AMD: We Are On Track With Steamroller Micro-Architecture in 2013.

Discussion in 'Frontpage news' started by Pill Monster, Apr 1, 2013.

  1. Pill Monster

    Pill Monster Banned

    Messages:
    25,211
    Likes Received:
    9
    GPU:
    7950 Vapor-X 1100/1500
    AMD Implies on Plans to Deliver Opteron Chips with Steamroller Cores This Year

    Source: Xbitlabs


    Advanced Micro Devices has released a presentation for investors that puts the launch of the new Steamroller micro-architecture in 2013. While the company did not unveil what chips will feature Steamroller high-performance x86 cores this year, it implied that the new generation of server-class Opteron chips will be based on Steamroller, the micro-architecture AMD puts a lot of hopes on. A Curious Case of Steamroller Plans

    The Sunnyvale, California-based company published a slide called “AMD Opteron Technology: Delivering multiple generations of greater functionality and improved performance” in its Q1 2013 investor presentation, which clearly puts release of Steamroller in 2013. In the previous version of AMD enterprise roadmap, the company clearly stated that its Opteron “Abu Dbabi” powered by Piledriver high-performance x86 cores will be its focus for 2013 and 2014. The next-generation of Opteron processors was planned to be introduced for the second half of 2014.
    [​IMG]
    As it appears, AMD has either dramatically changed its Opteron plans and the new chip will actually be launched this year (which either suggests compatibility with current sockets or means that AMD intends to unveil a new platform for server microprocessors with support for PCI Express 3.0 and improved functionality), or just wants to show that it is on-track with the micro-architecture, which will power different chips.
    Earlier this year AMD updated its roadmap for client-class personal computers. It reiterated plans to launch its code-named Kabini and it is a new-generation accelerated processing unit with Steamroller x86 general-purpose core as well as Radeon graphics engine based on GCN architecture.
    AMD Steamroller: Expectations

    AMD revealed a lot of details about the Steamroller at Hot Chips conference in August, 2012. Just like in case of the Bulldozer architecture, Steamroller x86 cores - which will power AMD's future high-performance Opteron and FX chips - will be located inside dual-core modules and therefore processors on its base should be similar by design with Orochi and Viperfish, with some minor exceptions that will not be truly important (new memory controller, different internal buses additional tweaks, etc) foe x86 performance. The main improvements will be independent instruction decoders for each core within a module, better schedulers, larger and smarter caches, more register resources and some other enhancements.
    [​IMG]
    One of the reasons why dual-core Bulldozer modules [the same may be said about Piledriver] are not completely efficient is because they have only one instruction decoder for two ALUs and one FPU. With steamroller, AMD not only incorporated two decoders per module, but also increased instruction cache size (to lower i-cache misses by 30%), enhanced instruction pre-fetch (the number of mis-predicted branches is down by 20% compared to Bulldozer ) as well as improved max-width dispatches per thread by 25%. AMD believes that Steamroller will provide 30% improvement in ops per cycle.
    [​IMG]
    AMD also advanced single-core execution by implementing 5%-10% more efficient scheduling, incorporated higher-capacity register files and performed some other tweaks. It should be noted that while integer pipes of Steamroller will not be too different from existing ones, the floating point pipe will be a bit redesigned. In general, AMD promises that both integer and floating point per-core performance of Steamroller will be higher than they are today with Bulldozer micro-architecture.
    [​IMG]
    One of the interesting features of AMD Steamroller will be its ability to disable unused parts of L2 cache. Since not all apps are cache-bound, this may result in decreased power consumption and/or AMD's ability to boost clock-speeds of its microprocessors dynamically.
    [​IMG]
    It is noteworthy that AMD decided to talk about its Steamroller micro-architecture that will be utilized inside microprocessors made using 28nm process technology approximately a year or more ahead of their roll-out​
     
    Last edited: Apr 1, 2013
  2. IcE

    IcE Don Snow

    Messages:
    10,693
    Likes Received:
    79
    GPU:
    3070Ti FE
    When are they going to roll out Pile Driver based APU's with GCN onboard graphics? That would be a real bank earner for them.
     
  3. BLEH!

    BLEH! Ancient Guru

    Messages:
    6,402
    Likes Received:
    421
    GPU:
    Sapphire Fury
    Good for AMD. Gunna build a mini-ITX build using one of their APUs soon.
     
  4. -Tj-

    -Tj- Ancient Guru

    Messages:
    18,097
    Likes Received:
    2,603
    GPU:
    3080TI iChill Black
    Now this will be the real deal, esp. those single threaded improvements. :)
     

  5. PNeV

    PNeV Guest

    Messages:
    2,069
    Likes Received:
    2
    GPU:
    MSI GTX 1070 Armor
    I too would like one of their Kaveri APU's, would make an awesome PC to put under my TV!
     
  6. mohiuddin

    mohiuddin Maha Guru

    Messages:
    1,007
    Likes Received:
    206
    GPU:
    GTX670 4gb ll RX480 8gb
    When will steamroller come for desktop?
     
