Question How ahead is Intel in CPU design compared to AMD?

[exquisitechar]

after a long time, we can not find a icelake server CPU leaks
it should be out IMO, but it isn't...

We got a Geekbench result just today, actually.

 

[Adonisds]

I really don't know much about this subject, and that's why I asked for help here, but here's my take:

https://www.reddit.com/r/Amd/comments/dw7msh
I found the link above that simplifies things a lot for me. Assuming that what Locuza said is right, then Ice Lake's 10nm and the 7nm HPC used in Zen 2 have nearly exactly the same density. He says 2MB of L3 in Sunny Cove uses 1.33mm², 0.7% more than in Zen 2 where it uses mm² 1.32mm². I don't know if that means that the rest of the chip would also be 0.7% bigger using Intel's process, or if each part would have a different density compared to TSMCs's 7nm, but I'm assuming every part is 0.7% bigger in Intel's 10nm.

He also says that a Sunny Cove + its L2 uses 5.09mm² while a Zen 2 core + its L2 uses 3.67mm². So that should mean that a Sunny Cove core uses 37.6% more transistors.

Apparently each Sunny Cove core has access to 2MB of L3, while each Zen 2 has 4MB. I have no idea how much having twice the L3 benefits a CPU, so I'll ignore that. Obviously there are also other factors to consider that I also don't know how to properly consider so I'll also ignore those.

I read somewhere that it's one of the semiconductor "laws" that if you multiply the number of transistors of a CPU core by x, it's expected that the performance will multiply by sqrt(x). I can't remember the name of the law. Does anyone know it?

So if you do that here, 37% more transistors should predict 17.3% more performance.

Since, as many said here, Sunny Cove IPC is around 15% higher, it doesn't seem that the Sunny Cove design is much more efficient than Zen 2.

That also ignores that one design might clock higher than the other because it has a deeper pipeline or something, since I have no way of estimating and correcting for that.

Initially I was taking a look at die images of Sunny Cove and Zen 2 to try to compare them, but I didn't know how to select parts of each die that had the same units to make a fair comparison. So that reddit link helped a lot.

There are many things that can be improved here obviously.

 

[inf64]

As per Ian from AT Zen2 has around 7% higher IPC than Skylake. Since Sunny Cove is around 18% faster than Skylake that implies that Sunny Cove is 1.18/1.07~=1.1 or 10% faster than Zen2 core. Zen3 core will be mostly IPC gain so it's almost certain that Zen 3 will have higher IPC than Sunny Cove. Zen3 vs Tiger Lake will be a close call but then we do have Zen4 coming in 2021 and Golden Cove potentially coming out the same year. I think that it will be a close battle except in AVX512 with AMD having a huge lead in desktop and server segments while intel having a slight lead in mobile(pure x86, not counting the iGPU).

 

[RetroZombie]

If we could get both companies to design a CPU core now on the same process and using the same number of transistors, who would be ahead and by how much?

I would say intel major trump card was the foundries. They where ahead everyone else by one process node, something like up to 3 years if i would use time.

Everyone was saying it was a combination of both process and design advantage, but interestingly that with the foundry issues they lost both.
Five years is a lot even small companies with smaller resources allow amd, apple, arm and maybe others to surpass intel.

Amd having epic at 2x faster than xeon at almost half the power says a lot, thanks to tsmc 7nm.
If amd would design zen2 still at GF 12nm would not be so great but still great, where the process is important but the design too.

It's interesting that intel roadmap had skylake to live from two quarters up to four quarters in 2015...

Compare that with the intel new roadmap and it's interesting:

It's like a dart game, you always win no it matter where you hit

 

[Adonisds]

I would say intel major trump card was they foundries. They where ahead everyone else by one process node, something like up to 3 years if i would use time.

Everyone was saying it was a combination of both process and design advantage, but interestingly that with the foundry issues they lost both.
Five years is a lot even small companies with smaller resources allow amd, apple, arm and maybe others to surpass intel.

