AMD Offers Full Details and Performance of Zen-Based Naples Server Platform (hothardware.com) 138
MojoKid writes: AMD lifted the veil this morning on architecture details and performance expectations of its next generation Zen-based server platform, codenamed Naples. Naples is an up to 32-core, 64-thread variant of Zen, targeted at enterprise and data center applications. The processors will feature eight-channel DDR4 memory controllers (with up to 16 DIMMs attached per CPU), with support for up to 4TB of memory and 128 lanes of on-chip PCI Express connectivity. In a 2P (dual processor/dual socket) configuration, Naples offers up to 64 physical cores (128 threads), access to 32 DIMM slots, and aggregate 16 memory channels. Versus a 2P Intel Xeon E5-2699A V4 based server, the 2P Naples setup ends up with double the memory channels, a higher total memory capacity, more cores (20 more physical cores, 40 more threads), and 48 more available PCI Express lanes. AMD's performance comparisons at its tech day event pit a 2P Naples server with 512GB of DDR4 RAM up against a 2P Intel Xeon E4-2699A V4 configuration with 384GB of RAM. The Naples system had a higher memory capacity and that memory was clocked much higher too -- 2400MHz versus 1866MHz. The Naples system has more cores, and with SMT on, can ultimately process more threads as a result. The AMD Naples system also has double the memory channels, further improving peak memory bandwidth. In its demos, AMD used a seismic analysis workload, which involved multiple iterations of 3D wave equations. According to AMD, the test taxes the entire system, including CPU cores, memory and I/O. In this demo, the AMD server system completed equations roughly 2.5x faster than the dual-socket Intel Xeon server. Expected price points weren't given, but Naples processors and servers should be available in Q2 this year.
But can it... (Score:2)
Re: (Score:2)
Re: But can it... (Score:2)
Maybe this monster CPU could finally handle 640x480?
Re: (Score:2)
Gonna get one of these and trick it out to run my toaster.
Quad Sockets: (Score:2)
The article states you can run dual sockets, but will this be supported in quad sockets (provided someone makes a board to support it), or are you also limited by the way it used PCI Express lanes?
Re: Quad Sockets: (Score:5, Insightful)
There is a massive benefit in situations such as a lot of parallel tasks using the same dataset shared in memory. I/O only matters with getting things in and out and a lot of tasks are CPU bound instead of I/O bound. The demo task they used, processing of seismic data, is a good example of that sort of task. Think of things like applying the exact same filter to a few million audio tracks for an idea of how parallel the tasks can be, and think of mixing based on location and time of millions of audio tracks to get an idea of how good it is to do it all on the same machine with shared memory.
Re: (Score:3)
"i cant wait to apply this filter to a million audio tracks." ... said no one ever
Communications network (e.g. phone) surveillance, seeking keywords. Yup, it gets done. A lot.
Re: (Score:1)
Back when I had to interact with servers running those platforms, there were some very strange interactions between MySQL and the NUMA arr
Re: (Score:1)
Re: (Score:3, Informative)
Fun fact: Quad sockets are SUPER-RARE.
Go ahead. Try and Google for a quad motherboard. They haven't made them since like... the Athlon 64-era Opterons. The boards are MASSIVE. You can't fit 128 DRAM slots on a single board. (And who wants to BUY that many?) There's too many traces and the board becomes super expensive from the extra routing layers. The board is also going to flex under any kind of weight, and the larger the footprint the more important the mounting becomes (going from casual "bolt it to the
Re: (Score:3)
Really? You should tell all the people that are making quad socket servers today.
Maybe you should try following your own advice.
Re: Quad Sockets: (Score:3, Insightful)
Still rare, still expensive, still fragile, still hard to service and upgrade, still not worth it except for some edge cases.
Quad (and octa) core servers are the moon rockets of IT. Yes, they do exist. No they are not common. Only for very, very special missions.
