Intel Skylake / Kaby Lake

[beginner99]

If AMD can release Ryzen II at 7nm before Intel gets anything out the door at 10nm, then Intel will need a revolution to save them.

Not really. Same as n Athlon days. AMD won't have enough fab capacity. Intel would just need to take a small hit on margins but biggest piece of x86 would still be intel. Even in P4 days intel had more than 50% market share.

 

[tamz_msc]

It will, if you factor in time. For example, rendering and encoding jobs could take hours at a time, if not days.

Only if you're doing professional work under time constraints. If your life doesn't depend on turning out 11 videos instead of 10 within a deadline, it isn't going to hurt you in any way.

Some of the Fortran codes that run on my lab's workstations take anywhere from 12 to 30 hours to complete. It hardly matters if the code I leave to run overnight finishes at 7 AM the next day, or 8 AM.

Pros have multiple systems to avoid downtime.

 

[beginner99]

It will, if you factor in time. For example, rendering and encoding jobs could take hours at a time, if not days

True but that only matters to professionals that make money from this and if that is the case, they will go with the fastest solution they can afford regardless of performance/$.

 

[tamz_msc]

So, Let me get this right:
1) You are excluding the non-gaming tests where Zen was behind Skylake by a large amount (28.1%, 40.9%, and 23.1%).
2) Then you throw out gaming tests.
3) Then you conclude that since the remaining tests show Zen behind by 8% that there is no difference?

Do you expect anyone to take you seriously?

1. Define "gaming IPC". Go on. Try.
2. What are those tests that I exclude? 7-Zip and WinRAR, which are memory bandwidth limited and Ryzen is dual-channel. Keep Lightroom if you want - results don't change significantly. Given that DxO Optics Pro does a much better job, its more due to typical Adobe lethargy when it comes to software optimizations, more than AMD lacking in performance.
3. Want more data? Here, have a look at The Stilt's results from the Ryzen: Strictly Technical thread:

Excluding AVX2, Kaby Lake is 9% faster than Zen.

It's pathetic when some people don't even try to understand the reasons behind benchmark numbers, when to exclude extreme data points, and above all don't even bother finding out what's happening at the other camp to get a hold of the overall picture.

 

[JoeRambo]

Excluding 7-Zip, WinRAR and Lightroom, which is completely fair because Skylake-X is behind Broadwell-E in these tests, we have Skylake-X 8-core roughly 8% ahead of Ryzen 8-core, clock-for-clock, thread-for-thread.

Great logic here, Ryzen gets destroyed in these tests, but Skylake-X regresses in performance versus Broadwell-E, so it's fair to exclude them. LOL.

I personally also love how "certain" members of this forum go around spreading Intel power usage numbers: usually Intel/AMD power usage is compared by certain "dimensionless" power load, like Prime95, and Skylake-X predictably gets nasty consumption, above TDP if BIOS is relaxed etc.
What is kept silent, is that in analogous high power loads like y-Cruncher, power usage numbers are also huge for Intel, but Skylake-X is more than 3 times faster (8C vs 8C).

 

[tamz_msc]

Great logic here, Ryzen gets destroyed in these tests, but Skylake-X regresses in performance versus Broadwell-E, so it's fair to exclude them. LOL.

I personally also love how "certain" members of this forum go around spreading Intel power usage numbers: usually Intel/AMD power usage is compared by certain "dimensionless" power load, like Prime95, and Skylake-X predictably gets nasty consumption, above TDP if BIOS is relaxed etc.
What is kept silent, is that in analogous high power loads like y-Cruncher, power usage numbers are also huge for Intel, but Skylake-X is more than 3 times faster (8C vs 8C).

Epyc 'destroys' Skylake-SP in STREAM Triad. I wonder why. 🙄

Exceed L1D with large data sets in real-world code and AVX2 lead vanishes into thin air. So much for being 3X faster.

 

[JoeRambo]

Exceed L1D with large data sets in real-world code and AVX2 lead vanishes into thin air. So much for being 3X faster.

Goalpost shifting much eh? Actually 8C Ryzen is just 1.6 times faster versus 4C Skylake coupled with 2133 memory. So IPC is in fact not that great?

And vector triad is low on computation and high on memory bw / caching / prefetch requirements, AMD's 512KB L2 is certain playing huge part here. One has to wonder how 1MB of L2 in Skylake-X and AVX512 would do? 2-3 Cores of Skylake-X would be surely be enough to match 8 of Ryzen's.

 

[Arachnotronic]

Why does every one of these threads devolve into an Intel vs AMD debate?

We should have one thread for the Intel vs AMD stuff, then separate threads to talk about Intel stuff and AMD stuff, respectively.

