AMD Naples coming Q2 2017

[MakeAMDGreatAgain]

Just now, Fetzie said:

Enterprise (as in the people that will actually buy these processors) doesn't listen to the people that might bitch about how a benchmark is unfair.

They are obviously not making a smart decision when not taking benchmark baselines into account then.

[Fetzie]

Just now, MakeAMDGreatAgain said:

They are obviously not making a smart decision when not taking benchmark baselines into account then.

You misunderstand me.

 

A company is shopping for new hardware, and has two options.

 

22 core with HT

32 core with HT

 

They don't give a shit about how well the 32 core one performs when you disable 10 of its cores because they can't buy that. They want a comparison between the two CPUs AS THEY CAN BE BOUGHT. Arguably, the test with equal cores is useless, because it isn't representative of the product.

[BachChain]

8 minutes ago, MakeAMDGreatAgain said:

They are obviously not making a smart decision when not taking benchmark baselines into account then.

You mean the smart decision to chose the product that performs better in their workload without caring about something as meaningless as Intel's feelings?

[MakeAMDGreatAgain]

5 minutes ago, Fetzie said:

You misunderstand me.

 

A company is shopping for new hardware, and has two options.

 

22 core with HT

32 core with HT

 

They don't give a shit about how well the 32 core one performs when you disable 10 of its cores because they can't buy that. They want a comparison between the two CPUs AS THEY CAN BE BOUGHT. Arguably, the test with equal cores is useless, because it isn't representative of the product.

Exactly my point though. Just because of "AS THEY CAN BE BOUGHT" doesn't make the benchmarks not unfair.. There has to be a baseline somewhere.

[2FA]

2 minutes ago, MakeAMDGreatAgain said:

Exactly my point though. Just because of "AS THEY CAN BE BOUGHT" doesn't make the benchmarks not unfair.. There has to be a baseline somewhere.

You think it's unfair because you're stuck in the consumer mindset where single threaded appiications matter. The enterprise landscape solely cares about how much multithreaded performance a CPU can give regardless of core count or clockspeed. 

[Fetzie]

5 minutes ago, MakeAMDGreatAgain said:

Exactly my point though. Just because of "AS THEY CAN BE BOUGHT" doesn't make the benchmarks not unfair.. There has to be a baseline somewhere.

And the baseline is the spec of the CPU. Which is 32 physical cores/ 64 logical cores in single-socket mode and 64 cores / 128 logical cores in dual-socket mode. Not "32 physical cores/ 64 logical cores in single-socket mode (ten physical cores / 20 logical cores disabled)" and "64 cores / 128 logical cores (20 physical cores / 40 logical cores disabled) in dual-socket mode".

 

The demo with equal cores and memory speed is completely irrelevant. It has nothing to do with the product. The "baseline" that you want is the full-fat, no holds barred test at the end of the video. That's the one that people care about. The other is just teabagging Intel.

 

These aren't gaming benchmarks, where the core usage diagram looks something like this:

 

These are applications that can have the workload spread out over thousands of cores, in hundreds of blades. That is the target market for ZEN/Naples. Nobody is playing WoW, GTA5 or CS on these CPUs.

[Wolther]

9 minutes ago, MakeAMDGreatAgain said:

Exactly my point though. Just because of "AS THEY CAN BE BOUGHT" doesn't make the benchmarks not unfair.. There has to be a baseline somewhere.

It's pretty fair. That's like saying an athlete shouldn't play a sport because it's too unfair that he runs faster than everyone else, even though they're in the same league. 

They took the 2 best cpus from both sides and compared them, not understanding why this comparison shouldn't have been made. They showed that they have more cores, and that's why they won.

[MakeAMDGreatAgain]

Just now, DeadEyePsycho said:

You think it's unfair because you're stuck in the consumer mindset where single threaded appiications matter. The enterprise landscape solely cares about how much multithreaded performance a CPU can give regardless of core count or clockspeed. 

No. I think it's unfair because the benchmark QUITE LITERALLY was 44 vs 64 cores. My mindset is irrelevant here.

No reason to show that benchmark other than to show off and belittle Intel. When in all actuality, it actually doesn't because the benchmark by nature is not even a fair one.

 

Might look cool during the presentation, but in the real world that's not the case.

[WkdPaul]

8 minutes ago, MakeAMDGreatAgain said:

Exactly my point though. Just because of "AS THEY CAN BE BOUGHT" doesn't make the benchmarks not unfair.. There has to be a baseline somewhere.

Your argument makes no sense whatsoever, comparing products, even if slightly different, is and always was a way to see if you got what you paid for.

 

Comparing CPUs, GPUs that are in the same price point but with different core count and/or memory has always been and will always be.

 

Comparing dissimilar products that are in a certain price bracket, or have a specific market makes sense, the same goes with all other products out there, fridges, ovens, cars etc...

[MakeAMDGreatAgain]

4 minutes ago, wkdpaul said:

Your argument makes no sense whatsoever, comparing products, even if slightly different, is and always was a way to see if you got what you paid for.

