Question- is AMD outmaneuvering Intel?

[Topweasel]

Taking advantage of Intel's bad situation is the main point of AMD playing their cards right. Even then, Intel's woes won't last forever. Intel still has a market cap of almost $200 billion dollars, and they can spend a hell of a lot more on R&D than AMD could. Also, Intel is still way ahead of AMD on mobile devices, especially with 10nm Ice Lake on the way.

That said, I'd imagine that Intel is going to stagnate for a bit in both desktop and server markets. Outside of Ice Lake right now, Intel doesn't have anything else to get excited about.

It really depends on what is in the bag for Intel with Icelake. Right now they are having success and undervolting their Coffeelake CPU's and running them in mobile devices. They keep the clock/IPC/and now Core count advantage over AMD's mobile product range. If AMD has a mobile Zen 2 based product out this year or early next year much of that can change. Zen is ridiculously efficient, clocks should improve greatly on the low end with 7nm, Zen 2 might be at worse tying Intel in IPC (so possibly falling behind in IPC against Icelake) at worst case scenario if core count matches its at best a wash. But with Icelake there is currently and possibly for 10nm's life a significant clock and efficiency loss. Which will probably continue to be a wash on that end. The only thing that is really hurting AMD is their cadence for mobile products.

 

[AnandThenMan]

Taking advantage of Intel's bad situation is the main point of AMD playing their cards right. Even then, Intel's woes won't last forever. Intel still has a market cap of almost $200 billion dollars, and they can spend a hell of a lot more on R&D than AMD could.

Could? In 2018 Intel spent $13 billion on R&D AMD $1.4 billion.

 

[Markfw]

Could? In 2018 Intel spent $13 billion on R&D AMD $1.4 billion.

Not saying that this is true at AMD/Intel, but the company I retired from wasted a TON of money, and it is POSSIBLE that AMD at that spening limit could actually beat Intel in real actual progress. Big corporations do that, like the federal government. Small s usually more efficient.

 

[sandorski]

Not saying that this is true at AMD/Intel, but the company I retired from wasted a TON of money, and it is POSSIBLE that AMD at that spening limit could actually beat Intel in real actual progress. Big corporations do that, like the federal government. Small s usually more efficient.

Intel has a broader Portfolio in which to spread their R&D $ as well. I'm sure they are outspending AMD by a wide margin on CPUs, but it's certainly not 10-1(at least at this time).

 

[Topweasel]

Not saying that this is true at AMD/Intel, but the company I retired from wasted a TON of money, and it is POSSIBLE that AMD at that spening limit could actually beat Intel in real actual progress. Big corporations do that, like the federal government. Small s usually more efficient.

It's like the look line from ID4. 10k for a hammer 20k for toilet seat. He was talking about miss reporting costs to creat slush fund to keep Area 51 up and running. But the idea still holds. The bigger you are the more steps in the development process the smaller return on money spent.

That doesn't include the fact the Intel has so many more hands in more jars. The Fabs have their own R&D. Intel is going to be spending a lot more on GPU development than AMD and won't have anything to show for it for a while. Dozens of Mono dies that they have to develop and test without a single dollar spent on making the cores themselves any better. Cometlake isn't a free chip. That's not to mention niche products like Xeon-phi.

So they end up with a smaller ROI performance wise and might not be spending that much more than AMD in the markets that AMD is trying to compete in.

 

[beginner99]

Could? In 2018 Intel spent $13 billion on R&D AMD $1.4 billion.

Intels R&D costs include the one for process tech which is large. It also includes anythign from modems, 3dxpoint, fpgas, specialised AI chips,....hence the actually difference spent on CPUs will be far smaller.

 

[AnandThenMan]

Not saying that this is true at AMD/Intel, but the company I retired from wasted a TON of money, and it is POSSIBLE that AMD at that spening limit could actually beat Intel in real actual progress. Big corporations do that, like the federal government. Small s usually more efficient.

Agreed. Now this notion that Intel could simply spend whatever they want and "crush" AMD has been brought up many times over the years which has always begged the question, why doesn't Intel do just that? I'd say the most common response is they did not and don't want to for various reasons including to avoid the eye of regulators.

IMO the most important thing AMD did was create AMD64, without it Intel would have gone with a new instruction set architecture and locked AMD out for good.

