AMD Naples preview, benchmarks, prices and discussion

[raghu78]

I am creating this thread to start off a place where we can share any Naples related information. Today is AMD's preview of Naples.

https://www.computerbase.de/2017-03/amd-naples-cpu-benchmarks/
https://videocardz.com/67086/amd-previews-naples-server-processor-coming-q2-2017

AMD Zen's design seems to be server focussed and so I am curious to see how Naples fares against the current Broadwell-EP and how its going to stack up against Skylake-EP.

 

[krumme]

AMD loads Naples, Anandtech is booting Johan
http://www.anandtech.com/show/11183...aples-cpus-for-1p-and-2p-servers-coming-in-q2

 

[SketchMaster]

This is a good start, but I'll really get excited when we know power and feq stats.

 

[William Gaatjes]

16 dimms over 8 memory channels and 128 pcie lanes and all control signals and power planes and ground. I am curious how much pins the mcm package has underneath..
That will be a big socket...

 

[ksec]

I expect this to be quite popular for VPS / Cloud instances. Lots of Cores and Lots of Memory in 1U. ( Lets hope they drop the name Opteron)

 

[dark zero]

I am creating this thread to start off a place where we can share any Naples related information. Today is AMD's preview of Naples.

https://www.computerbase.de/2017-03/amd-naples-cpu-benchmarks/
https://videocardz.com/67086/amd-previews-naples-server-processor-coming-q2-2017

AMD Zen's design seems to be server focussed and so I am curious to see how Naples fares against the current Broadwell-EP and how its going to stack up against Skylake-EP.

Seems that Broadwell-EP is gonna get murdered due low chip quality, but Skylake EP will retake the crown.

 

[bigboxes]

So, when can we expect actual SKUs, pricing, power and freq info? Also, what socket are we expecting?

 

[biostud]

according to computerbase it is +4000 pin.
https://www.computerbase.de/2017-03/amd-naples-cpu-benchmarks/

 

[bjt2]

According some videos I saw, the 128 PCIex lanes are split in 64 true PCIex and 64 links for infinity fabric (with same low level electrical protocol, but different, coherent, high level protocol) for connecting to other CPU in second socket.

So

1P Naples: 8 memory channel 16 dimms @ 2400MHz plus 128 PCIex 3.0 lanes (versus 12 1866 DIMMs and 40 PCIex lanes of concurrent INTEL)
2P Naples: 2x8 MC and 32 DIMMS @ 2400MHz plus 2x64 PCIex 3.0 lanes, with 2x64 PCIex lanes adapted as infinity fabric for socket linking (versus 24 1866 DIMMs and 80 PCIex lanes of concurrent INTEL)

According to such videos the speed of IF links is similar to PCIex 3.0. Old slides promised 100GB/s for link BW. 10GBs PCIEx 3.0 64x is 80GB/s. So IF has higher clock?

 

[lefty2]

From the specs, it looks like AMD will score a home run on this one.
There's just one thing, will they have the actual servers ready to roll on launch, or do OEMs like Dell and HP need another few months to incorporate Naples into their product?

 

[FIVR]

Those benchmarks must be terrifying for intel's CTO... 2.5X performance with lower platform cost and more memory and I/O

 

[Mechanical Man]

Those benchmarks must be terrifying for intel's CTO... 2.5X performance with lower platform cost and more memory and I/O

Not to downplay AMD's nice result. But it's pretty clear that most of it is due to memory bandwith. I do believe that there is lot of usage scenarios where it is the most important thing for performance, but don't get hyped up and expect this to be same for compute intensive workloads. This must be taken as it is. Best case scenario at memory bw limited test case.

 

[Atari2600]

Lets hope they drop the name Opteron

Why?

It's a name with recognition in the industry...

 

[Atari2600]

Best case scenario at memory bw limited test case.

Probably.

But 2.5x is a large margin to pull back to parity!

(and parity is of no use to Intel given their pricing structure)

 

[Glo.]

