Slashdot Mirror
AMD Launches Piledriver-Based 12 and 16-Core Opteron 6300 Family
MojoKid writes "AMD's new Piledriver-based Opterons are launching today, completing a product refresh that the company began last spring with its Trinity APUs. The new 12 & 16-core Piledriver parts are debuting as the Opteron 6300 series. AMD predicts performance increases of about 8% in integer and floating-point operations. With this round of CPUs, AMD has split its clock speed Turbo range into 'Max' and 'Max All Cores.' The AMD Opteron 6380, for example, is a 2.5GHz CPU with a Max Turbo speed of 3.4GHz and a 2.8GHz Max All Cores Turbo speed."
6200 series have shared FPU (floating point unit). Which means that there are less FPUs that there are processing cores. To multiply two floating point numbers cores are waiting in a queue until FPU is free to use, this happens when all cores are calculating at the same time. If you are doing intensive calculations this is going to be slower than if you used 6100 series. 6100 series have dedicated FPU for each core.
I know this because we were recently buying a new cluster for calculations using YADE software.
How, here's the question: how about 6300 series, is there a dedicated FPU?
#
#\ @ ? Colonize Mars
#
I'm not even sure how you could post a story about AMD, what with it's recent decline this entire last decade, and not directly compare them to intel.
Are these even desktop or server chips? It's been so long since I bought AMD, I really couldn't tell you which line Piledriver sits in anymore, or if they've consolidated them.
The general gist I've read is that AMD is cheaper than Intel, and in the past has been "more green" due to power consumption, but with Ivy Bridge, your bang for the buck and much, much smaller lithography process has given intel the advantage in both areas.
moox. for a new generation.
Will it contribute to its survival ? Or is this one in a long series of convulsions accompanying AMD's bleeding to death ?
Religous speak to God. Insane are spoken to by God. When all shut up, one can finally hear Shostakovich in peace
Piledriver is the architecture, like Intel's Ivy Bridge is the architecture.
These are server chips. Best case, these are finally faster than their pre-Bulldozer parts in real, consumer desktop use. They will not beat an 8 core Sandy Bridge Xeon in FP-heavy applications, and power consumption is, at best, on the same level as the Xeons.
All they can do is work like crazy on their next line (Steamroller as it?) so they're truly competitive again.
Are these even desktop or server chips? It's been so long since I bought AMD, I really couldn't tell you which line Piledriver sits in anymore, or if they've consolidated them. The general gist I've read is that AMD is cheaper than Intel, and in the past has been "more green" due to power consumption, but with Ivy Bridge, your bang for the buck and much, much smaller lithography process has given intel the advantage in both areas.
Server chips. Opteron was always, and always has been about servers.
I am not a business owner and do not operate servers myself. For home usage, a low price CPU with adequate power will kick intel "we-cripple-all-but-i7-features" anytime in value for my $. I do not do geek pissing contests.
Tomorrow is another day...
Last year we bought some servers with 6 core cpu's
The. SQL 2012 came out with per core licensing
I did some quick math and its cheaper to buy new servers with 4 core cpu's than license SQL 2012 for 12 cores per server
AMD has lost performance/watt recently. These are intended as server chips (G34 socket, not AM3+).
These might bring back performance per watt, as AMDs have seemed to scale better in the the multi-CPU per box / multi-core per CPU segment recently.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
These Opteron models are the new server line from AMD. The desktop version based on the same architecture (the Trinity alluded to in the summary) closed some of the gap against Intel. But Intel remains the market leader on single core performance, performance per core, and power utilization. AMD continues to push the number of cores upward more aggressively, but there's not many workloads where that matters enough for their slim advantage to result in a net win. And the lower efficiency means that sometimes even having more cores doesn't aggregate into enough speed to be a useful alternative. That leaves AMD to compete on pricing. And the CPU is a relatively small part of the total budget on larger servers. Load up a Dell 815 for example and you'll find the CPU pricing seems small compared to what filling its RAM capacity up costs. And then there's reliable storage, at a while higher price level altogether.