He said 'rare' not 'nonexistent'
Re: (Score:2)
OP specifically said quad socket motherboards hadn't been made since the Athlon 64 era (10+ years ago). I provided two examples of current generation quad socket servers that are on sale today.
Sure, Google and Facebook probably don't run quad socket servers (they're not built for that kind of thing), but your bank probably does.
Re: (Score:2)
His bank probably runs on mainframes.
Re: (Score:3)
Still rare, still expensive, still fragile, still hard to service and upgrade, still not worth it except for some edge cases.
I bought a bunch about 6 years ago (not the 10 the OP claimed). They were quad socket Opterons with a quarter terabyte of RAM. Rarer than 2P, but available from a well known manufacturer (SuperMicro) via vendors already on our systems. Not rare. They weren't expensive either, well, not in terms of bang for buck doubly so considering the space taken up (they're about double the density
Re: (Score:2)
Knock yourself out:
http://www.thinkmate.com/syste... [thinkmate.com]
Got one of their dual Opteron 16-core (2.8ghz) machines right now, the performance is ~20% better than our dual 8-core (16 thread) Intel machine for our workloads. I'm dying to grab a 64 core machine that can handle 128 threads. 2x that? It would be awesome if we could afford it.
Sam
Re: (Score:2)
Cores or sockets? If you actually mean cores just about every Xeon processor made this decade is quad core or better. Most i5 and i7s are quad or better as well. On the AMD side has there been a model this decade that's not quad core or better?
He actually said "SUPER-RARE". Having boards/servers readily available from Dell, Supermicro, HP,
Re: (Score:2)
HP DL370 - industry standard 2S Xeon socket server
HP DL570 - industry standard 4S Xeon socket server
Sure, getting hold of a mass produced 4S motherboard is difficult, and these 4S servers aren't cheap but if you are intending on buying $20k of processors to run on it, the difficulties can also be solved by throwing money at the problem.
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
The biggest cost is being hit with per core licensing, I'd love to throw more hardware at our system if it'd stay constant. But a doubling in cores deals a doubling in license costs and rarely a doubling in real world performance so unless you really need it all in one package.... no.
Re: (Score:2)
That all depends on what software you're running, only certain things are licensed per core... Some things are licensed per socket, while a lot of licenses are per machine or entirely unrestricted.
Re: (Score:2)
Erm, I'm pretty sure the 2699 is the top of the line for their Xeon DPs. At worst it is the 3rd best from what I can tell, and no one has benchmarks that I can find of the two newer processors that might replace it, the E7-8890 and E7-8894. As they were probably able to borrow or rent the 2699's from someone and they weren't going to blow 20,000+ to purchase equipment for a single benchmark, it's unsurprising they didn't go for the two newer chips.
Re: (Score:2)
Incidentally, even assuming that the two Kaby Lakes are better, they're almost certainly not better enough over the 2699 to make up the performance deficit given the Kaby Lake chips' price difference to the 2699.
Re: (Score:1)
pci-e lane madess next to intel (Score:2)
now a 1s system with 16 core 32 threads with 64 lanes + more for storage? is better then what Intel has can make for a killer 3-4 way SLI system with room for 10-GIG-e more then 1 pci-e storage disk / etc.
The mac pro should move the this give it the lanes (Score:2)
The mac pro should move the this give it the lanes needed for 2 storage cards + 2 video cards + 6-10 TB 3 buses + dual 10 GIG-E.
Re: (Score:2)
While we're dreaming, I'd ask for a Mac Pro in a standard tower chassis that can handle standard PCIe cards. And a pony. But instead we'll keep using Linux boxes.
Now OVH... (Score:2)
just has to get these and rent some cheap AMD boxes. I'll take one to run my Xonotic server...
But (Score:2)
It runs slow for games.Everyone knows at 800 X 600 resolution gaming is the most important thing when evaluating a CPU.
Finally, some Xeon competition (Score:5, Interesting)
The Xeon line has been pretty lame for the last couple of years. It's nice to see some competition.