Intel has a broader Portfolio in which to spread their R&D $ as well. I'm sure they are outspending AMD by a wide margin on CPUs, but it's certainly not 10-1(at least at this time).

Even if we say conservatively that Intel is out spending AMD 2:1 that should in theory be more than enough to put AMD far in their rear view mirror. I wonder how long they can continue with their fab business model that is hurting them badly IMO. Intel has too many things going at once from a logistics standpoint very difficult to run a lean business. Consider auto makers they don't actually make most of the car it is better to out source components from companies that specialize.

See Tesla as a company that is finding this out the hard way.

 

[TheELF]

Agreed. Now this notion that Intel could simply spend whatever they want and "crush" AMD has been brought up many times over the years which has always begged the question, why doesn't Intel do just that? .

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but...

Now intel made an F version of the 9900k, how much space ,power and thermals does that free up?How many cores could they fit into that envelope theoretically?
Intel can keep trolling AMD with 14nm and come out on top even on absolute performance,even for people that take one single metric and stick to it as if it's the one and only metric that counts(CB) and this wouldn't even cost that much it's not any new tech or anything, they could just design a new chipset that they would sell on top of that as they do almost every year.
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

Spoiler

 

[BigDaveX]

When you stop to really think about it Intel only has the best products whey their competition is barely above water, if they have healthy competition then Intel finds itself behind.

Uh, you do realise you could say the same about literally any company which exists in a duopoly-type situation? Including AMD itself?

Besides, there have been at least two occasions where Intel has had some healthy competition from AMD - P4 Northwood vs Athlon XP, and Nehalem vs Phenom II - but has still been ahead overall.

 

[moinmoin]

In another thread we were talking about Intel throwing money at 10nm to get it to work. Intel's biggest issue right now is that a lot of their R&D in their core business is interdependent. The lack of progress on 10nm ensured that their advancements in core designs is stuck in the node limbo, with Ice Lake on mobile possibly being the first real outlet of them since Skylake. Note how we also missed one whole iGPU gen due to Cannon Lake with its 10th gen iGPU only shipping completely disabled. That's a lot of R&D money down the drain.

Like with the redundancy in their chip and product design AMD seems to prepare for all eventualities in R&D as well, putting adaptability first. Note how GloFo dishing its 7nm didn't result in any big endless story like Intel's 10nm. Instead AMD seemed well prepared to go with TSMC. And since then it became obvious AMD is interested in following TSMC's fast cadence of nodes which could potentially be faster than even the 2 years per node cadence Intel managed until 2011.

 

[DrMrLordX]

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but…

What are you going on about? y-cruncher supports AVX2, and neither Summit Ridge nor Pinnacle Ridge support that properly (using the AVX2 binary provides no performance advantage). Skylake/Kabylake/Coffee Lake has a huge throughput advantage running highly-parallel, SIMD-optimized software over older Zen/Zen+ CPUs thanks to Zen/Zen+ only having 128-bit FMACs. None of what you said about "IPC advantage" applies in a bench like that, or in 3DPM v2.x either (which uses OpenMP so it is sort-of vectorized).

Now intel made an F version of the 9900k, how much space ,power and thermals does that free up?

You'll find out if/when Intel finally launches a chip with more than 8 cores on a consumer socket. In the meantime, none, since unless you're using the iGPU on a 9900k for something, it's power-gated and draws no power.

 

[beginner99]

In another thread we were talking about Intel throwing money at 10nm to get it to work. Intel's biggest issue right now is that a lot of their R&D in their core business is interdependent. The lack of progress on 10nm ensured that their advancements in core designs is stuck in the node limbo, with Ice Lake on mobile possibly being the first real outlet of them since Skylake. Note how we also missed one whole iGPU gen due to Cannon Lake with its 10th gen iGPU only shipping completely disabled. That's a lot of R&D money down the drain.

I think they learned their lesson as they stated that future core designs will be process independent. I don't think R&D is fully down the drain. I'm sure lessons were learned for Gen11 and Xe. But yeah it's possible once they get to 7nm and new cores there will be a Conore-like moment.

Like with the redundancy in their chip and product design AMD seems to prepare for all eventualities in R&D as well, putting adaptability first. Note how GloFo dishing its 7nm didn't result in any big endless story like Intel's 10nm. Instead AMD seemed well prepared to go with TSMC.

