1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Vishera Packs Quad-Channel DDR3 IMC, G34 En Route Desktop?

Discussion in 'News' started by btarunr, Jan 20, 2012.

  1. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
    AMD is making changes. I would take any roadmap with a grain of salt at this moment.
    I can easily see Quad-Channel memory for Piledriver Desktop CPU's. Bulldozer was suppose to be Quad, but AMD made last minute changes.

    In order for AMD to further better compete, Piledriver may be fully based on there server CPU's but with AM3+ and/or FM2.
     
  2. Patriot

    Joined:
    Oct 27, 2009
    Messages:
    232 (0.11/day)
    Thanks Received:
    25
    Location:
    Republic of Texas
    you didn't read...

    IL was never going to be dual and isn't dual.
    That roadmap says it is...

    IL IS bulldozer...
    its a family

    That slide is all sorts of mixed up.
    I still don't think you understand current sockets or families of cpus.
     
  3. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
    I do understand, I get what you are saying now :D
     
  4. seronx

    seronx

    Joined:
    Jul 10, 2010
    Messages:
    994 (0.56/day)
    Thanks Received:
    219
    Location:
    USA, Arizona
    FX(Bulldozer) was fully based on the Server CPUs(Interlagos/Valencia) but was on AM3+

    FX(Piledriver) will be based on the Server CPUs(Terramar/Sepang) but will be on AM3+/FM2/C2012(to satisfy certain somebodies)
     
    Super XP says thanks.
  5. OOZMAN

    OOZMAN New Member

    Joined:
    Nov 24, 2011
    Messages:
    160 (0.13/day)
    Thanks Received:
    20
    Location:
    Brisbane, Australia
    What on earth did you just say?

    "all these people winging about BD better be folding or cryunching 24/7 as if your pc sits doing nowt while you workin then is only on to game and surf and your still moaning about BD your not right in the head"

    Still figuring that one out...

    *facepalm*

    You cut the productivity results out.
    [​IMG]

    Ah yes, the old 'Look at this one screenshot of a random game from a random site that's not consistent with any other site's results of BD competing with Sandy, making it an overall equally good gaming CPU.'

    Seen that one a few times... ;)
     
    ensabrenoir says thanks.
  6. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
    I was comparing gaming benchmarks. Also find me other benchmarks from different sites with the same games Hardwareheaven used, and you will see results are similar.
     
  7. pantherx12

    pantherx12 New Member

    Joined:
    Jan 2, 2009
    Messages:
    9,714 (4.17/day)
    Thanks Received:
    1,699
    Location:
    ENGLAND-LAND-LAND
    To be fair, gaming (unless it really is a CPU hog like a strategy game with crazy unit amounts) shouldn't be how to determine if a CPU is good or not.
     
  8. xenocide

    xenocide

    Joined:
    Mar 24, 2011
    Messages:
    2,162 (1.42/day)
    Thanks Received:
    470
    Location:
    Burlington, VT
    In the image posted comparing the 990x and BD CPU, you used the same Civ V result twice since it was one BD one, the listing clearly stacked up on Benchmarks that favoured BD. Not to mention it was being compared to an older less efficient and powerful architecture. Lets see what happens when we use the same source and compare CPU's that are in the same Price\Performance region;

    http://www.anandtech.com/bench/Product/434?vs=288

    The i5-2500k beats it in every single game, and costs less.

    A good number of Computer Enthusiasts are gamers, so they walk hand in hand quite often. For general purposes just about any CPU these days would work xD
     
  9. eidairaman1

    eidairaman1

    Joined:
    Jul 2, 2007
    Messages:
    14,166 (4.92/day)
    Thanks Received:
    2,052
    Lets not turn this into another vs thread shall we?

     
  10. xenocide

    xenocide

    Joined:
    Mar 24, 2011
    Messages:
    2,162 (1.42/day)
    Thanks Received:
    470
    Location:
    Burlington, VT
    I'm not, just pointing out why his post was misleading. I'm interested in seeing how well AMD can improve BD, which I think was very underwhelming. Vishera should be interesting.
     
