platforms

 

[ShintaiDK]

On the contrary -- I think it's the other way around. AMD is looking for little niches in the market where they won't be competing with Intel. It does sense for them to specialize in niches instead of fighting an uphill battle against a vastly larger competitor.

If they are looking for niches, why are they going up against even more competition? AMD is a nobody in the ARM segment. And they focus on something nobody really wants to buy.

Fracturing an already miniscule R&D budget into a custom ARM design as well as x86 is on a bulldozer scale of fail.

 

[mrmt]

Most people would get their info on our roadmap from the all hands meetings, and in those the execs always talked about how great our chips were, and Intel was never mentioned (and Nvidia was only mentioned when it was something bad for them). So, if an employee didn't go out and do their own research, they'd probably believe everything was going great (except for the layoffs and lack of bonuses of course...).

The marketing and commercial teams are supposed to have detailed comparisons between their products and competition products, and the financial guys are supposed to have detailed comparisons as well because they must sell the company's strategy to banks and relevant investors.

But I'm quite baffled to know that AMD executive team was acting like the proverbial ostrich and was hiding its head beneath the sand while their product stack was being imploded on the market, this even after the CEO was fired. Tells much about the panic that should have ensued in the upper management levels.

 

[turtile]

If they are looking for niches, why are they going up against even more competition? AMD is a nobody in the ARM segment. And they focus on something nobody really wants to buy.

Fracturing an already miniscule R&D budget into a custom ARM design as well as x86 is on a bulldozer scale of fail.

It barely cost them anything compared to x86. They are sharing the same infrastructure.

And who has anywhere near the IP of AMD when it comes to servers (only considering ARM players)....

 

[JDG1980]

I don't expect Zen to miraculously leapfrog Intel, but I do expect (and hope) it will close the gap somewhat. Currently, AMD's Steamroller core barely beats Core 2 Duo in IPC, and can't match Nehalem on that front. Nehalem was released by Intel in late 2008. This is absolutely disgraceful - AMD is literally more than six years behind on the CPU front. That's why they are doing a new clean sheet design; the Bulldozer core is fundamentally flawed, just as NetBurst was for Intel. At a minimum, I expect Zen to match Sandy Bridge in IPC, and beat it in power efficiency (if only because AMD will be using a better process; remember, SB was done on 32nm, while Zen will be on 16nm FinFET). If AMD can't beat a design released by Intel in 2011 with a clean-sheet project slated for release in 2016, then they might as well hang up their hat.

The claims that Zen is going to be a design focused on low power makes no sense. AMD already has the cat cores for that, and they're quite successful if you count the console design wins. Of course Zen will have to take efficiency into account, but there would be no point in releasing something that isn't considerably better than the construction equipment cores in terms of performance. It just wouldn't be competitive.