I think it was the other way around. AMD ditching GF in favor of TSMC leading to GF having to ditch 7nm. Combined with the chiplet /IO die that was a genius move. They can fullwill WSA with IO dies while at the same time limiting the WSAs impact as GF doesn't offer a leading edge node anymore.

 

[maddie]

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but...

Now intel made an F version of the 9900k, how much space ,power and thermals does that free up?How many cores could they fit into that envelope theoretically?
Intel can keep trolling AMD with 14nm and come out on top even on absolute performance,even for people that take one single metric and stick to it as if it's the one and only metric that counts(CB) and this wouldn't even cost that much it's not any new tech or anything, they could just design a new chipset that they would sell on top of that as they do almost every year.
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

Spoiler

"So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower."

I can't even begin to understand most of this line. IPC heavy software and software with less IPC?

What is being said here?

 

[Markfw]

"So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower."

I can't even begin to understand most of this line. IPC heavy software and software with less IPC?

What is being said here?

It sounds to me like he is trying to talk the new AMD chips down any way he can, even if it makes no sense. Wait until benchmarks come out, then we can refute this with proof.

 

[Topweasel]

It sounds to me like he is trying to talk the new AMD chips down any way he can, even if it makes no sense. Wait until benchmarks come out, then we can refute this with proof.

It also ignores any advancement made with Zen 2 over Zen and Zen+. I get his point (now that I select show ignored content), that Zen is a bit wider then Skylake but still doesn't get more done when running benchmarks that test IPC. Other programs that don't try to fill the pipeline also don't run particularly well on the CPU due to clock disadvantages. He is using extreme information maybe even mostly pulled out his bum, but the point does stand and he isn't wrong. But Zen isn't a well massaged arch that has been in development since 2010, based on an arch started in the late 90's. It's what Zen 2 from the information provided shall prove. That there is a bunch of things AMD can do to Zen to help its performance, with AMD doing in 2 years what it took Intel 7-8 to do.

 

[Kenmitch]

AMD looks to currently have the upper hand. AMD is leaner and meaner than Intel so I'd imagine it's much easier to concentrate on the big game plan and the hurt they have aimed at Intel. Interesting times ahead for sure. Worst case scenario those who still prefer Intel will get a better deal on their offerings.

 

[HutchinsonJC]

I don't really view it as AMD "out maneuvering" so much as just seeing it as AMD playing a fun and exciting game of advertisement with what they have achieved.

It's not like Intel has come out with anything with such marked improvements in years recent and it's not like Intel has sat as underdog and struggled competing where all of a sudden they rose to a level of maybe actually trading blows.

Benchmarks of *our* choice will be a deciding factor of how successful it'll be relative to Intel's offerings.

I think the KeyNote at Computex and Lisa's role there really helped to create excitement. The collective whole of the enthusiast community hasn't really been *this* excited about new CPUs coming out since Zen was first announced. The two most exciting CPU announcements in the last decade are just from the past few years, and both of those from AMD.

For someone that don't see themselves as a fan of either camp, I'm just glad to see technology progress... to not sit stagnate. And for me, that's the most exciting element of all of this.

 

[lobz]

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but...

Now intel made an F version of the 9900k, how much space ,power and thermals does that free up?How many cores could they fit into that envelope theoretically?
Intel can keep trolling AMD with 14nm and come out on top even on absolute performance,even for people that take one single metric and stick to it as if it's the one and only metric that counts(CB) and this wouldn't even cost that much it's not any new tech or anything, they could just design a new chipset that they would sell on top of that as they do almost every year.
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

Spoiler

At least after this post of yours I don't ever have to wonder again about the intention behind your comments on this forum, and I thank you for that.

 

[Thunder 57]

Intel can keep trolling AMD with 14nm and come out on top even on absolute performance...

Intel is trolling AMD so hard on 14nm they just decided to sit on 10nm and keep revising 14nm over and over. Oh wait...

I could give you a list of reasons why AMD's immediate future looks brighter than Intel's, but it would be a huge waste of my time. I never understood it. Most people seem to be able to research the reviews and buy whatever is "best" at the time, or a coin flip if they are similar. I don't care whether that be AMD/Intel, AMD/NVIDIA, Ford/Chevy, etc. Yet a select few see one that can do no wrong, and will always find some fault in the other no matter how good it may be, real or imagined.

 

[chrisjames61]

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but...