 

Comparing CPUs, GPUs that are in the same price point but with different core count and/or memory has always been and will always be.

 

Comparing dissimilar products that are in a certain price bracket, or have a specific market makes sense, the same goes with all other products out there, fridges, ovens, CAS, etc...

I already know the whole "you get what you pay for" shtick. But what I am saying is, even that is true (I AGREE, THE first AMD Benchmark tea-bagged Intel in an equal core comparison). This DOESN'T make the 44 vs 64 benchmark fair. Just because something performs better at a better price point is absolutely IRRELEVANT to the validity of a benchmark.

 

So please stop trying to distort native benchmarks just because it's a "fair price comparison".  

[Fetzie]

4 minutes ago, MakeAMDGreatAgain said:

No. I think it's unfair because the benchmark QUITE LITERALLY was 44 vs 64 cores. My mindset is irrelevant here.

No reason to show that benchmark other than to show off and belittle Intel. When in all actuality, it actually doesn't because the benchmark by nature is not even a fair one.

 

Might look cool during the presentation, but in the real world that's not the case.

The benchmark that actually belittles Intel is the one you think is the fair comparison. It is saying "We disabled a third of the cores and reduced memory clocks by 23% and we still beat you NANANANANA *sticks tongue out and blows a raspberry*".

 

Just now, MakeAMDGreatAgain said:

I already know the whole "you get what you pay for" shtick. But what I am saying is, even that is true (I AGREE, THE first AMD Benchmark tea-bagged Intel in an equal core comparison). This DOESN'T make the 44 vs 64 benchmark fair. Just because something performs better at a better price point is absolutely IRRELEVANT to the validity of a benchmark.

 

No, the benchmark clearly showed that the Naples CPUs are better in this specific scenario than the Intel Xeons were. If you have a dataset similar to this benchmark that can be processed in half the time, then that is important and valid information. It doesn't matter what you think is fair.

[2FA]

1 minute ago, MakeAMDGreatAgain said:

I already know the whole "you get what you pay for" shtick. But what I am saying is, even that is true (I AGREE, THE first AMD Benchmark tea-bagged Intel in an equal core comparison). This DOESN'T make the 44 vs 64 benchmark fair. Just because something performs better at a better price point is absolutely IRRELEVANT to the validity of a benchmark.

So they shouldn't benchmark at all? What are you even trying to get at here? 

[MakeAMDGreatAgain]

Just now, Fetzie said:

The benchmark that actually belittles Intel is the one you think is the fair comparison. It is saying "We disabled a third of the cores and reduced memory clocks by 23% and we still beat you NANANANANA *sticks tongue out and blows a raspberry*".

That's your personal opinion though, atleast it was the equal amount of cores as compared to the 44 vs 64? But that's none of my business, right?

[WkdPaul]

Just now, MakeAMDGreatAgain said:

I already know the whole "you get what you pay for" shtick. But what I am saying is, even that is true (I AGREE, THE first AMD Benchmark tea-bagged Intel in an equal core comparison). This DOESN'T make the 44 vs 64 benchmark fair. Just because something performs better at a better price point is absolutely IRRELEVANT to the validity of a benchmark.

??? except that's the whole point!

By that logic, each and every CPU and GPU benchs are now irrelevant?

[BachChain]

1 minute ago, MakeAMDGreatAgain said:

I already know the whole "you get what you pay for" shtick. But what I am saying is, even that is true (I AGREE, THE first AMD Benchmark tea-bagged Intel in an equal core comparison). This DOESN'T make the 44 vs 64 benchmark fair. Just because something performs better at a better price point is absolutely IRRELEVANT to the validity of a benchmark.

 

So please stop trying to distort native benchmarks just because it's a "fair price comparison".  

"benchmarks are only valid if the products are identical in literally every aspect"

[MakeAMDGreatAgain]

 

Just now, BachChain said:

"benchmarks are only valid if the products are identical in literally every aspect"

Yep, 100% agree, especially when it comes to the amount of cores/threads.

[RagnarokDel]

7 hours ago, AnonymousGuy said:

Also, launching a server part < 3 months after launching the desktop part indicates it's not really a server part with a ground-up server architecture.  4 desktop parts glued together does not a server make.

Or they were working on it all along. It has 8x channels for memory, it's not the same thing but what was the issue with Ryzen exactly? It outperforms a 6900k for half the price. The issues with Windows 10 scheduler are not on AMD's hands but Microsoft's. 

[WMGroomAK]

16 minutes ago, Fetzie said:

A company is shopping for new hardware, and has two options.

 

22 core with HT

32 core with HT

 

They don't give a shit about how well the 32 core one performs when you disable 10 of its cores because they can't buy that. They want a comparison between the two CPUs AS THEY CAN BE BOUGHT. Arguably, the test with equal cores is useless, because it isn't representative of the product.