Amd having epic at 2x faster than xeon at almost half the power says a lot, thanks to tsmc 7nm.
If amd would design zen2 still at GF 12nm would not be so great but still great, where the process is important but the design too.

It's interesting that intel roadmap had skylake to live from two quarters up to four quarters in 2015...

Compare that with the intel new roadmap and it's interesting:

It's like a dart game, you always win no it matter where you hit

I've always wondered what Intel's chip designers were doing during all the time 10nm was being delayed. Did they use the extra time to try to make planned microarchitectures a little better?

 

[maddie]

I found the link above that simplifies things a lot for me. Assuming that what Locuza said is right, then Ice Lake's 10nm and the 7nm HPC used in Zen 2 have nearly exactly the same density. He says 2MB of L3 in Sunny Cove uses 1.33mm², 0.7% more than in Zen 2 where it uses mm² 1.32mm². I don't know if that means that the rest of the chip would also be 0.7% bigger using Intel's process, or if each part would have a different density compared to TSMCs's 7nm, but I'm assuming every part is 0.7% bigger in Intel's 10nm.

That should really be stated as maximum theoretical density. In practice, real world layouts often come nowhere close to these numbers. Heat dissipation, high frequency targets, etc all lower the transistor density.

 

[Adonisds]

That should really be stated as maximum theoretical density. In practice, real world layouts often come nowhere close to these numbers. Heat dissipation, high frequency targets, etc all lower the transistor density.

You probably didn't read it all. He analysed die photos of both designs. How is pictures of consumer silicon "maximum theoretical density"?

 

[Arkaign]

I hope future Intel designs don't try to cram iGPU into the entire consumer lineup. Being able to choose between say an i9G 10C/20T vs an i9K 12C/24T would be a welcome option. The dang IGPUs are gigantic now, and represent basically all of the transistor count increases all the way from i7-860 era to 7700k. Would have been interesting to see what the socket 115x could have achieved without that dragging on things.

At the same time I feel like AMD could do more by adding a Vega chiplet to more AM4/AM5 consumer CPUs. The Ryzen 1300 for example is really hard to justify as a budget work PC option due to needing a GPU to go with it. This is true even up towards higher end options. I'd like to build office PCs using Ryzen 3600, but it's less attractive if I also have to buy a new GPU to go with it (can't really get away with used or old stock for business clients).

 

[dmens]

I've always wondered what Intel's chip designers were doing during all the time 10nm was being delayed. Did they use the extra time to try to make planned microarchitectures a little better?

They spent most of their time getting friendly with politically powerful managers to avoid being caught in the next round of layoffs.

 

[Carfax83]

I don't possess enough technical knowledge to answer with definite precision how far ahead (or behind) Intel is compared to AMD. Personally though as a longtime computer enthusiast and gamer, I have greatly preferred Intel's design decisions over the years, and other than the disastrous 10nm node, I like what they've done and I'm hoping they can get back on their feet soon so I can update my rig. Large monolithic dies with big multi level cache, high IPC and clockspeed plus an integrated memory controller for low latency and wide vector SIMD have made PC performance better than ever in my opinion in the applications I care about, ie mostly gaming. My 6900K at 4.3ghz is quite old by tech industry standards, yet I can still easily hit and sustain triple digit framerates in many of the most demanding and performance intensive PC games and the CPU is not bottlenecking my Titan Xp.

Also, I am close to 50ns for memory latency with quad channel DDR4 3400.

I'm not totally convinced about AMD's decision to embrace chiplet based designs. Although the Zen 2 core is strong, the step backward for memory latency is disconcerting to me. Sure the massive L3 cache helps a lot to hide memory latency, but that can only get you so much. It will be interesting to see what improvements if any Zen 3 will have in that area.

 

[mikk]

Sunny Cove will compete with the Ryzen 4000 series and I don't think Intel is ready for that....