Now intel made an F version of the 9900k, how much space ,power and thermals does that free up?How many cores could they fit into that envelope theoretically?
Intel can keep trolling AMD with 14nm and come out on top even on absolute performance,even for people that take one single metric and stick to it as if it's the one and only metric that counts(CB) and this wouldn't even cost that much it's not any new tech or anything, they could just design a new chipset that they would sell on top of that as they do almost every year.
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

[/SPOILER]

This post is actually hard to read let alone understand.

 

[fkoehler]

How much more "crushing" do you want?
Skylake can execute 4 mops choosing out of 8 instructions
Zen can execute 6 mops choosing out of 10 instructions
So ZEN has at least 20% more IPC and still is barely on par and maybe a little bit faster (on same clocks) on very IPC heavy software,while software with less IPC runs a lot slower,like 25 to 50% slower... in ycruncher single thread intel is about 90% faster.This already settles it ZEN needs 20% more core just to reach equity,but...

So, what exactly are you arguing again?

Intel's had 10+ years of being King Of the Hill, with money gushing in like a spigot which has afforded them the ability to spend multiple billions on R&D far in excess of anything AMD can even dream of.
And even then, they've had to resort to bribery to keep the Young-Gun down, manipulate industry standards like the compiler and benchmarks used, and ultimately, cheat on how to actually process data the ISO way in an effort to maximize perceived speed/efficiency/IPC.

AMD gets their $hit together, and all of a sudden we have $190 8-Core procs, a year or more later and they're First To Market with PCIe4, and now a 12-Core proc, with IPC apparently roughly on par or better than Intel.

No doubt Intel is slaving away on Arch. R&D even with 10nm process in the dumper.
However, anyone with half a dram of common sense in their head can see that the days of Intel just riding the gravy train and doling out this or that tidbit to justify their pricing structure, are done.
At least for a little while now.

The above are relatively rational facts or positions, made from a non-fanboi pov.
You seem to have some inner angst thats crying out for you to persuade many of us that we're just not seeing reality the way you see it.
You seem to be trying to convince us that Up is Down, but not bringing much proof.

 

[moinmoin]

I think they learned their lesson as they stated that future core designs will be process independent. I don't think R&D is fully down the drain. I'm sure lessons were learned for Gen11 and Xe. But yeah it's possible once they get to 7nm and new cores there will be a Conore-like moment.

The problem with making core designs process independent is that that's just one half of the whole work, you still need working processes to move to. Intel went all out with 10nm and failed spectacularly. Considering they essentially use their foundries only in-house TSMC's quick cadence of nodes makes no financial sense to Intel, a single node has a harder time making back the R&D and buildup necessary if it's quickly superseded. But TSMC's quick cadence of nodes also better ensures that the process is actually going to work and (relatively) on time, something Intel has a hard time with since 2011.

I think it was the other way around. AMD ditching GF in favor of TSMC leading to GF having to ditch 7nm. Combined with the chiplet /IO die that was a genius move. They can fullwill WSA with IO dies while at the same time limiting the WSAs impact as GF doesn't offer a leading edge node anymore.

AMD couldn't ditch GloFo even if they wanted due to the WSA. They very likely wanted dual sourcing since 7nm at GloFo was uncertain and unproven (just look at their track record compared to TSMC), and GloFo eventually bailed out since unlike TSMC they have no other big consumers that make the whole process worth the financial effort.

 

[beginner99]

AMD couldn't ditch GloFo even if they wanted due to the WSA.

AMD doesn't have to use the wafers but pay for them even if unused. There was no need to do anything at GF from a contract point of view. AMD could have done everything at TSMC for the last decade with the downside of paying GF for unused wafers. Of course that was financially not viable back then.

My Point is WSA includes paragraphs about GF having to supply a leading edge node and if they can't, the fine AMD would have to pay would be a lot lower. AMD might have made a bold move to move everything to TSMC and betting on the fact that GF would then not be able to develop a 7nm node as not financially viable even with the WSA and hence invalidating large parts of the WSA.

AMD must have made the decision for TSMC 7nm at least 1-2 years before GF announced they are canning 7nm.

 

[jpiniero]

AMD must have made the decision for TSMC 7nm at least 1-2 years before GF announced they are canning 7nm.

Based upon the rumors, my belief is that they were intending to go with GloFo first but decided to use a different design based at TSMC and then port that back to GloFo to satisfy the WSA.

Now GloFo probably made up their mind to suspend efforts on 7 nm long before it became public.