Not to downplay AMD's nice result. But it's pretty clear that most of it is due to memory bandwith. I do believe that there is lot of usage scenarios where it is the most important thing for performance, but don't get hyped up and expect this to be same for compute intensive workloads. This must be taken as it is. Best case scenario at memory bw limited test case.

Second reason is that AMD marketing team likes to nitpick in worst possible way benchmarks.

 

[nathanddrews]

I hope that AMD can reuse the tweaks and bug fixes of the Ryzen launch to make the Naples rollout smoother.

Anyone else unable to resist saying "nipples"?

 

[imported_jjj]

Not to downplay AMD's nice result. But it's pretty clear that most of it is due to memory bandwith. I do believe that there is lot of usage scenarios where it is the most important thing for performance, but don't get hyped up and expect this to be same for compute intensive workloads. This must be taken as it is. Best case scenario at memory bw limited test case.

The entire point was to show that more memory BW matter in some cases, nothing more. Some folks didn't quite got that from the start but that's on them.
The perf and efficiency are already known anyway. What is less clear is how it scales to 4x8C and then 2 sockets.

 

[Mechanical Man]

The entire point was to show that more memory BW matter in some cases, nothing more. Some folks didn't quite got that from the start but that's on them.
The perf and efficiency are already known anyway. What is less clear is how it scales to 4x8C and then 2 sockets.

That is what i said. Its just that some guys see the "2,5X" and get hyped up how freaking fast and performant it is, then when they see the real performance will go all apes that AMD somehow defrauded them. That already happened with Ryzen line.

 

[bigboxes]

That is what i said. Its just that some guys see the "2,5X" and get hyped up how freaking fast and performant it is, then when they see the real performance will go all apes that AMD somehow defrauded them. That already happened with Ryzen line.

So, you just created an account just to poo pah AMD? Your just as bad as the new accounts created just to pump up Ryzen. If you think that Ryzen is a failure then your just a troll. Why don't we wait for the release to pass judgement.

 

[Mechanical Man]

So, you just created an account just to poo pah AMD? Your just as bad as the new accounts created just to pump up Ryzen. If you think that Ryzen is a failure then your just a troll. Why don't we wait for the release to pass judgement.

Cant you comprehend what i said? I just said that don't go overlooking what the test is showing. It shows limited case and tells nothing else. I also did not say anywhere that Ryzen is somehow bad. I said some ppl got overhyped and then reacted as it was AMD's fault when it was them who did not see what they were looking at.

Same thing you are doing thinking i somehow diss AMD. I do not, read again.

 

[William Gaatjes]

according to computerbase it is +4000 pin.
https://www.computerbase.de/2017-03/amd-naples-cpu-benchmarks/

I wonder how big that mcm will be.
Around 90mmx 80mm, i would guess...

 

[Atari2600]

That is what i said. Its just that some guys see the "2,5X" and get hyped up how freaking fast and performant it is, then when they see the real performance will go all apes that AMD somehow defrauded them. That already happened with Ryzen line.

To be honest, if it ends up "just" 1.10X faster than Intel's, then the only people who will be underwhelmed will be idiotic fanbois whose opinion's are best ignored anyway.

 

[SketchMaster]

Really, all AMD has to do to impress the market is meet intel's performance & watt for 1/2 the cost; anything after that is just gravy.

 

[nathanddrews]

Personally, I can't wait to see Naples 720p gaming benchmarks... Too soon?

 

[Mechanical Man]

To be honest, if it ends up "just" 1.10X faster than Intel's, then the only people who will be underwhelmed will be idiotic fanbois whose opinion's are best ignored anyway.

I do expect per socket performance to be pretty much greater on same power or about par with greatly better power use (with some outliers).

I do think it will be a hit on big part of server customers based on what we saw on ryzen, specially Stilts technical review.

I just don't want to see the all out hyping that happened with ryzen, then when it fails to deliver that 2,5X performance, ppl get dissapointed while it most likely still is better than intels xeons, atleast where the performance per socket matters.