Friday, January 17th 2014

AMD Readies 16-core Processors with Full Uncore

AMD released developer documentation for a new processor it's working on, and the way it's worded describes a chip with 8 modules, working out to 16 cores, on a single piece of silicon, referred to as Family 15h Models 30h - 3fh. This is not to be confused with the company's Opteron 6300-series "Abu Dhabi" chips, which are multi-chip modules of two 8-core dies, in the G34 package.

What's more, unlike the current "Abu Dhabi" and "Seoul" chips, the new silicon features a full-fledged uncore, complete with a PCI-Express gen 3.0 root complex that's integrated into the processor die. In what's more proof that it's a single die with 8 modules and not an MCM of two dies with 4 modules each, the document describes the die as featuring four HyperTransport links; letting it pair with four other processors in 4P multi-socket configurations. Such systems would feature a total core count of 64. There's no clarity on which exact micro-architecture the CPU modules are based on. Without doubt, AMD is designing this chip for its Opteron enterprise product stack, but it should also give us a glimmer of hope that AMD could continue to serve up high-performance client CPU, only ones that can't be based on socket AM3+.

Source: Planet3DNow.de
Add your own comment

92 Comments on AMD Readies 16-core Processors with Full Uncore

#1
Dent1
by: harry90
All amd Needed to do was to improve the single core performance of their cpu's. who needs more than 8 cores? Just enhance the IPC, single core performance by 40-60% and they could compete with intel!!!
What makes you think AMD care about competing with Intel on a performance level? The objective is to capture a larger market share and thus increase revenue for share holders. No point having the best performing product if nobody buys it.

AMD is probably doing whatever is cost effective for them and the most beneficial to them in the long term, although we probably can't see it now their key shareholders sat in a meeting and agreed this strategy, and right or wrong this was their best solution.


by: newtekie1

Honestly, I don't see a need for AMD to increase single threaded performance to meet Intel. The reason is that AMD's single threaded performance is good enough. Yes, they lag behind in benchmarks, but in real world use there really isn't anything that is single threaded that AMD can't handle. Saddly, games are still heavily dependent on single threaded performance, but most modern games still run perfectly well on AMD processors despite this, because AMD's single threaded performance is good enough. There are a few exceptions, StarCraft II comes to mind, because it is extremely CPU heavy and extremely single threaded.
I agree. Improving single threaded performance should be secondary. There isn't a single game or application that the average desktop user can't do.

When the need for more cores becomes necessary the work AMD did on their multiple module design will pay off. Even Intel's hyper threading wasn't successful at first, it took lots of trial and error and a decade later we all see the benefits. Same thing with AMD module principle.
Posted on Reply
#2
Vinska
by: harry90
Who needs more than 8 cores? Just enhance the IPC, single core performance by 40-60% and they could compete with intel!!!
I need. Most of the workloads I do are very heavily threaded. So I would trade my 8-core to, for example, a [hypothetical] CPU of the same architecture with 16 cores while having 35% lower clocks anytime.
Posted on Reply
#3
Eukashi
Please 4GHz 4M/8C 256SP-GCN APU on Socket FM2+.
When I record PC games, more CPU core is required for x264VFW.
Posted on Reply
#4
Lionheart
by: RCoon
I sincerely doubt this will ever come to any other market besides Opteron. If it does, enjoy the 50000W TDP.
Every thread I see you in you're always bashing AMD, give it a rest o_O
Posted on Reply
#5
eidairaman1
This is definitely a preview of what AMD has in store, first Kaveri, now Hexadecimal Core- maybe the introduction of it to AM3 or an 8 core model Steamroller for AM3+ then Hexadecimal core on the next Desktop socket
Posted on Reply
#6
TRWOV
by: RCoon
I sincerely doubt this will ever come to any other market besides Opteron. If it does, enjoy the 50000W TDP.
More like 150w, if done on 28nm.