Re: (Score:2)
The Xeon line has been gimped forever. Can't do a 2S 4GPU config. Memory architecture limitations makes it impossible. AMD, OTOH, could handle it no problem.
Re: (Score:1)
Obligatory Post for /. Old Farts (Score:4, Funny)
Imagine a Beowulf Cluster of these processors!
Re: (Score:2)
And covered in hot grits...
Re: (Score:1)
Re: (Score:2)
Considering how the CPU market has stagnated the last 10 years I can believe that, yes.
Re: (Score:2)
2. ???
3. Profit!
Price? (Score:2)
Anyone?
Re:ECC? (Score:5, Interesting)
The Ryzen CPUs support ECC, but the parts haven't gone through the server level validation for it. It's up to motherboard manufacturers whether or not to support it, and I expect we'll see "home server" 300 series boards with ECC support in the near future.
The Naples parts will absolutely support ECC. It's pretty much mandatory for server parts these days.
Re:ECC? (Score:4, Interesting)
The Ryzen CPUs support ECC, but the parts haven't gone through the server level validation for it. It's up to motherboard manufacturers whether or not to support it, and I expect we'll see "home server" 300 series boards with ECC support in the near future.
Historically one of the huge advantages of AMD is that ECC has been available on much lower end parts than Intel. I hope that continues.
Re: (Score:3)
Ryzen supports ECC. Even on the desktop, if the mobo supports it.
Re:ECC? (Score:4, Interesting)
The article doesn't mention support for error correction for the RAM. If a bit gets flipped somewhere in a massive research project running on such a system with 4 to 8 TB RAM, the entire project could be ruined.
ECC, anyone?
All the AMD processors support ECC. You need to explicitly criple it like Intel does not to. It need to be certified with the motherboard though, and most consumer motherboards don't do that, and many also don't provide the tools to access if any error-corrections are done.
ECC and VM support (Score:2)
As a long-term AMD user, I was blown away when a system was "upgraded" from a dual-core to quad-core chip started behaving even worse with VM's, which led me to discover that the Quad-core chip didn't support VT.
Looking into it more, I discovered that Intel does this a *lot*, with sprinkles of support for this and that across a huge range of chipsets. With AMD, I haven't found a chip without AMD-V/VT-x in forever.
Now this was a bit of an older CPU for sure, but it seems that even newer Intel CPU's one has t
Re: (Score:2)
Looking into it more, I discovered that Intel does this a *lot*, with sprinkles of support for this and that across a huge range of chipsets. With AMD, I haven't found a chip without AMD-V/VT-x in forever.
Now this was a bit of an older CPU for sure, but it seems that even newer Intel CPU's one has to be very careful that options like VT-D and others aren't neutered out.
I sometimes pay a little extra to make things easier for me. But with Intel, they make CPU selection harder even when you pay more >.<
Re: (Score:2)
Their AM1 boards do not. They saved some COGS by only wiring up 64 of the 72 data lines. Very disappointing!
Strategy masterpeice (Score:1)
Always our new vs your old (Score:2, Insightful)
A proper comparison would be between AMD's 2017 cpu Naples against Intel's 2017 cpu E5 V5 not 2016 E5 V4, as is done here.
Re: (Score:2)
Found the Intel fanboy. How do you expect them to compare to a processor you can't actually get?
Re: (Score:2)
More importantly, why would they blow 20,000+ for a single benchmark?
These should be good server CPU's (Score:2)
Based off of what we saw in the Ryzen benchmarks, the new Zen CPU architecture is really good at handling multi-threaded workloads. That doesn't help Joe Sixpack much when he's surfing the web or playing Call Of Duty, but it will help with highly threaded server applications like Java application servers and databases.
Maybe AMD should have done the server CPU's first and then did the desktop models. There is probably more profit margin in the Enterprise market anyway.