The rule of thumb I've been using for the last year, based on benchmarking of CPU heavy database work, is that I expect a 32 core AMD server to be about as fast as a 24 core Intel one, while using significantly more power. The 40% performance per watt gain claimed here--from AMD's own hand-picked best case scenario benchmark--is only enough to make the Intel performance and gap decrease in size, not go away. We'll see if these new Opterons benefit from the re-engineering work done recently more than the desktop ones did; so far it doesn't look good.
AMD states that "an 8% increase is enough to no longer need to widen their product line to 20 core CPUs". Guess what AMD, when your CPUs perform like shit compared to other server based CPUs, an 8% increase of shit means nothing. Another 4 cores could have compensated for the terrible performance your CPU provides in comparison to Intel's E5 line, particularly for highly concurrent based programs. An 8% only warrents that our company purchases the new 4 socket Intel based motherboards. Yeah I know, "But hey, Intel CPUs cost much more... blah blah blah", if you're living in your mother's basement, then certainly Intel's CPUs cost too much for your world of warcraft 10hour/day "work". But when you're actually relying on the CPU performance, as we do due to scientific, financial, and industrial simulations and applications, then Intel's nearly 50% superior performance more than pays for the cost of the CPUs within a week.
MD continues to push the number of cores upward more aggressively, but there's not many workloads where that matters enough for their slim advantage to result in a net win.
I disagree: that's exactly what Xeon and Opteron are about. What differentiates those two from the Core and Phenom processors is that the former have multiple crazy fast and very expensive low latency links to allow glueless multi socket systems. Once you've got an (16)8 (hyper)thread Xeon or 16 core opteron and have more than one socket, you're already expecting a workload to scale to 32 distinct units.
Basically these chips only make sense for pretty parallel work loads.
Load up a Dell 815 for example and you'll find the CPU pricing seems small compared to what filling its RAM capacity up costs.
I use this as my go-to online quoter.
http://www.woc.co.uk/default6.aspx?nquoter=13
I have no affiliation except that I've bought such machines from them before.
Maxing out the RAM (512G) costs £3300. Maxing out the processors costs £2300. It's not quite as much, but it's substantially over half the price.
That said, I've heard rumours that the new opterons can drive 32G DIMMS, in which case you could load it up with 1T RAM for the low, low price of £30,000. In which case, your point certainly stands.
The 40% performance per watt gain claimed here--from AMD's own hand-picked best case scenario benchmark--is only enough to make the Intel performance and gap decrease in size, not go away
True, but the Opterons are substantially cheaper. If you factor in lifetime cost including bang for buck, power and cooling, it's basically a wash and really dependent on the specific workload.
If they have closed the gap this much, then they will be a substantially cheaper option overall.
SJW n. One who posts facts.
Intel calculate their TDP based on full load which isn't necessarily maximum power use.
AMD calculate their TDP based on maximum power use.
Yes, they have a shared 256 bit FPU, but that can be split into two 128 bit parts. So no, multiplying two floating point numbers in two threads is performed immediately and simultaneously, the cores do not wait at all. I measured this on a previous generation Opteron 6234, the performance loss caused by running two threads on two cores of the same module vs two cores in different modules was barely measurable, 3%.
My wallet called, it said your fanboy attitude will not change it's mind about buying an AMD CPU.
Good to see they're not tapping out yet.
If this is not about a button on my case, then I don't care what that old buzzword is regurgigated into.
I wasn't clear enough on what I meant by number of cores. AMD's strengths when they did well in the server market (2003 to 2009) included more sockets, more cores per socket, and higher memory bandwidth to each socket. At this point the only one of those leads they maintain is that they still cram more real cores onto a socket than Intel does. Presuming the number of sockets is the same, I was suggesting that AMD's higher core count per socket doesn't give them much of a real-world advantage. As you suggested, the multi-socket situation isn't different enough between AMD and Intel for it to be a competitive advantage for either anymore; that's fairly level now.
An Intel server with 8 cores and the current generation of HyperThreading is not necessarily any slower than these new AMD ones with 16 real cores. There are times you run into memory bandwidth issues at the top end of concurrency, and Intel has been the leader on that since Nehalem in 2009. At the low end of active cores, sometimes there is just one thing you want to run really fast, and there Intel's Turbo approach is still better than AMD's. The middle area where AMD is at least competitive--lots of cores active but not constrained by memory bandwidth--is not that wide of a range of server workloads.