  11. seronx

    seronx

    Joined:
    Jul 10, 2010
    Messages:
    994 (0.56/day)
    Thanks Received:
    219
    Location:
    USA, Arizona
    They are not going to improve it the way you like it ;)

    FMA is here to stay...
     
  12. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
  13. nt300

    nt300

    Joined:
    Jan 20, 2010
    Messages:
    868 (0.45/day)
    Thanks Received:
    160
    Location:
    Toronto, ON. Canada
    Good to see AMD doing better with Piledriver :)
     
  14. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
    Except 2 of them. And yes I fully agree when it comes down to Price/Performance the FX-8150 fails, but the FX-8120 does not due to the fact you can easily OC the sucker. Though overall platform cost Intel is approx" $100 to $200 more, though considering all the upgrades people do, that is not a lot of extra cash.
     
  15. Aquinus

    Aquinus Resident Wat-man

    Joined:
    Jan 28, 2012
    Messages:
    7,411 (6.12/day)
    Thanks Received:
    2,839
    Location:
    Concord, NH
    Memory and cache latencies are too high and and the length of bulldozer's pipeline is just simply too long, which is why BD's IPC is horrible. The best way to do branch prediction is not having to do it in the first place, which a long pipeline doesn't help. Intel is perfectly happy with their (13 stage?) pipeline as everyone can see and they've perfected their branch predictor. Bulldozer has a long way to go if it ever wants to compete with SB and beyond.
     
  16. sergionography

    Joined:
    Feb 13, 2012
    Messages:
    284 (0.24/day)
    Thanks Received:
    40
    having a long pipeline has nothing to do with olipc and doesn't mean it will have low ipc, its just that long pipelines allow for good scaling at high frequency, and that approach is usually taken to make smaller cores with lower resources, this approach is done when companies have manufacturing restrains or when wafers are limited and what not, in amds case its a bit of both and its cheaper because the die ends up smaller than making a full 8 core CPU with all big cores
    Also its because amd is redefining cores while the shared resources in the modules are subject to be replaced with other solution, like the fpu unit to do floating point operations will be the job of the integrated gpu in the future, that explains why amd is focusing mainly on integer performance while on the gpu they are heading towards gpgph and x86 capability.

    As for piledriver I have to note that it's said trinity core are already 25% faster than llano, when the highest trinity is clocked at 3.8 and highest lano being 3.0 that suggests the frequency brings most of that improvement + like 5-7% ipc assuming these numbers are accurate piledrivef is already 15%-20% faster than bulldozers ipc, and add to that it's ability to clock higher hopefully then there u got something
     
  17. Aquinus

    Aquinus Resident Wat-man

    Joined:
    Jan 28, 2012
    Messages:
    7,411 (6.12/day)
    Thanks Received:
    2,839
    Location:
    Concord, NH
    Yes, but branch prediction on a longer pipeline is harder to make and harder to correctly predict. Lets say you miss-predict an instruction, now you have to dump EVERYTHING in the pipeline and start where you're supposed to be, so every extra stage you have you will have a performance hit. Intel has a small pipeline and have had this design since the Core/Core 2 series, 14 stages. All it takes is 1 miss-prediction on the branch predictor to make a large-pipeline system have garbage for performance. On top of that AMD's L2 and L3 are too big. If you read up, Intel's L3 is actually multiple L3s connected using a ring bus. So smaller segments to access have lower latency.

    Also check out Intel's changes to their CPUs. Faster cache, improved branch prediction, power reductions. What is AMD doing? Focus on heterogeneous computing. The CPU market isn't quite ready for that.

    Edit: Unless you're running a server. As a system admin I would *love* to get my hands on an Interlagos-based server.
     