Not yet available. Once it is available we could compare it with Willow Cove and see how big the deficit is for AMD.

 

[lobz]

Curiosity, fun, I could learning something about CPU design.

Do you often visit question threads just to tell the OP that you don't like the question?

He actually told you his opinion too.

 

[lobz]

As per Ian from AT Zen2 has around 7% higher IPC than Skylake. Since Sunny Cove is around 18% faster than Skylake that implies that Sunny Cove is 1.18/1.07~=1.1 or 10% faster than Zen2 core. Zen3 core will be mostly IPC gain so it's almost certain that Zen 3 will have higher IPC than Sunny Cove. Zen3 vs Tiger Lake will be a close call but then we do have Zen4 coming in 2021 and Golden Cove potentially coming out the same year. I think that it will be a close battle except in AVX512 with AMD having a huge lead in desktop and server segments while intel having a slight lead in mobile(pure x86, not counting the iGPU).

Zen 3 vs Tiger Lake is almost nonsensical, since Tiger Lake is just mobile. The underlying uarch will be released to desktops as Rocket Lake, which will still be on 14 nm, but a much bigger core than Skylake. Like 25-30% less profit for intel on each chip sold.

Also, anyone who expects Golden Cove in 2021 is living under a rock.

 

[lobz]

I would say intel major trump card was they foundries. They where ahead everyone else by one process node, something like up to 3 years if i would use time.

Everyone was saying it was a combination of both process and design advantage, but interestingly that with the foundry issues they lost both.
Five years is a lot even small companies with smaller resources allow amd, apple, arm and maybe others to surpass intel.

Amd having epic at 2x faster than xeon at almost half the power says a lot, thanks to tsmc 7nm.
If amd would design zen2 still at GF 12nm would not be so great but still great, where the process is important but the design too.

It's interesting that intel roadmap had skylake to live from two quarters up to four quarters in 2015...

Compare that with the intel new roadmap and it's interesting:

It's like a dart game, you always win no it matter where you hit

You mean it's like a phony dart game, where you can always say to the investors that you won no matter how hard you missed

 

[lobz]

I don't possess enough technical knowledge to answer with definite precision how far ahead (or behind) Intel is compared to AMD. Personally though as a longtime computer enthusiast and gamer, I have greatly preferred Intel's design decisions over the years, and other than the disastrous 10nm node, I like what they've done and I'm hoping they can get back on their feet soon so I can update my rig. Large monolithic dies with big multi level cache, high IPC and clockspeed plus an integrated memory controller for low latency and wide vector SIMD have made PC performance better than ever in my opinion in the applications I care about, ie mostly gaming. My 6900K at 4.3ghz is quite old by tech industry standards, yet I can still easily hit and sustain triple digit framerates in many of the most demanding and performance intensive PC games and the CPU is not bottlenecking my Titan Xp.

Also, I am close to 50ns for memory latency with quad channel DDR4 3400.

I'm not totally convinced about AMD's decision to embrace chiplet based designs. Although the Zen 2 core is strong, the step backward for memory latency is disconcerting to me. Sure the massive L3 cache helps a lot to hide memory latency, but that can only get you so much. It will be interesting to see what improvements if any Zen 3 will have in that area.

1: Skylake-X is the worst example for IPC in gaming, as it falls even behind consumer Skylake.
2: "Sure the massive L3 cache helps a lot to hide memory latency, but that can only get you so much." - that can only get you so much, that Zen 2 has a higher IPC in the vast majority of workloads. Only so much.

 

[lobz]

Not yet available. Once it is available we could compare it with Willow Cove and see how big the deficit is for AMD.

Deficit for AMD? Good luck with that. Or did you mean strictly mobile up to 4 cores?

 

[liahos1]

They are not. AMD has clearer vision, up-to-date methodology and smarter people.

 

[scannall]

Not yet available. Once it is available we could compare it with Willow Cove and see how big the deficit is for AMD.