Piledriver is the architecture, like Intel's Ivy Bridge is the architecture.
These are server chips. Best case, these are finally faster than their pre-Bulldozer parts in real, consumer desktop use. They will not beat an 8 core Sandy Bridge Xeon in FP-heavy applications, and power consumption is, at best, on the same level as the Xeons.
That's true. A 16-core Opteron has the same FP width as an 8-core Xeon, and a higher TDP for a given clock.
On the other hand, we buy almost all AMD because it lets us build cheap 1U or 2U 4-socket servers with 512GB of RAM each. 4-socket Intel chips (E5-4600 or E7) are much more expensive; mid-range servers work out to 50% more for Intel, and high-end servers about 80% more for equivalent speed.
There are times you run into memory bandwidth issues at the top end of concurrency, and Intel has been the leader on that since Nehalem in 2009.
Certainly on the desktop. I thought on the server they both have quad channel DDR3 per socket. Of course, that gives intel a higher per-core bandwidth.
I thought that the 6200 series support slightly higher clocked memory (1866) compared to Intel (1600).
I've not cheecked more thoroughly than looking up a few figures though.
SJW n. One who posts facts.
"Athlon 64" was released in December 2003 that beat P4 in all regards, price, power consumption, performance. Intel recovered from it only in January 2006 with first "Core" CPU. How does that make AMD to be in "decline entire last decade" pretty please?
Rubbish about AMD having only one competitive time: The T-Bird Athlons offered much better bang for the buck and performance than Netburst based P4's.
Anyone who followed computer architecture knew this. Athlon XP's were generally considered better all around chips than Netburst based P4's until the P4's hit > 3.2+ Ghz. The only way Intel could even stay competitive with the T-Bird Athlons and Early Athlon XP's was stuff like the Extreme Edition.
Likewise, early dual core Netburst based products P4's weren't great chips, they were massive hot power-hogs. First generation Opterons were a much better choice for many server workloads. The general advice now is to look at your workload. If your choice is an Opteron versus a Xeon look at the system cost along with your workload. If the Opteron is cheaper then you can buy more memory (If that helps you workload). Hyper transport was significantly better than the Netburst based Xeons for multiprocessing.
For single threaded Integer+FPU performance Intel is definitely better than AMD right now.
If your running prebuilt binaries that aren't heavily threaded then Intel Xeons are your best bet, but don't forget approaching this as a systems engineering problem. If your workload loves memory then you may well be better off with an Opteron with lower single threaded performance, but a lower cost, and throw the cost delta towards more memory.
If cost is no object then your best bet is a Xeon. As much as people love to beat up AMD we have them to thank for decent x64 performance on the desktop. If they hadn't made a killer chip with the Opteron then Intel would have tried to push us all to Itanium which had great theoretical performance and great FP workload performance for apps the compiler could schedule decently for but sucked eggs running integer apps with a decent number of branches... Look at the Spec CPU Perl benchmark results for an idea how much Itanium sucked at that stuff.
"With this round of CPUs, AMD has split its clock speed Turbo range into 'Max' and 'Max All Cores.' "
Remember Episode 200 of Stargate and the "Set Weapons to Maximum!" line?
AntiFA: An abbreviation for Anti First Amendment.
With respect, why are you even commenting on this if you have not got that much out of the summary? I'll try though, these are for the sort of servers where a lot of tasks are done in parallel and it's a big deal since the best comparable Intel chips are 10 core, 2GHz and horribly expensive. That may change but Intel doesn't seem so interested in that end of the market for now and have let AMD undercut them by several multiples of the price ($9000 for 64 cores vs $80,000 for 80 cores back in January).
It could be argued that 10 Xeon cores act like 20 opteron cores, but that really depends on exactly what the tasks are.
Yes, but you are going to be getting the same amount of RAM and almost always the same type no matter which way you go. The price of a CPU really matters once you go beyond a couple of sockets. Also fuck Dell since you just get the whitebox of the week with a Dell badge on it - Supermicro and a long list of others will give you something better far cheaper and may even give you support from someone based in your own country that speaks your own language.
My anecdotal experience is that Intel's total cost of ownership is quite steep.