by: eidairaman1
This is definitely a preview of what AMD has in store, first Kaveri, now Hexadecimal Core- maybe the introduction of it to AM3 or an 8 core model Steamroller for AM3+ then Hexadecimal core on the next Desktop socket
According to AMD's roadmap, Vishera 32mn is as far as AM3+ gets.
Posted on Reply
#7
arbiter
by: newtekie1
Honestly, I don't see a need for AMD to increase single threaded performance to meet Intel. The reason is that AMD's single threaded performance is good enough. Yes, they lag behind in benchmarks, but in real world use there really isn't anything that is single threaded that AMD can't handle. Saddly, games are still heavily dependent on single threaded performance, but most modern games still run perfectly well on AMD processors despite this, because AMD's single threaded performance is good enough. There are a few exceptions, StarCraft II comes to mind, because it is extremely CPU heavy and extremely single threaded.
AMD single thread is good enough? Yea really? How many games at this point in time for example uses more then 4 threads? Heck even a lot of programs don't really use much more then 2. Less you get in to encoding graphic design kinda stuff that really helps a ton in. AMD really should put some R&D in to increasing performance cause don't think to many games will span much past 4 cores on best top side which puts AMD behind a bit, on top of that they use 50% more power then the competitors cpu. Yea AMD cpu looks good cause initial cheaper cost but over year or 2 that cost even's out when it add's up in an eletric bill. AMD needs to get i would say either same performance with lower wattage or same wattage with say around 30% boost in single thread work loads. But that game seems to change quick when the 6/8core haswells come out in next 6ish months.
Posted on Reply
#8
newtekie1
Semi-Retired Folder
by: TRWOV
More like 150w, if done on 28nm.
The current 16-Core parts are 115w on 32nm, so I'm guessing these would likely be similar but with higher clocks.

by: arbiter
AMD single thread is good enough? Yea really? How many games at this point in time for example uses more then 4 threads? Heck even a lot of programs don't really use much more then 2. Less you get in to encoding graphic design kinda stuff that really helps a ton in. AMD really should put some R&D in to increasing performance cause don't think to many games will span much past 4 cores on best top side which puts AMD behind a bit, on top of that they use 50% more power then the competitors cpu. Yea AMD cpu looks good cause initial cheaper cost but over year or 2 that cost even's out when it add's up in an eletric bill. AMD needs to get i would say either same performance with lower wattage or same wattage with say around 30% boost in single thread work loads. But that game seems to change quick when the 6/8core haswells come out in next 6ish months.
There isn't anything that relies on single threaded performance, games included, that don't run well on AMDs. There was a time when software was outpacing hardware, but we've reached a point where the hardware has caught up and software has sat stagnant. There are a lot of people still running modern games on Core 2s(heck one of my gaming rigs is a Celeron E3300) and they are still working just fine. And the FX series is beyond Core 2 in single threaded performance. So, yes, AMD's single threaded performance is good enough.
Posted on Reply
#9
RCoon
Gaming Moderator
by: Lionheart
Every thread I see you in you're always bashing AMD, give it a rest o_O
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
Posted on Reply
#10
Frick
Fishfaced Nincompoop
by: RCoon
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
That is because you let yourself believe the hype. You're blaming AMD for your personal failures!

Seriousy though, newtekies idea with a unified socket is great. I'd buy into that.
Posted on Reply
#11
Ravenas
by: RCoon
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
Are you telling me that I am completely disappointed with my 8350? Ha ha ha. Please stop trying to speak for people on this thread.
Posted on Reply
#12
Mindweaver
Moderato®™
I'll take a 16 core desktop! :toast: These should make great crunchers/folders.
Posted on Reply
#13
Dent1
by: RCoon
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
I'm happy with my 2009 Athlon II X4. Still waiting for games to take advantage of it. Can't say I'm disappointed with AMD. Actually I'm happy that I didn't have to change boards and I was able to drop in this fantastic piece of circuitry and its lasted 4 years and going strong.

No need to feel disappointed for me. You can have your disappointment back. Here.
Posted on Reply
#14
Vinska
by: arbiter
AMD single thread is good enough? Yea really? How many games at this point in time for example uses more then 4 threads? Heck even a lot of programs don't really use much more then 2. Less you get in to encoding graphic design kinda stuff that really helps a ton in.
Okay, most games use 4 threads or even less. But that doesn't mean other cores are useless. I can play games with at worst case negligible (due to no turbo) and at best case no performance loss while recording gameplay at 1920x1080@40 (or more) with superb quality straight into h264 with enough compression to keep the filesize relatively tiny (using ffmpeg+libx264). Meanwhile, a friend with a 4-core Intel said it becomes a bit lacking, reducing performance and making him record at slightly lower video framerates for it to keep up. Video transcoding speed is also very good, as long as using sane software. As @[USER=69279]Mindweaver[/USER] already mentioned, for crunching/folding – more cores, the better. Very important for me – compiling. Compiling benefits greatly from increased core count and pretty much scales linearly. [Re]Compiling larger projects can take very long on just a few cores. As a software engineer / programmer, I often need to recompile large projects several times a day. Especially when doing regression tests, where it can easily need a over dozen recompiles. Thus, when I moved from my dual-core to a octa-core, there was much rejoicing due to reducing compile time of a certain project I work on from ~ one hour to less than five minutes. Also, it is disappointing that in the Windows world a lot of software is still poorly threaded. While on Linux (the OS I use 99% of the time), things tend to be more threaded.