  18. sergionography

    Joined:
    Feb 13, 2012
    Messages:
    284 (0.24/day)
    Thanks Received:
    40
    well that is true but phenom for example was in order execution and had a shorter pipeline aswell just like intel but now bulldozer has out of order execution so that helps a bit, but over all your right, all these things need some good fine tuning which have huge impacts hence what we are hearing about trinity along with its improvements, not to mention amds years of R&D investing into bulldozer(if there wasnt something realy brilliant about it they wont even bother), bulldozer is a great architecture over all its just that amd failed at the execution mainly because i think they rushed it or were forced to do so(i have a believe that if you think rather small or realistic and succeed its better than thinking big and planning big then end up failing or doing a half job, which is what amd did), i think what they shouldve done is take the sharing thing slowly with less sharing on their first iteration untill bulldozer overall becomes familiar and that way they wont do such a dramatic change all at once because that has its downsides and with amd being in a really tough spot that makes it worse, but what do you expect from a company that didnt have proper management at the time, but owell i guess it is what it is, as for heterogeneous computing that is for 2014 and on, by then it should be doing good, not to mention amd is slowly moving there which. doing this will allow more potential out of hardware (now adays hardware is way ahead of software, as most software cant even utilize an old core2quad or phenom x4 properly with its full potential) just look at game consoles, they still come up with new titles that run on over 6-7 year old hardware while still being fairly good overall, while on computer you cant even run those games even on low res with hardware that old because most computer games are ported from console so they run less efficiently on it, heterogeneous computing in a way is a new standard presented by amd to software designers which helps utilize for the hardware.
     
  19. Aquinus

    Aquinus Resident Wat-man

    Joined:
    Jan 28, 2012
    Messages:
    7,411 (6.12/day)
    Thanks Received:
    2,839
    Location:
    Concord, NH
    The Phenom series had a better IPC, however things need to be put into perspective. Bulldozer's "cores" aren't as efficient (IPC wise,) however you have more cores. This prevails in heavily multi-threaded tasks such as video encoding and benchmarks show the FX-8150 to be strong in these areas. The problem you run into is when you have an application (games are great examples of this being the primary issue for bulldozer) that only utilizes a handful of threads, you have the i7 prevailing because of the IPC.

    Bulldozer didn't fail, it's just geared towards servers more than anything, because in server applications Valencia and Interlagos run without skipping a beat because you need to handle multiple clients (multi-process by design in most cases,) but for newer games on video cards as fast as the 7970, you will see (like on Crysis 2,) that the CPU benefit between the i7 2600k against the FX-8150 is minimal because it can harness the power bulldozer has.

    Now to wrap up everything with one last piece of brain-food, if you have a Bulldozer CPU, watch your CPU utilization on applications that run relatively poorly (not to say it is running poorly, but relative to the 2(5/6)00k I'm willing to bet that the Bulldozer has more CPU time available in the background than Intel's chips do, which allows you to do things like video encoding while you play a video game. That is AMD's focus, and with Bulldozer looking strangely like a GPU design, I would guess that this is the first step towards a heterogeneous computing platform where separation of GPU and CPU (even at the die level,) doesn't exist, but that is incredibly far down the road, even now with bulldozer being a "flop."

    As a system administrator, on a workstation I would rather have an FX-8150 because of the benefits 8-cores bring to a developer or admin. If I were gaming, I would rather have an i5 25(5/0)0k even though the i7 has more power, there are some cases where the i5 is faster than the i7 and I'm willing to bet that has to do with the size of the L3 cache (smaller tends to have lower latency), and if the machine did both, I would want an i7.

    Edit: Out of order execution has been around for a long time and it really depends on the CPU implementation weather it needs to have it or not (with x86 it's practically a given.)