I had an ASRock AM2+/AM3 motherboard I purchased in 2010 for $80. I also purchased a Athlon 64 X2 5200 Brisbane 2.7GHz for it. Mind you, that's not the Athlon x2, that's the Athlon 64 x2 which is 1-2 generations behind of the one you're probably thinking of.
I also had another computer with an Intel Core 2 Duo 2.3GHz which ran a bit faster, but not by much.
The only upgrade path without a new motherboard for the Core 2 system was to get a Core 2 quad, which is still ridiculously expensive ($180+) for its age and not competitive.
On the other hand, AMD designed their architecture so that the Athlon 64 x2, Phenom, Phenom II could all use the same motherboard. That's at least three generations of chips. I got a 3.4GHz Phenom II quad-core black edition at newegg for $100, dropped it into my 3-year-old motherboard, overclocked it to 3.7GHz and have been very happy with my "bang for the buck." Sure, it isn't competitive cycle-per-cycle, but as a consumer, I mostly want it for video work (which it does great), photography work (which it does great), and gaming (which it does great).
That's what I would call bang for the buck, personally.
From a CPU design standpoint, Intel may be winning at the moment, but AMD has been looking to the future with their processor designs.
Intel has been pushing relatively the same design methodology and making incremental improvements since Core 2. AMD tore down their designs and started much closer to "from scratch." They didn't expect to be market leaders in performance with Bulldozer, and they refined it with Piledriver a good deal. This "half-core" strategy they've built on is actually a very good idea; it hasn't been done before quite this way, and they're still learning lessons. One of the best advantages, though, is the ability to put more cores onto a single chip.
Piledriver hit the performance improvement AMD claimed it would when Bulldozer was released. People compare it roughly to Intel i5's now. If they keep on their performance increase road map, there's no reason to think that Piledriver II or whatever won't be comparable to i7's, and after that, they're in step with i9's or whatever Intel names their Core line next, if not ahead.
As far as FP performance goes, AMD's higher-end CPU-integrated Radeons put Intel's offerings to shame, and give FP performance to the processor that the general-purpose cores couldn't hope to match (if used properly).
Intel's policy of keeping prices high on older processors and changing sockets frequently between generations doesn't help their total cost of ownership, either. You also have to include upgrade paths for a few years, motherboard/RAM costs in that, etc.
I recently bought one of their non-server Trinity APU processors specifically to be used for my HTPC. The power footprint is low enough that it fits in a shoebox sized enclosure and the integrated Radeon graphics mops the floor with anything from Sandy/IvyBridge and all at a lower cost. I use it to crank out 1080p video, send audio to my AVR and the kids use it to play games with a fair amount of eye candy turned on and at a playable resolution and frame rate.
Would I buy any current AMD processors for a server farm? Probably not.
Best,
Would I buy any current AMD processors for a server farm? Probably not.
The predecessors of this series, the Opteron 6200 is used in quite a few supercomputers. Actually I counted 21 Opteron bases systems in the last supercomputer top 100 list.
It supports turbo boost, virtualization (vt-x and vt-d), speedstep, etc.
Why do you need quad socket? Have you compared the supermicro boards with two E5-2600 Xeons and up to 512 GB RAM as 16 DIMMs? They even have ones with onboard dual 10G-BaseT in case that helps cut costs for your datacenter networking...
Yea, looks like Sandy Bridge EX is cancelled:
http://vr-zone.com/articles/ivy-bridge-ep-and-ex-coming-up-in-a-year-s-time--the-multi-socket-platform-heaven/15488.html
Ahahaha. IBM's POWER7+ CPU is the fastest general purpose server CPU you can buy. Go look at any benchmark, and it is frequently close to 2x the throughput of Intel's best server CPU (while being on a less advanced silicon process node).
This crap from AMD would not stand the slightest chance.
They're about as good as a mid-range Xeon while consuming more power?
AMD has lost performance/watt (on x86 at least) when they fired all the engineers who knew how to make performance/watt critical decisions. Then they hired a bunch of copy-pasters from India and China. I'll bet the CEO responsible got a hefty bonus for this "presidential" decision.
$
I struggle to max out my turion x2 dual compiling and re-encoding video...but then I refuse to buy computers, always get hand-me-downs etc but still i'm always amazed by whats out there these days...shared FPU's or not