Re: (Score:2)
The problem is that AMD never had great adoption in the enterprise market. Before dropping out of the performance market, AMD's enterprise offerings were slim. While the Opteron CPUs (x86_64) were considered decent and worth their money at the time, the bigger issues were the motherboard and chipsets. They had a couple of bad server chipset releases and the number of compatible and available motherboards decreased. This maked it hard to find servers that used AMD server CPUs. This lead to poor adoption rate
Re: (Score:2)
AFAIK they did ok with the K8 servers and their derivatives. The issue is they released the high power consumption K10 processors around the time server people started taking notice about power consumption. Other than some HPC applications like supercomputers the K10 it wasn't that popular and eventually not even there.
Let me get this straight (Score:2)
So let me get this straight, in order to show competing platforms they put it up against a 22 core E5-2699A but then hobbled the Xeon with both LESS RAM and SLOWER RAM? Um... that's not really a very fair comparison now, is it?
Don't get me wrong, I like AMD as much as the next guy and I am very interested in probably making my next home-server build a Ryzen with ECC... but at least compete on a level playing field. The E5-2699A supports DDR4-2400 as well but instead they decided to hobble it with DDR4-1866?
fabric shrink in the washboard abs (Score:2)
There are moments when I understand what other people are griping about.
AMD Prepares 32-Core Naples CPUs for 1P and 2P Servers: Coming in Q2 [anandtech.com]
Sad, Ian, sad.
Re: (Score:2)
No. It's four chips, with 8-cores per chip (i.e. 16 threads per chip), per processor. I assume they're going to use a multi-chip module for that. The thing with multi-chip modules is that you can use a lot faster interconnect than you would otherwise because of the smaller traces and more specialized packaging.
Meanwhile Consumer CPU side is gimped to hell (Score:2)
"The processors will feature eight-channel DDR4 memory controllers (with up to 16 DIMMs attached per CPU), with support for up to 4TB of memory and 128 lanes of on-chip PCI Express connectivity."
Meanwhile, Ryzen on the Desktop has a shitty 16 lanes of PCI-E with an additional 8 possible with the mobo manufacturer adding another bridge. Meanwhile, my FX-9350 has over 30 lanes of PCI-E connectivity, for SLI, multiple M.2 drives, and multiple USB connectivity ports and dual gigabit ethernet.
Ryzen is gimped so
It's an overdue competition... (Score:1)
Their CPUs are way overpriced. Note their 2016 company profits hit $7.5B on $17.5B revenue! REALLY?!
AMD has priced Ryzen, their answer to the i7, at less than half the price with higher performance.
This surely shows that Intel is likely gouging the market; which is unethical monopolistic antitrust behavior.
Re: We keep getting faster processors... (Score:2)
So, in short, if you can pay, you can have?
Re: We keep getting faster processors... (Score:2)
Let me correct that for you: "I dont need faster processors, I need a faster network"
The rest of us have our own hopes and dreams
Re: (Score:2)
That seems to be mostly a problem in the US. In Europe, slow Internet is something you usually find in the countryside, but not in cities. Market failure?
Re:We keep getting faster processors... (Score:5, Insightful)
Chattanooga tried to set up a city owned ISP a few years ago because the local monopoly was terrible, and the state government legislated to prevent them doing it.
That's what happens when money dominates politics.
Re:We keep getting faster processors... (Score:5, Insightful)
Chattanooga did set up a city owned ISP, EPB, the Electric Power Board of Chattanooga. EPB offers some of the fastest, most reliable, Internet service at extremely competitive prices. So when EPB began thinking about expanding to other underserved communities in the Chattanooga area, it was a real threat to the incumbent duopoly: Comcast and AT&T.
The Tennessee legislature couldn't prevent Chattanooga from setting up its own ISP, but what they could do is prevent EPB from expanding their service area beyond the city limits. So areas just outside Chattanooga have some of the slowest Internet service, and it's all thanks to lobbyists for AT&T. Thank you, Tennessee Republican legislature, for protecting the profits of AT&T by protecting them from competition by a city owned utility.