    Edit 2: Once again I will re-iterate that the FX-8150 is adequate (but only adequate) for single tasks, but when you throw multi-tasking and multi-threading at it, that price point looks much better and much more reasonable. Also I will once again quote Guru3d's Hilbert Hagedoorn, "If anything, this little article proofs once again that investing money in a faster graphics card will gain you better game performance compared to investing in a faster CPU. The performance difference in-between a 1000 USD Core i7 3960X compared to a 320 USD Core i7 2600K processor is extremely small, something you'd never notice unless measured. So we say, stick to a modern mainstream quad-core processor and the differences really aren't that big in the overall framerate, especially at 1920x1080/1200. Yes we know it's that weird penumbra, the higher you go in resolutions, the slower your processor may be. Remember, once you pass 1920x1080/1200 the GPU is almost always the bottleneck, not your processor." Feel free to check out the CPU scaling review with the 7970 here: http://www.guru3d.com/article/radeon-hd-7970-cpu-scaling-performance-review/1
     
    Last edited: Feb 22, 2012
  20. sergionography

    Joined:
    Feb 13, 2012
    Messages:
    284 (0.24/day)
    Thanks Received:
    40
    well I mentioned bulldozer was a fail because it didnt meet amd's expectations, a bulldozer module is much buffer than a phenom II core in certain areas, and when it isnt sharing resources the idea was to get the benefits of a full core(bigger than that of phenom)
    however bulldozer is somewhat more efficient aswell considering some benchmarks where phenomII had more hardware for the required tasks yet bulldozer performed the same(im not sure if it was the alu's or something)
    but overall i would say it just wasnt fine tuned enough
    but i agree with you that bulldozer at its current state is good at servers and multitasking, but amd had more in mind, high frequency was one way to achieve good single thread, but amd had more in mind.

    ive seen many reviews with people running bulldozer with every other core disabled(running each module as if its a core) and performance gains in ipc where impressive if you ask me, however amd didnt advertise that much because they expected higher frequencies with turbo, to were running 2 threads on a module allowing turbo to kick in will gain more performance than running the threads on 2 different modules, so instead of having 20%higher ipc amd will simply increace the frequency by 20% or more(however that goal was not achieved well as using turbo with one module or no turbo on 2 modules gave the same results)

    also another note about the long pipeline and high latency is to allow the sharing, meaning while one integer core is idle waiting for the next cycle the other one would be receiving data from the shared resources in the module, this is how bulldozer achieves that 80% performance of a real core(or 80% the performance of the module) but like you said mispredictions and timing does cripple things a bit so amd should work on getting the perfect harmony between the hardware
    once amd get things right with this architecture then you will get the perfect blend of multithread and single thread performance because if you ask me bulldozer is a very interesting architecture, if its done right non of the cpu parts would be running and wasting energy between the integer core cycles without doing any productive work
     
    Last edited: Feb 23, 2012
  21. Super XP

    Super XP

    Joined:
    Mar 23, 2005
    Messages:
    2,807 (0.76/day)
    Thanks Received:
    550
    Location:
    Ancient Greece, Acropolis
    Holly Reading lol, both of you make great points.

    Bulldozer is not a fail, it's just geared more for servers. AMD made an error with this, though I can see Piledriver fixing this and performing much better with even better thermals.
     
    sergionography says thanks.
  22. Aquinus

    Aquinus Resident Wat-man

    Joined:
    Jan 28, 2012
    Messages:
    7,411 (6.12/day)
    Thanks Received:
    2,839
    Location:
    Concord, NH
    Thank you. :) I don't think piledriver's IPC will improve enough to catch up with IVB, however the more they optimize a module, the more cores they can shove on the same area. Multi-threading is the future. :)
     
  23. eidairaman1

    eidairaman1

    Joined:
    Jul 2, 2007
    Messages:
    14,166 (4.92/day)
    Thanks Received:
    2,052
    We dont know anything at this point till its launched, n who knows Piledriver mayhave been in the works since bulldozer was launched n AMD market dept had no choice but to unveil Bulldozer.
     