 

[Zucker2k]

1. Define "gaming IPC". Go on. Try.
2. What are those tests that I exclude? 7-Zip and WinRAR, which are memory bandwidth limited and Ryzen is dual-channel. Keep Lightroom if you want - results don't change significantly. Given that DxO Optics Pro does a much better job, its more due to typical Adobe lethargy when it comes to software optimizations, more than AMD lacking in performance.
3. Want more data? Here, have a look at The Stilt's results from the Ryzen: Strictly Technical thread:

Excluding AVX2, Kaby Lake is 9% faster than Zen.

It's pathetic when some people don't even try to understand the reasons behind benchmark numbers, when to exclude extreme data points, and above all don't even bother finding out what's happening at the other camp to get a hold of the overall picture.

I don't understand your rant, especially since you've only shown that Ryzen does indeed trail Skylake in ipc. If you need to be reminded, here's what started what shouldn't even be a debate in the first place:

"Are you sure about that? Coffee lake is basically Skylake IPC and Skylake chips when clocked at the same as Ryzen often lose in many tests. clock per clock Ryzen has great IPC."

Is this what you are defending? By the way, in the more efficient x265 benchmark, the difference is closer to 20% between Ryzen and Skylake. Also, Skylake runs at a higher frequency in the real world. Finally, these clock for clock comparisons are not the best way to judge IPC. Although it may sound counterintuitive, chip designs are not done in a vacuum. Process (tuning) can and does influence ipc designs and decisions so the best ipc test is to just ran each test at stock, imho.

 

[TheF34RChannel]

Why does every one of these threads devolve into an Intel vs AMD debate?

We should have one thread for the Intel vs AMD stuff, then separate threads to talk about Intel stuff and AMD stuff, respectively.

When I log on and see a ton of new messages I get excited. Then when I look I see all this AMD vs. Intel stuff and I face palm. The continuous, persistent, neurotic need for certain AMD enthusiasts to come into Intel threads, cherry-pick and glorify their evidence astonishes me. Excellent idea, to have separate threads and everybody will be happy.

 

[dullard]

3. Want more data? Here, have a look at The Stilt's results from the Ryzen: Strictly Technical thread:
Excluding AVX2, Kaby Lake is 9% faster than Zen.

I'll go with your latest quote then. You just proved that Zen IPC is less than Kaby Lake. Thanks for proving my point.

 

[tamz_msc]

Finally, these clock for clock comparisons are not the best way to judge IPC.

😵

Goalpost shifting much eh? Actually 8C Ryzen is just 1.6 times faster versus 4C Skylake coupled with 2133 memory. So IPC is in fact not that great?

And vector triad is low on computation and high on memory bw / caching / prefetch requirements, AMD's 512KB L2 is certain playing huge part here. One has to wonder how 1MB of L2 in Skylake-X and AVX512 would do? 2-3 Cores of Skylake-X would be surely be enough to match 8 of Ryzen's.

The moment you exceed 4 cores on Ryzen you hit inter-CCX penalties, which is reflected here.

In general, in HPC you hit bandwidth constraints much before you hit computation constraints. This is even more relevant because Broadwell and Haswell fall significantly short of maximizing the theoretical 64B/cy L1 bandwidth in triad-like benchmarks.

Y-cruncher's author has said that the gains from AVX512 over AVX2 is usually 20-40 percent.

Yet you claim that 2 Skylake-X cores with AVX512 is worth 8 Zen cores.

 

[caswow]

Not really. Same as n Athlon days. AMD won't have enough fab capacity. Intel would just need to take a small hit on margins but biggest piece of x86 would still be intel. Even in P4 days intel had more than 50% market share.

https://www.youtube.com/watch?v=osSMJRyxG0k

you sound like an intel exec. 🙄

 

[tamz_msc]

When I log on and see a ton of new messages I get excited. Then when I look I see all this AMD vs. Intel stuff and I face palm. The continuous, persistent, neurotic need for certain AMD enthusiasts to come into Intel threads, cherry-pick and glorify their evidence astonishes me. Excellent idea, to have separate threads and everybody will be happy.

You should have seen what this thread was like before your join date then, when it was first revealed that Skylake-X would have a revamped cache hierarchy. How it will be so fast that everything will be left in the dust.

Of course we now know that turned out.

 

[SocialJusticeWarrior]

Skylake X turned out to be the fastest processors available.

 

[TheF34RChannel]

You should have seen what this thread was like before your join date then, when it was first revealed that Skylake-X would have a revamped cache hierarchy. How it will be so fast that everything will be left in the dust.

Of course we now know that turned out.

Why are you here? Seriously, no hate intended, why are you here?

 

[JoeRambo]

Yet you claim that 2 Skylake-X cores with AVX512 is worth 8 Zen cores.

2-3 fixed it with my actual claim* for You.

* based on y-Cruncher 30s vs Ryzen 100s numbers on their page.

