Friday, January 20th 2012
AMD Vishera Packs Quad-Channel DDR3 IMC, G34 En Route Desktop?
AMD might be a little sore that its "Zambezi" FX processor family based on its much-hyped "Bulldozer" architecture didn't quite meet the performance expectations of a ground-up new CPU architecture, but it doesn't want to take chances and build hype around the architecture that succeeds it. From various sources, some faintly-reliable, we have been hearing that the next-generation of high-performance desktop processors based on "Piledriver" architecture, codenamed "Vishera", will pack five modules or 10 cores, and will be structured essentially like Zambezi, since Piledriver is basically a refinement of Bulldozer architecture. The latest leak comes from the Software Optimization Guide for AMD 15h family (read here), which was picked up by CPU World while most of us were busy with CES.
CPU World compiled most of the features of what it suspected to be AMD referring to its future processors based on the Piledriver architecture, that's "Vishera" (desktop high-performance), "Terramar" (high-density server), and "Sepang" (small-medium business server) parts. The three are not the first chips to be based on Piledriver, AMD has a new mainstream desktop and notebook APU in the works codenamed "Trinity", which is en route for a little later this year. Trinity basically has an identical CPUID instruction-set as Vishera, Terramar, and Sepang, confirming their common lineage compared to today's "Bulldozer" architecture. The most catchy detail is of Vishera featuring 4 DDR3 channels.The plot thickens where "HyperTransport Assist feature" is listed as being supported on Vishera. HT Assist is a feature found on AMD's enterprise socket G34 processors, which facilitates better inter-die communication between the two dies of a typical socket G34 Opteron processor. The G34 (LGA1972) package is a multi-chip module of two quad-core, six-core, or four-module dies, which combined have four DDR3 memory channels, and a number of HyperTransport links to communicate with neighbouring sockets and the system's chipset. Could this be the first indication that AMD wants to take on Intel LGA2011 HEDT (high-end desktop) using Vishera chips in the G34 package? It will be a while before we find out.
Apart from using common silicon between client and enterprise platforms, AMD does have a history of colliding the two.
Source:
CPU World
CPU World compiled most of the features of what it suspected to be AMD referring to its future processors based on the Piledriver architecture, that's "Vishera" (desktop high-performance), "Terramar" (high-density server), and "Sepang" (small-medium business server) parts. The three are not the first chips to be based on Piledriver, AMD has a new mainstream desktop and notebook APU in the works codenamed "Trinity", which is en route for a little later this year. Trinity basically has an identical CPUID instruction-set as Vishera, Terramar, and Sepang, confirming their common lineage compared to today's "Bulldozer" architecture. The most catchy detail is of Vishera featuring 4 DDR3 channels.The plot thickens where "HyperTransport Assist feature" is listed as being supported on Vishera. HT Assist is a feature found on AMD's enterprise socket G34 processors, which facilitates better inter-die communication between the two dies of a typical socket G34 Opteron processor. The G34 (LGA1972) package is a multi-chip module of two quad-core, six-core, or four-module dies, which combined have four DDR3 memory channels, and a number of HyperTransport links to communicate with neighbouring sockets and the system's chipset. Could this be the first indication that AMD wants to take on Intel LGA2011 HEDT (high-end desktop) using Vishera chips in the G34 package? It will be a while before we find out.
Apart from using common silicon between client and enterprise platforms, AMD does have a history of colliding the two.
229 Comments on AMD Vishera Packs Quad-Channel DDR3 IMC, G34 En Route Desktop?
Once again everyone is ignoring what bulldozer is good at, multi-tasking and multi-threading. Maybe some multi-threaded benchmarks where different tasks are performed at the same time would show where BD can flex its muscles.
Now I'm just playing devils advocate, don't rip my head off for defending AMD, because AMD is a good company and they really are trying. Give it time, and don't forget, AMD isn't solely a CPU vendor so they have time to improve the architecture. As threading becomes more prevalent being able to squeeze more cores in the same area will be more useful than a higher IPC, because clocks can only go so high when using Si. Not to say IPC isn't important, but more "cores" can improve performance much more in applications that can utilize it.
Stop looking in the past and look at the future. Multi-core systems are everywhere and software vendors will want to take advantage of that. Also notice the i7 2600k's performance and the 2500k's performance. HyperThreading doesn't give you nearly a full core worth of performance, Bulldozer scales almost LINEARLY.
If AMD can work out the branch predictor issues and latency issues on the cache, it will be a very worth while platform for the price.
Edit: ...and Xeno, I'm not disagreeing with you, I completely agree. I'm just tired of people bashing Bulldozer where it really isn't a bad platform, there just isn't enough software to take advantage of it yet. A great example would be video encoding on 8 cores, it will keep up with the 2600k no problem.
The FX-8xxx CPU's can be viable, but with Intel's offerings usually being just as cost-effective and often even more so, BD does seem like a pretty bad platform. Hopefully they can fix all this with Piledriver, and software will start using more threads, but for the time being, it's just not a great offering.
by that time tho, well have other tech that isnt used yet, its called evolution, and before you start chatting that its worse then phenom etc etc, man evolved from ape yet comparitively were shit at climbing but i would not call man a FAIL nuff said
and if i was to buy an FXBD it wouldnt be so i can run poorly threaded software well ,it would be to run future and present multi threaded(properly) games like Bf3 etc
I play games, I run a lot of VMs, and I love my video and Zambezi (8-core of course,) would suit my purposes very well and should perform similarly to the 2600k for my purposes.