  24. Aquinus

    Aquinus Resident Wat-man

    Joined:
    Jan 28, 2012
    Messages:
    7,411 (6.12/day)
    Thanks Received:
    2,839
    Location:
    Concord, NH
    Piledriver isn't introducing major architectual changes, so it isn't possible for the IPC to me improved by that much unless the branch predictor and cache latencies really were worse than we previously thought.

    True, but this is just another revision of the BD architecture so we can make some guesses.

    It was, just like how Bulldozer has been in the works for the last 5 years or so, it takes a lot of time to develop a CPU.

    Because SB came out? Maybe, but this was a huge project that was going to take multiple cpu iterations to work out the kinks. Saying that AMD has no choice is completely correct. Why would you develop a new CPU architecture and not release it? I'm not convinced you know what you're talking about. I would do some more research if I were you.

    Honestly, if you learn what is actually going on in a BD module, you will see that each core simply doesn't do as many instructions per clock because of the hardware inside each module. The Phenom II was capable of 4 ops per clock per core, where BD only does 3 ops per clock per core. Now overall, the theoretical performance of 4 ops per clock on 4 or 6 cores is lesser or similar to 8 cores with 3 ops per clock, but this is only one of the many changes that alters how a CPU reacts under certain conditions and for single-threaded applications performance suffers. However if you look at all modern applications designed in anno 2011/2012 you will see that threading is becoming more and more prevalent and CPUs with more cores will validate where BD shines.

    Once again, I've done a ton of research on Bulldozer, and it isn't a bad platform, it's just not designed for games, and certainly not games from a few years ago. However you will see games like Crysis 2 which can utilize 8 threads and you will see next to no improvement at resolutions of 1600x1200 and higher.

    Reading benchmarks doesn't make you knowledgable of a platform, once again, do some research about what BD actually is, how it works, and why it suffers under certain workloads.

    I will wrap this up by saying that AMD's total assets are just shy of 5 billion USD with approximately 11 thousand employees in 2010. Intel's total assets (as of 2010,) were over 71 billion dollars with just over 100 thousand employees. So tell me, who has more resources to develop a faster CPU? All things considered, I think AMD is doing damn well.

    Cheers.
     
  25. sergionography

    Joined:
    Feb 13, 2012
    Messages:
    284 (0.24/day)
    Thanks Received:
    40
    yes but amd was in such a bad shape financialy so 5 years for them wasnt even enough r&d


    in my opinion piledriver will actualy be what amd expected of bulldozer (a sandy bridge killer) it was only when amd got the real silicon that they realized this wasnt possible, thats y it was too late for the PR to go back, but my point is the engineers expected alot from bulldozer and im sure they did for a reason and im not only talking about multithreading performance.
    however since ivy bridge is only a die shrink from sandy bridge maybe that will give amd some room to breath

    i encourage you to read these details again, it is phenom II that does 3 instructions per cycle while bulldozer does 4, however those 4 are shared between 2 integer cores, so in theory a bulldozer module is 25% faster than a phenom II core, and with the ability to reach higher frequencies, sandy bridge on the other hand does 5 instructions per cycle
    thats in theory, however in reality both phenom II and sandy bridge are done so well that in practice they almost perform just like in theory, bulldozer on the other hand doesnt. however the fact that each integer core in practice performs better than a 2 instruction per cycle core in theory(considering 4ipc divided between 2 cores makes 2ipc per core) and closer to a 3 instruction per cycle core(10% slower than phenom II ipc) tells you that they are using less hardware for more performance which is a good start
    but if they get the architecture closer to theory and the sharing becomes almost perfect in timing and whatnot, then you will get 4ipc per core with like 60% less hardware because this hardware in typical cpu's stays idle consuming enegry waiting for the next cycle on the integer core.
    if that ever happens then clocking bulldozer 20% higher will match it with sandy bridge, but with less hardware meaning the ability to pack more cores in the die(that in reality this advantage will only even things out as intel almost always has the advantage in the manufacturing process)
     

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)

Share This Page