While I agree for the most part that a company shopping for what is available at the moment Naples release will probably go with the 32 core over the 22 core, one of the big issues is really going to come down to timing of release...  If AMD rushes this release, they may hit some of the same stumbling blocks that they've hit with Ryzen CPUs and software/hardware support.  If they hold off too long on their release, they may give time for Intel to get the upcoming Skylake-EP Xeons out, which may prove just as competitive (except for price point?)(Hate this many qualifiers in a sentence.)  

[Fetzie]

Anyway. What is really interesting to me is how it performs when you are clustering 1000 of these machines together. Given that is the more likely use-case for this kind of processor.

[leadeater]

5 hours ago, MakeAMDGreatAgain said:

Red herring, not relevant unfortunately. Bottom line is 44 vs 64 cores is unfair

Ah no, why does core count matter, price does. What if 64 AMD cores is the same cost, what if it's less. Who cares if its 44 cores vs 64 cores, nobody who is going to buying these which includes me.

 

Fair comparisons is a misnomer, no such thing in this market so don't even bother to expect it.

[leadeater]

4 hours ago, Fetzie said:

You misunderstand me.

 

A company is shopping for new hardware, and has two options.

 

22 core with HT

32 core with HT

 

They don't give a shit about how well the 32 core one performs when you disable 10 of its cores because they can't buy that. They want a comparison between the two CPUs AS THEY CAN BE BOUGHT. Arguably, the test with equal cores is useless, because it isn't representative of the product.

Thank you exactly, someone who is actually thinking about how people buy products and isn't obsessing over benchmarks and non real world scenarios and configurations.

[MakeAMDGreatAgain]

1 minute ago, leadeater said:

Ah no, why does core count matter, price does. What if 64 AMD cores is the same cost, what if it's less. Who cares if its 44 cores vs 64 cores, nobody who is going to be these which includes me.

 

Fair comparisons is a misnomer, no such thing in this market so don't even bother to expect it.

I agree with you on the first part. Price does matter, but just because the price:performance ratio is better (as the first benchmark tea-bagged Intel), doesn't mean that the benchmark could NOT be "unfair".  I'm just saying we need to look at the benchmarks from an objective baseline. The initial first benchmark was good. Equal cores, an equal benchmark. And AMD won. That gets me hyped.

 

Then, seeing a 44 vs 64 core benchmark right after, and AMD still winning creates a feeling of something fishy is going on because why do they need to show that they beat Intel with 20 more cores? Of course they will.  

 

There's a point where you don't kick people when they are down. That saying applies in the technical world as well.  Beat them fair and square as AMD did in the first benchmark. I personally think the continue tea-bagging was unwarranted but obviously we all have differing opinions, to each his own

[leadeater]

25 minutes ago, MakeAMDGreatAgain said:

Then, seeing a 44 vs 64 core benchmark right after, and AMD still winning creates a feeling of something fishy is going on because why do they need to show that they beat Intel with 20 more cores? Of course they will.  

 

There's a point where you don't kick people when they are down. That saying applies in the technical world as well.  Beat them fair and square as AMD did in the first benchmark. I personally think the continue tea-bagging was unwarranted but obviously we all have differing opinions, to each his own

But that's not how it works, AMD made a design for 32 cores so that is what they have. They also did a lot of work on getting the per core performance increased which they achieved, they have no idea from the outset how well they can get their architecture to but they can set goals.

 

One of the design goals is pretty evident that 32 cores is the full design limit, that is what they were doing from the outset. This has nothing to do with anything other than they have a 32 core CPU and need to show it. You don't not show it just because the competition doesn't have a 32 core part, too bloody bad this is what AMD has plain and simple.

 

Not showing the full 64 (32) cores is by far a much worse move and would lead to so many questions being asked to the point that it would have been a disservice to AMD to do so. What if 64 (32) cores performs worse in some tests since the 44 core bench used higher clocks due to disabled cores? What if there are L1/L2/L3 cache issues and only using 44 hides this?

 

Show what you have not what the competition has. Playing to the competition is a losing move, look forward not back.

[leadeater]

It's probably also worth keeping in mind that these benchmarks been shown are specifically to show Naples at it's best, much like Ryzen proof will be launch reviews where most likely it's going to be on par or slightly less than Intel's 44 cores in common workloads today.

 

The argument of things being optimized for Intel is far more applicable in the server market, particularly when it comes to virtualization as everything is very tailored to Intel's technologies and not AMD's. 

[Liltrekkie]

Does this mean there will be a 32 gamers 1 CPU video????

 

All joking aside, if AMD can actually pull this off, be power efficient with this many cores, and provide all these benefits WITHOUT the 9 million fixes Ryzen will need, then they just may score a win and switch a lot of data centers to AMD. 

 

There's 4 things enterprise markets care about. Power efficiency, multi core performance, reliability, save rack space.

 

I will gladly switch each and every one of my servers to run the top Napels chip, I would save money by not having to use so many servers due to the higher core count. 

 

I'm living the dream.