  7. deltatux

    deltatux Guest

    Messages:
    19,040
    Likes Received:
    13
    GPU:
    GIGABYTE Radeon R9 280
    AMD's only problem in their Bulldozer designs have been single threaded performance the entire time. Once they solve it, the FX would be a true competitor to Intel's offerings, assuming Intel doesn't figure out a new way to leap frog AMD's performance.

    The FX's multithreaded performance is known to be very good, so all there is left is the single threaded performance.

    They also need a new chipset to support PCIe 3.0, I know we don't need PCIe 3.0 but there's a lot of marketing value to it because most mainstream people only look at the name of the technology, if it looks new, they'll buy it. Maybe the new chipset could support a newer HyperTransport speed too.

    deltatux
     
    Last edited: Apr 1, 2013
  8. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    Waaaaaay too optimistic. That's the same thing as saying once they stop being bad they'll no longer be bad. To "fix" their criminally stupid single threaded performance they'd have to make a new architecture. There is no real solution using the nutjob architecture they chose in favour of advancing K10. Just imagine if they had stuck with K10, Intel would have actually had some competition right now.

    With Bullpile they'll never catch up to Intel. I'd bet Steamroller won't even catch up to Ivy Bridge, and Haswell will already be out. It'd be optimistic to imagine they'll catch up to Sandy Bridge.

    Edit: Oh and "Q1 2013" my foot. That would mean they would have to release it this month at latest.
     
    Last edited: Apr 1, 2013
  9. IPlayNaked

    IPlayNaked Banned

    Messages:
    6,555
    Likes Received:
    0
    GPU:
    XFire 7950 1200/1850
    Hey we got a CPU engineer here! Just how much does Bulldozer/piledrivers relatively poor branch prediction affect the single threaded performance, and what would you do to remedy that?

    It didn't say anything about a Q1 release. It said that this was mentioned in a Q1 investors' meeting.
     
  10. Taint3dBulge

    Taint3dBulge Maha Guru

    Messages:
    1,153
    Likes Received:
    12
    GPU:
    MSI Suprim X 4090
    Come on now, you know that the piledrivers are hand n hand with almost all aps and games compaired to the i5 and just a bit behind on the i7 but selling for the price of a i3..

    I know you want to think that intel is beyond better but truth to the matter is right now the chips arnt that far apart.. right?
     

  11. deltatux

    deltatux Guest

    Messages:
    19,040
    Likes Received:
    13
    GPU:
    GIGABYTE Radeon R9 280
    Ivy Bridge isn't THAT much better than the FX series, only single threaded performance is lacking in FX CPUs at the moment, and in gaming, they're pretty much in the same ball park, except for in dual threaded games like Starcraft II or older games. With multithreading, depending on the application, the FX beats out the Core i5 which is in the same price range.

    Once AMD keeps on tweaking their architecture, they're gonna have a good competitor. Remember, AMD isn't going after the performance crown and for the price that they are selling it for, it has great potential to beat out the Intel equivalent for the same price in future revisions.

    By making the instruction decoder separate for each core inside each module, instructions would be decoded faster since the same instruction decoder don't need to feed two cores. I'm quite optimistic with Steamroller. If AMD can really hit the 30% performance increase, then AMD can catch up, but that's all on paper until we test them.

    Honestly, if Bulldozer was what Piledriver is now, I wouldn't have bothered with switching over to Intel.

    deltatux
     
    Last edited: Apr 1, 2013
  12. Chillin

    Chillin Ancient Guru

    Messages:
    6,814
    Likes Received:
    1
    GPU:
    -
    This is just an April Fools.
     
  13. Noisiv

    Noisiv Ancient Guru

    Messages:
    8,230
    Likes Received:
    1,494
    GPU:
    2070 Super
    Single threaded and Power Consumption.

    I know plenty of you guys think that power consumption is overrated, but that alone is what gets you in or out of the laptops, you know... that thing that gets you lots of money
     
  14. DSparil

    DSparil Guest

    Messages:
    3,295
    Likes Received:
    33
    GPU:
    GeForce RTX 3080
    I really look forward to Steamrollers desktop release! If they deliver on the promised performance benefit, it would be a great leap from Bulldozer. Probably wouldn't be worth it to go from Piledriver to Steamroller though.

    Way too optimistic? Not at all and I completely agree with Deltatux. Where AMD has been hurting is single threaded performance, and once they shore up those issues, AMD will be golden. You don't think Steamroller will "catch up" to Ivy Bridge? News flash, Piledriver has already done that! In fact, as Delta mentioned, Ivy Bridge i5's are already being beaten in many benches with Vishera chips that cost less $$. Haswell may be another story, but we'll see.

    lol, I hope not!
     
  15. deltatux

    deltatux Guest

    Messages:
    19,040
    Likes Received:
    13
    GPU:
    GIGABYTE Radeon R9 280
    Their APUs don't really consume too much power, their TDP is around 17-35W which is about the same as Intel's TDP (not including their U series Core i3s and i5s)...