Sure, I can get the same or even better MT performance with a 6-core HT'ed Intel. But for what? 2x the price or even more? Thanks, but no thanks.

P.S. +1 to what newtekie1 said

by: arbiter
on top of that they use 50% more power then the competitors cpu. Yea AMD cpu looks good cause initial cheaper cost but over year or 2 that cost even's out when it add's up in an eletric bill. AMD needs to get i would say either same performance with lower wattage or same wattage with say around 30% boost in single thread work loads. But that game seems to change quick when the 6/8core haswells come out in next 6ish months.
Okay, I kept deciding on not pointing this out, but I will. Since people tend to needlessly bash AMD for inefficient design when it comes to power consumption.
Example, sorted from most to least watts allocated to a single core:
code:
i7-3820 – 4 cores, 130W TDP; 130 / 4 = 32.5W per core
FX-4350 – 4 cores, 125W TDP; 125 / 4 = 31.25W per core
FX-9590 – 8 cores, 220W TDP; 220 / 8 = 27.5W per core
i7-3970X – 6 cores, 150W TDP; 150 / 6 = 25W per core
i7-2600K – 4 cores, 95W TDP; 95 / 4 = 23.75W per core
FX-4320 – 4 cores, 95W TDP; 95 / 4 = 23.75W per core
i5-2450P – 4 cores, 95W TDP; 95 / 4 = 23.75W per core
i7-4770K – 4 cores, 84W TDP; 84 / 4 = 21W per core
FX-6350 – 6 cores, 125W TDP; 125 / 6 = 20.83W per core
i5-3550 – 4 cores, 77W TDP; 77 / 4 = 19.25W per core
i3-2330M – 2 cores, 35W TDP; 35 / 2 = 17.5W per core
FX-6300 – 6 cores. 95W TDP; 95 / 6 = 15.83W per core
FX-8350 – 8 cores, 125W TDP; 125 / 8 = 15.625W per core


OH SNAP it appears that if we consider how much TDP is allocated to a single core, it doesn't look like AMD CPUs are inefficient – the power per core is quite low, relatively. Which is only possible if the cores are efficient enough.

by: RCoon
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
>completely disappointed me and everyone else
>implying
>implying


You are implying too much, sir.
P.S. I totally love it.

by: Ravenas
Are you telling me that I am completely disappointed with my 8350? Ha ha ha. Please stop trying to speak for people on this thread.
Yeah, what He said. When You, @[USER=104854]RCoon[/USER], say "and everyone else", You take on quite a bit of responsibility, Ya know...
Posted on Reply
#15
RCoon
Gaming Moderator
Seems like there's a lot of people kidding themselves in this forum. Waste of my time.
Posted on Reply
#16
TRWOV
by: RCoon
I own an 8350, I can 'opinion away' all I like. I was an amd fan through and through, then they messed up and completely disappointed me and everyone else.
I'm so disappointed that I went and bought only 4 of them.




by: Vinska

OH SNAP it appears that if we consider how much TDP is allocated to a single core, it doesn't look like AMD CPUs are inefficient – the power per core is quite low, relatively. Which is only possible if the cores are efficient enough.
Intel's TDP isn't calculated in the same way as AMD's TDP. Intel's TDP is rated higher than the actual value because it's a worst case scenario. AMD's TDP is the average power draw during a set of test workloads.


Let's not kid ourselves. Intel CPUs are more energy efficient than AMD CPUs, that's a given, but AMD CPUs aren't terribly inefficient considering their core count.
Posted on Reply
#17
fullinfusion
1.21 Gigawatts
by: RCoon
Seems like there's a lot of people kidding themselves in this forum. Waste of my time.
Then leave Mr downer! o_O
Posted on Reply
#18
Vinska
by: TRWOV
Intel's TDP isn't calculated in the same way as AMD's TDP. Intel's TDP is rated higher than the actual value because it's a worst case scenario. AMD's TDP is the average power draw during a set of test workloads.
LOL. I thought it was the other way round.
And my FX-8320 only goes close to TDP when overclocked, where it hovers around 120-123 W on full load, as reported by internal sensors or whatever sh*t.
And when I overclock More Than I Should™, it appears to drop my voltage on load to zealously keep the power consumption below 124.75W no matter what. Unless I disable lotsa stuff and turn on several overrides (can't find a better word) that my previous mobo didn't even have. (So my previous mobo was zealously keeping the power draw like this all the time)
So, from what I saw with my own eyes, saying that AMD's TDP is "average power draw" must be very much false.