Re: (Score:3)
So market failure caused by anti-competitive politics. And that in the "home of capitalism". Pretty bad. I wonder how much corruption is involved. Corruption kills societies.
Re: (Score:2)
The trick is it isn't corruption. The US has completely legalized what many other countries would call corruption. Companies can just throw money at US politicians or their campaigns in various ways and it's all legal.
Re:We keep getting faster processors... (Score:4, Insightful)
It doesn't magically become not corruption when it becomes legal.
Re: (Score:2)
Indeed. It actually is much worse than illegal corruption. This state of affairs explains a few things though.
Re: (Score:3)
It becomes corruption when politicians pass laws and regulations because of that money.
I dont live in the Seattle area so I dont know which politicians are responsible for granting the Seattle monopolies without some sort of public compensation (such as contractually good internet service) but I do know that if I lived there I would know exactly who was responsible. Its probably the city council thats holding the keys to the franchise
Re: (Score:2)
Nor does it become corruption because money was thrown at politicians.
It becomes corruption when politicians pass laws and regulations because of that money.
You are right about that. But who keeps throwing money at politicians without getting anything back? At the very least, the recipients would need to do full, public disclosure about everything they got and explain conclusively with any decision they make why they were not influenced by that money.
Re: (Score:2)
You are right about that. But who keeps throwing money at politicians without getting anything back?
Corporations wouldn't. Which is why we know for a fact that the problem isnt corporations. The problem is the politicians selling laws and regulations.
Re: (Score:2)
There's plenty of blame to go around. Just because the politicians are corrupt doesn't exonerate those who bribe them, legally or not.
Re: (Score:2)
Just because the politicians are corrupt doesn't exonerate those who bribe them
If you refuse to lobby the law maker but your competition does not refuse to do so...
Everybody but the politicians are exonerated. Everybody but the politicians are victims.
Re: (Score:2)
It is corruption, but legalized corruption. I would call that super-corruption or the like, because the corrupt have managed top make their depravity legal.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
EPB offers some of the fastest, most reliable, Internet service at extremely competitive prices.
Are you including the taxes in those competitive prices or are you just pretending that tax money falls from the sky?
Re: (Score:2)
cities have no place running a business. name one thing that government does right?
Capitalize the first letter of their sentences?
Re: (Score:3)
That seems to be mostly a problem in the US. In Europe, slow Internet is something you usually find in the countryside, but not in cities. Market failure?
Broadband have monopolist tendencies everywhere, but I think the real market failure in the US is because the biggest providers are cable networks. You're basically asking them to saw off the branch they're sitting on. Here in Norway telcos and xDSL was big and power companies were the first to push fiber, cable is also a player but they must remain competitive. They still want to make profit but if they can make more money selling you even faster broadband they'll do it. The mean download speed is now 47 M
Re: (Score:3)
That's off-topic. There's nothing about CPU upgrades that demand a comparison to internet speeds. There are TONS of workloads that don't require LIVE access to a saturated internet connection.
And those that do? BUSINESSES PAY FOR IT.
My brother wasn't overseeing terabytes of medical info crossing through his building in a day, on a user-grade cable modem.
Re: (Score:1)
Re: (Score:2)
Microsoft is in Redmond, not Seattle proper. The outlaying areas are all served with symmetrical gigabit FttH, while Seattle itself in many neighborhoods is still stuck on shit POTS lines
Re: (Score:2)
Hey. Be happy you're the Graceland of Grunge and the Capital Of Crappy Coffee.
Don't be greedy!
Re: (Score:2)
I was shocked to discover that the Seattle train station had no WiFi. The Seattle area is the home of two technology giants, Microsoft and Amazon; they should do better. If Microsoft can sponsor WiFi on the Space Needle campus they can also cover the train station.
That same station was also a dead spot for T-Mobile cell coverage. That would be bad enough in any major city, but Seattle is the home city of T-Mobile US. It is one of those old stone buildings that eat RF, but it's a location that should get spe