You also mensioned that not many applications use multiple threads, you're very correct... but it will not always stay that way. It is already changing and games like Crysis 2 and Battlefield 3 show little to no difference between CPUs.
and because intel barely scales a maximum of 30% with hyperthreading while amd bulldozer scales 80% compared to the performance of a module running one thread
Yes, I said logical threads because you're right, BD doesn't have specific cores, it just has extra hardware to run two threads in tandem. Intel's HyperThreading doesn't even really run two threads in parallel using HT, it is just using unused parts of the CPU when certain instructions are being executed.
AMD created a module knowing that they could run two threads while only using a fraction of the die space of a real full blown core. Now as far as floating point ops are concerned, applications using FMA3 and other new FP extensions will enable BD to do two FP calculations instead of one because of the size of the floating point unit on BD. (2 single-precision ops or 1 double-precision op.)
When push comes to shove, BD is more scalable and Sandy Bridge has a better IPC.
Think for a moment though, AMD has time to improve their IPC now that they have a scalable platform. Intel's cores are huge and on a chip the size of a SB-E imagine how many modules AMD could fit on there.
Note: Intel will find out that once they hit 16nm that making their CPUs smaller will be rather difficult due to quantum tunneling, so they need to find out how to reduce the size of their cores before too long.
and as amd modules become more optimized and reach better scaling you will end up with hardware that is running with every bit in it doing productive work
as you know its usualy the integer core doing most of the work while other hardware waits on it between its cycles, amd pretty much made a second integer core to receive data from the same hardware while the other integer core is crunching the data, so tuning these cores for the perfect latency to get this going will only mean scaling gets better
and as you said amd has so much room to improve ipc, for all you know they can pretty much create a module the size of 2 sandy bridge cores with 10 decoders and 512fpu or what not but that is exactly wat amd is moving away from.
Imagine if ATi (now AMD) and nVidia kept using pixel and vertex pipelines. They had criticism because a shaders were slower than the pipelines, but they were smaller and resulted in overall better performance after a couple of generations. CPUs are no different.
I would expect much more than 10% out of piledriver, I expect like 20% in ipc(the 10% that bulldozer missed, and the 10% that amd had in mind for pd) and if clockspeed hits the desired numbers then thats another 10-20% performance from higher clockspeed
and as scaling gets better and the architecture gets more optimized then multithreaded apps will perform even better than now
as for bulldozer it already exceeds phenom II except in certain cases using older instruction sets, but it clocks much higher so it still beats phenom II as you reach higher frequencies, comparing clock for clock isnt all practical since they are 2 different architectures
With all of this said, AMD is suffering in a single segment, high-end mainstream and enthusiast level chips which is the smallest portion of the market. Just wait and see what happens.
It seems, AMD is building on it's strengths such as they did in the past with the Athlon 64. Capitalize on Intel's weaknesses to succeed.
Just because software doesn't use multiple threads doesn't mean that it won't. Applications tend to use hardware that is available, not hardware that could be available. As the number of cores increases, more and more software will use multiple threads. AMD took a gamble with a completely new CPU design, and you can't expect the first revision to work perfectly and keep up with an architecture that Intel has been working on for 6 years (yes, Intel's architecture all started with the Core 2 lineup) so Intel has a lead on what they already have.
You have to give AMD time to improve this architecture and you have to give software companies time to optimize their software to take advantage of it.
Finally with all of this said, yes, Bulldozer is slower than Sandy Bridge, but that isn't stopping users from buying it. Also keep in mind Intel had something like 85% of the market, even before Bulldozer came out. Take that number with a grain of salt though, I haven't checked these numbers for a couple years.
if AMD came out with a 20Ghz 4 core CPU that was SLIGHTLY better than a 2600k but SLIGHTLY cheaper, would you not buy it just because the clock to clock is shit? no, you'd be an idiot... if it runs better at out of box speed vs out of box speed that is what matters. that or Max clock vs Max clock. you can't compare 4ghz to 4ghz on different architectures, it just makes no sense, its completely stupid to do so.
I'm trying to educate the population. I have a degree in Computer Science and I have a job as a Systems Administrator, what do you all have and do for work? :banghead: I know, some of the things I read here just annoy me. If people are really interested in this stuff, they should be trying to learn what is really going on instead of trying to show how big their e-peen is... they would be wise to listen.
Keep it up! It's refreshing to have someone who actually knows what they're talking about posting on these forums instead of the usual know-next-to-nothing gamers. :)
I don't have a degree in computer science, I'm just a dilettante trained in logical argument and analysis, learning, and refining what I learn here, as I go. :D
There is no sigma towards wanting to learn more or wanting to know how the stuff works. I just ask the these people ask instead of trying to interpret what it says on their own. There are people here who are willing to give their professional "6-sense".
Granted this thread is starting to run off topic and I think every angle has just about been hit. I would like to ask a moderator to lock this thread if no one has anything extra to add. :toast:
Edit: Inceptor, I like your signature. :)
www.fudzilla.com/home/item/26253-amd-fx-8350-vishera-cpu-production-in-q3 20% IPC improvement clock 4 clock over the Bulldozer is very impressive. :)
I am taking 20% based on the upcoming Trinitys base Piledrier CPUs. So the desktop should end up faster.