by: RCoon
Seems like there's a lot of people kidding themselves in this forum. Waste of my time.
Real smooth. That just shows You are out of arguments and don't have anything genuinely useful to say. Aww well...
Posted on Reply
#19
TRWOV
AMD themselves say that it's an average:
TEST CONDITIONS
Given the goal of representing typical power usage in real world conditions, environmental test conditions were chosen to reflect that aspect (room temp of 70°F, server’s fan heat sink used, closed case, etc.) The power for the cores, memory controller, and HyperTransport™ links was logged multiple times per second throughout the entire duration of the workload tested, and the time- averaged power consumption for that workload was calculated. The results across the suite of workloads are used to derive the ACP number. The ACP value for each processor power band is representative of the geometric mean for the entire suite of benchmark applications plus a margin based on AMD historical manufacturing experience.
http://www.amd.com/us/Documents/43761D-ACP_PowerConsumption.pdf

EDIT: That being said, AMD tests with lower binned parts so you could say that it's TDP is an average of the lowest binned CPUs.



It could be that your motherboard is undervolting your CPU, my GA-880GM-USB3 undervolts my 8350 to 1.28v when I set voltage on AUTO.
Posted on Reply
#20
Vinska
isn't that just for Opterons?
Posted on Reply
#21
micropage7
by: RCoon
I sincerely doubt this will ever come to any other market besides Opteron. If it does, enjoy the 50000W TDP.
yeah, one major problem ofAMD is power consumption
come on AMD
Posted on Reply
#22
Dent1
by: RCoon
Seems like there's a lot of people kidding themselves in this forum. Waste of my time.
So your saying this forum only makes good use of your time when we share your views and agree with you?
Posted on Reply
#23
theoneandonlymrk
I feckin love the power consumption card, really funny.
On Tpu most ardent members in actuality smash efficiency to the kerb in favour of Ghz or even Mhz gains , this is not Eco power up that's elsewhere and to the likes of me a few watts means nothing get over it.
Sdp is coming to Amd parts soon enough mark my words as intels subterfuge seams to have blinded their fans.
Far to busy kidding myself to rise to Rcoons Bs.
Posted on Reply
#24
NeoXF
As much as I'd like to see a (very plausible) 6M/12T/12 core Steamroller w/ L3 and PCI-Express 3.0 (that admitedly, currently would be platform-less)... I'm still rooting for APUs, now more than ever. Software developers and open sources projects need to get their shit together and cook us some HSA magic tho. So far we've only got 2-3 actual previews and a handful of promises from developers X and Y.

IMHO, HSA needs to be pushed from the ARM front too if it wants get full-scale traction.
Posted on Reply
#25
lilhasselhoffer
Allow me to make the performance arguments, so this thread doesn't become a pissing contest.

1) AMD chips have a higher TDP than Intel, so they must be more efficient.
No. AMD and Intel do measure chips differently. Between chips being measured differently, and completely differing architecture, efficiency cases can be made for both sides. There is no clear winner here.
2) Intel chips don't clock as high as AMD ones, so AMD makes better chips.
This one is generally true. If you're looking for bragging rights about the highest clock, then AMD wins. The reality is that both manufacturers' chips take huge amounts of power to do this. You don't run a CPU at peak frequencies constantly, unless you want a huge bill and rapidly deteriorating chip. For every day use, either manufacturer produces a relatively solidly performing chip.
3) Intel and AMD don't measure cores the same.
Absolutely. Intel has traditional cores, while AMD decided to share a component among the cores. A four core Intel chip doesn't match the 4 core AMD chip, a two core with hyper-threading chip doesn't match a 4 core AMD chip, and none of this matters. This is not a move for the consumer CPU market. In that market only a handful of program use more than a couple of cores. People using more cores are doing server related work, crunching, or running encoding software.


Now that the silly arguments have been made, can we get back on topic? AMD looks to be firing for the server market, without any bashfulness. Assuming this is the case, it seems like they are making a large step back into competing with Intel. This bodes well for more reasonably priced servers, but more importantly could be parlayed into something interesting on the desktop CPU front. Anyone care to comment on that, rather than on how much they think the current parts are either awesome or terrible?
Posted on Reply
Add your own comment