    Only on desktops do AMD CPUs take up more electricity but you won't really see the difference in the electricity bill, around a couple dollars difference per year.

    deltatux
     

  16. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    Which is the most important and precisely what I'm talking about. It's so behind that even with twice as many "cores" at a significantly higher frequency it can't beat Intel's offerings.

    It seems people want to plug their ears and not listen, pretty much everything is still single and dual threaded. Ever had an installation take a long time even on an SSD and wondered why it took so long? Single thread processing limit. People see 12% of the CPU being used and don't realize it's 1 thread maxed out and its time slices distributed among all 8 threads; it's still limited by the maximum of a single thread.

    The only way they'll achieve a 30% increase in performance is with increased clock speeds as well. It will not be anywhere near 30% faster clock for clock. I would be surprised if it was even 15%. Steamroller will still have only one FPU per module, they're still half-assed cores. They will have subpar performance in most programs which are dual threaded, they will have abysmal performance in single threaded programs and AMD will cherry pick artificial benchmarks and anomalies to make it look like an adequate product for the price. AMD fans will vouch that these FX chips are excellent, it may server their purposes, but I prefer the highest single thread performance I can get.

    In what? Some niche market that exists in someone's imagination? For my purposes, high performance per thread, and maximum performance in the $250-$400 range the only option has been Intel.

    You make it sound like I'm happy with this monopoly, at the very least the next chip I get is going to cost me $350 before taxes just because there is no competition. When overclocking is taken into account the gap increased even further. Even if AMD did manage to offer something in that price/performance range that gave overall the same performance, that would still be with 8 threads, I'm bound by too many single threaded programs.

    Let's see some of those benchmarks... the most powerful AMD chip that exists versus the standard 3770K - http://www.anandtech.com/bench/Product/697?vs=551

    4GHz 8 cores vs 3.5GHz 4 cores, as you can see, for me AMD is not an option, especially since I overclock.

    You have no idea how much I wish I'm wrong and AMD come through with some miracle, I'm just saying don't keep your hopes up. I would be shocked if AMD deliver half of what they promise.
     
    Last edited: Apr 1, 2013
  17. deltatux

    deltatux Guest

    Messages:
    19,040
    Likes Received:
    13
    GPU:
    GIGABYTE Radeon R9 280
    Well, NTFS is a multithreaded file system, so are many other UNIX file systems, so I/O access isn't single threaded, that's for sure. The only driver I know that's single threaded is the open source edition of ntfs-3g, Tuxera intentionally nerfed the performance just so that you'd buy their proprietary Linux NTFS drivers.

    Where single threaded performance is very important are mainly office productivity and older gaming engines. Even web browsers like Google Chrome are multithreaded thanks to its multiprocess sandbox.

    As for clock for clock performance, it is true that it takes a higher clock speed to achieve the same performance in terms of single threaded applications, but with highly threaded applications, the more integer cores the better. Remember the FPU is really 2 128-bit FMACs, that means technically both cores can lock into one of the 2 128-bit FMACs, only when it needs to process a 256-bit instruction like AVX is when it needs to share. Honestly, most software do not take advantage of 256-bit floating point instructions.

    The 30% increase is likely through slightly higher clockspeeds and a much improved uncore as Steamroller slides have shown.

    As single-threaded performance is starting to get replaced by multi-threaded performance, the case of the FX is stronger. I find that Ivy Bridge is still the most rounded micro-architecture, but I cannot discount the potential that the Bulldozer family of micro-architectures hold. AMD acknowledged that it's too multi-threaded bias right now and have been working to remedy that with these revisions and it's just going to serve well for the entire market in the future.

    Many gaming engines and generic software are going multi-threaded, at least AMD has gotten that covered, albeit a bit too early. My next rig is likely going back to AMD as what I do on my rig benefits when more cores are thrown at it.... unless Intel finds a way to leap frog AMD again.

    deltatux
     
  18. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    I didn't even think of that. But what I fear is that the most likely case of often using AVX in the near future will be when the CPU is stressed to the maximum... on two cores; emulators.

    I'm currently limited by dual threaded performance in emulators and I'd be afraid of using an AMD CPU for the same reason.

    When that scenario does roll around, one module's FPU is used entirely for AVX, it's not going to move onto a core from the next module to make use of another FPU, it's going to use both cores off of 1 module and become bottlenecked.
     
    Last edited: Apr 1, 2013
  19. Pill Monster

    Pill Monster Banned

    Messages:
    25,211
    Likes Received:
    9
    GPU:
    7950 Vapor-X 1100/1500
    @Neo- why are you so mad at AMD?
     
  20. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,780
    Likes Received:
    1,393
    GPU:
    黃仁勳 stole my 4090
    Sad, Pill, not mad, just sad.

    Edit: Unless you mean their GPU department as well, in that case I'm mad. :wanker:
     
    Last edited: Apr 1, 2013

Share This Page