 

[DrMrLordX]

Compared to that, Intel currently has no alternative high perf CPU design, is stuck on 14nm, with 10nm having worse perf characteristics. Much harder to turn things around.

It is harder, but I think we'll see Intel break through their rut on process tech sometime in the next year. They've spent a lot of time on 10nm which appears to be their next 90nm moment, of sorts. As for alternative high performance CPU designs . . . I dunno. In terms of what they've put into production, you're right. They don't have a PIII to lean on as a backup. And I don't think they're going to roll out the next iteration of Netburst either (or are they? dun dun dunnnnn). But none of us know what they've had in the lab that has never seen the light of day.

Maybe nothing at all.

Intel can just keep incrementally improving like they have and stay ahead of AMD, they don't need a revolution to save them.

AMD aside, what about Apple? Intel wants to keep pushing Core into lower and lower power envelopes to compete in that convergence sector where tablets and 2-in-1s compete with one another. There's the ever-present threat of server ARM looming in the shadows. Now is not the time for them to act like they have only one competitor, or no competition at all.

 

[PeterScott]

AMD aside, what about Apple? Intel wants to keep pushing Core into lower and lower power envelopes to compete in that convergence sector where tablets and 2-in-1s compete with one another. There's the ever-present threat of server ARM looming in the shadows. Now is not the time for them to act like they have only one competitor, or no competition at all.

I am not suggesting Intel act like it has no competitors.

I am merely answering the hyperbolic nonsense arguments about the threat of AMD destroying Intel because Intel has failed, and needs completely new architecture to save them. That simply isn't grounded in reality.

 

[tamz_msc]

Why are you here? Seriously, no hate intended, why are you here?

To counter people who like to construct dubious narratives because they like to post asymmetrically on Intel and AMD threads.

2-3 fixed it with my actual claim* for You.

* based on y-Cruncher 30s vs Ryzen 100s numbers on their page.

Whose page? Y-cruncher's?

AVX2 is still less than 50% faster on the 5960X@4GHz and 2666MHz memory compared to the 1800X at 3.7GHz and 2133MHz memory. AVX512 scaling is nowhere near it's theoretical expectations.

 

[DrMrLordX]

I am not suggesting Intel act like it has no competitors.

I am merely answering the hyperbolic nonsense arguments about the threat of AMD destroying Intel because Intel has failed, and needs completely new architecture to save them. That simply isn't grounded in reality.

I don't know how many people think AMD is going to "destroy" Intel. If anyone is going to destroy Intel, it will be Intel first and Apple second. Their path to salvation from THAT particular future is to continue innovating, rather than attempting to stay ahead of AMD through iterative improvements.

 

[PeterScott]

I don't know how many people think AMD is going to "destroy" Intel. If anyone is going to destroy Intel, it will be Intel first and Apple second. Their path to salvation from THAT particular future is to continue innovating, rather than attempting to stay ahead of AMD through iterative improvements.

Really? Where does Apple compete with Intel, and what does Intel need to do remain competitive with them?

 

[JoeRambo]

AVX2 is still less than 50% faster on the 5960X@4GHz and 2666MHz memory compared to the 1800X at 3.7GHz and 2133MHz memory. AVX512 scaling is nowhere near it's theoretical expectations.

First of all i was talking specifically about about Skylake-X taking 2-3 cores to match Ryzen's 8, second, if you actually look at that page ( and not just randomly pull convienient numbers that suit Your goalshifting habits), You will see 96.368s for Ryzen @ 3.7 and 30.06s for 8 cores of Skylake-X. It takes some consideration and extrapolation of scaling, but 3 Skylake-X cores @ same clock more than match Ryzen's eight, and therefore my claim is correct?

 

[tamz_msc]

First of all i was talking specifically about about Skylake-X taking 2-3 cores to match Ryzen's 8, second, if you actually look at that page ( and not just randomly pull convienient numbers that suit Your goalshifting habits), You will see 96.368s for Ryzen @ 3.7 and 30.06s for 8 cores of Skylake-X. It takes some consideration and extrapolation of scaling, but 3 Skylake-X cores @ same clock more than match Ryzen's eight, and therefore my claim is correct?

Wow, you pick the results in the write-up detailing the throttling issues on Skylake-X and then compare it to the Ryzen result in a different page that is marked specifically as a reference for benchmarks, and not only that - the computation that you mention is a different test and uses an algorithm that is completely different from what the default y-cruncher runs - Chudnovsky's algorithm vs BBP formula.

I'll let that slide.

Do you have *actual* examples of AVX-512 giving the purported uplifts over 256b and 128b versions? Or did Intel invent it only to calculate digits of irrational numbers?

 

[Phynaz]

To counter people who like to construct dubious narratives because they like to post asymmetrically on Intel and AMD threads.

Posting false information is against the forum rules and should be reported for moderation using the report button. This is an Intel topic, can you at least try to keep it to Intel?