Not sure there will be a deficit. Until parts are on the shelves, and in the hands of reviewers then marketing slides and speculation really don't mean a lot. I mean it's fun and all to guess, root on your favorite if you have one. But at the end of the day, the only numbers that count are written when the parts are shipping. From AMD or Intel.

 

[Carfax83]

1: Skylake-X is the worst example for IPC in gaming, as it falls even behind consumer Skylake.

Yeah because Skylake-X has comparatively much less L3 cache than mainstream parts, slower non inclusive L3 cache, and the mesh topology which all hurt its gaming performance.

That said, Cascade lake-X can easily be tweaked to offer excellent gaming performance with overclocking (both core and mesh) and using high speed DRAM. I really want to see a desktop Willow Cove part with more than twice the amount of L3 cache as Cascade lake-X, assuming the HEDT version would keep the same cache configuration as the mobile version.

2: "Sure the massive L3 cache helps a lot to hide memory latency, but that can only get you so much." - that can only get you so much, that Zen 2 has a higher IPC in the vast majority of workloads. Only so much.

I already said that Zen 2 is a strong core. But for the sake of fairness, many people forget that it's also going up against an almost 5 year old Intel core.

If Intel hadn't screwed up their 10nm process, a Willow Cove desktop variant would be significantly more performant.

 

[DrMrLordX]

1. By that definition Zen 3 is also ready, and has been for months.

True. If Intel and AMD could produce and sell their best cores today, it would be Zen3 vs. Willow Cove or Zen3 vs. Golden Cove. Zen3 should actually be far into QS by now. ODM shipments should start next month.

2. You're right, we don't know about Golden.

Exactly, it's an unknown. AMD's progress is easier to track. Zen3 is basically ready in the form of Milan. I would guess that Zen4/Genoa isn't really ready yet. Probably still working it out in simulation.

I just wanted to know how important is CPU design. Do two different teams, one with more resources than the other, produce nearly the same CPUs if they target the same ISA and process? Is what's going to define what CPU core will perform better mostly what company is using the more advanced process?

If that's what you wanted to know, you should have been looking at the ARM world instead. A13 vs A76. Obviously Apple's team produced the faster CPU. There are A76-based SoCs produced on TSMC processes idential to the TSMC process used for A13. Heck even A77 is slower than A13, apparently.

 

[tamz_msc]

Not yet available. Once it is available we could compare it with Willow Cove and see how big the deficit is for AMD.

Willow Cove has zero IPC improvements based on the leaks so far.

 

[coercitiv]

If Intel hadn't screwed up their 10nm process, a Willow Cove desktop variant would be significantly more performant.

The problem with this different timeline thinking of "what if X hadn't failed" is the weird butterfly effect. We know architectures are built with several production parameters in mind: cost, power, performance. So once you "go back in time" and change the outcome of node R&D, then you might influence architecture capabilities as well, because fixing a node is not done for free.

Let's explore. Let's say I get to go back in time and warn Intel of their 10nm fiasco. They believe me because I know the secret Intel password for time travelers. Armed with this knowledge Intel retargets 10nm parameters for 100% success chance. Better safe than sorry, right? Definitely better than this reality, but expect engineers to inform management that Cove arch was built with specific density in mind, hence in order to maintain the high margins they're used to... Cove will get some area adjustments.

So which Willow Cove should we compare Zen 2 against, the one we see today on a limping 10nm node... or the one they would have built for a realistic 10nm node?

Food for thought, even if you don't like time travel stories. It won't be long now until Zen 2 will face Willow Cove in the mobile space, I wonder if it will be the "real" one.

 

[dmens]

And how many years did you spend in the CPU design business?

 

[uzzi38]

Well they're capable of going two months without burning another roadmap, so I'd say his statement is correct.

 

[guachi]

AMD would be ahead because Intel would spend transistors on the die for a GPU.