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

Why doesn't AMD create a better CPU architecture to compete with Intel again?

Discussion in 'General Hardware' started by qubit, Jan 18, 2014.

  1. qubit

    qubit Overclocked quantum bit

    Joined:
    Dec 6, 2007
    Messages:
    10,864 (3.92/day)
    Thanks Received:
    4,212
    Location:
    Quantum well (UK)
    AMD's siamesed cores module architecture has been used for its CPUs since Bulldozer was released over two years ago in October 2011 to disappointing reviews.

    It's lead to AMD humiliatingly losing the performance war* against Intel with this design, so I've been wondering why they persist in using it to this day? They've tweaked it a bit here and there, but since it's been over two years now, surely they could design a better architecture instead and take a leaf out of Intel's book to make performance competitive again while being careful not to step on their patents?

    For all the usual and obvious reasons, I'd much rather see a strong AMD competing head to head with Intel every generation, rather than this performance monopoly that Intel enjoys.

    EDIT: To clarify, what I meant was that AMD lost the performance war when Intel released Conroe way back in 2006, but remained reasonably competitive, with aspirations of matching Intel's performance someday. However, the Bulldozer release cemented just how hopelessly far behind AMD were, completely humiliating them, with the gap only growing wider as time moves on.
     
    Last edited: Jan 22, 2014
  2. Solaris17

    Solaris17 Creator Solaris Utility DVD

    Joined:
    Aug 16, 2005
    Messages:
    17,722 (4.91/day)
    Thanks Received:
    3,990
    Location:
    Florida
    Wait AMD was aiming to beat Intel in performance?

    source?

    To be fair. I think alot of it has to do with R&D funds but more importantly I think certain types of architecture and transistor setups are actually owned intellectual property. and theirfor cant be used by someone else. Which probably explains why AMD hasnt used scalar architecture like nvidia and stays with supar scalar. Its more of a pigeon holed market from what I see.
     
    qubit says thanks.
  3. Steevo

    Steevo

    Joined:
    Nov 4, 2005
    Messages:
    9,000 (2.55/day)
    Thanks Received:
    1,621
    No, its more that AMD has had a ass backwards idea for years now and since they have devoted years of R&D time to it they are going to follow through, and they have an idea that it will cause them to win. Considering the slight IPC improvement in the latest offering and how their chips are now in consoles, and they are working on the software (so it seems) to make their hardware win.........I hope it works out for them only so there is good marketable competition again.


    Making a CPU is far beyond a transistor does A, or can add A plus B. It falls back to the amount of time electrons are moving through the traces to get from cache to the actual transistor, and then back. t his is why AMD chips have latency issues, and why they suck at non pipelined or heavily threaded pipelined instructions. Or instructions that don't fit in the pipeline that can be spoon fed to the CPU cores.


    They keep hoping for a software revolution to use GPU cores for these jobs, and for pure speed (Giggle Hurtz) to overcome their failings in IPC.
     
    10 Million points folded for TPU
  4. Fourstaff

    Fourstaff Moderator Staff Member

    Joined:
    Nov 29, 2009
    Messages:
    9,276 (4.54/day)
    Thanks Received:
    2,023
    Location:
    Home
    I think AMD has bigger worries: ARM processors. I can see tablets and phones replacing casual use computers, and it wouldn't be long before we get full "Office" suites, which will signal the start of office computers abnadoning x86. The transition is not going to be fast (imo fastest transition is 10 years at least), but once it happens x86 will be pushed straight into niche. Even servers are using graphics cards for their main firepower.
     
  5. qubit

    qubit Overclocked quantum bit

    Joined:
    Dec 6, 2007
    Messages:
    10,864 (3.92/day)
    Thanks Received:
    4,212
    Location:
    Quantum well (UK)
    I'm just musing, Solaris.

    Back in the day AMD certainly were competing with Intel for performance, as any rival CPU manufacturer would want to do. Command the top prices and brand prestige and recognition all go with it, obviously. Think back to 2005 when the Athlon 64 ruled the day. I bought one and then the dual core version, which were fantastic for the time.

    Instead, they seem to have just resigned themselves to this fate of underdog and have gone for the "value"market instead. They don't like to be there and neither do we. It sucks.

    Nice avatar. :cool:

    EDIT: thanks for your replies steevo, fourstaff. Saw them after I posted. Making a competing product certainly isn't a trivial matter, that's for sure.
     
    Solaris17 says thanks.
  6. micropage7

    micropage7

    Joined:
    Mar 26, 2010
    Messages:
    6,677 (3.46/day)
    Thanks Received:
    1,609
    Location:
    Jakarta, Indonesia
    personally i just feel ok with AMD processor performance but i just expect AMD would squeeze their power consumption.
    it would be great
     
  7. Solaris17

    Solaris17 Creator Solaris Utility DVD

    Joined:
    Aug 16, 2005
    Messages:
    17,722 (4.91/day)
    Thanks Received:
    3,990
    Location:
    Florida
    The big problem that I see is during the intel change in power around the penryn era AMD was looking total collapse in the face and Intel literally had billions to burn on R&D and they already had a solid product using P3 pipeline arc (slightly more advanced mind you) while AMD was attempting to still BE a company they burned what they could into the buyout of ATI to float on GPUs for a bit. They lost their job a week before the rent was due. Its been a long time coming for sure. but Intel just has the R&D lead. You have to understand intel shows us some pretty roadmaps but to the bearded elders with high level security badges that power point slide is old.

    Regardless of the flack I have given and AMD in general has gotten their cards are coming around to be more nvidia contenders then they used to when ATI lost that race a little after the 9800 years. I think they are taking the right course since it seems they are dumping all the improvements on GPUs atm. Intel has never had good performing IGPs its a fact that their arc simply sucks. AMD has the right of it since they seem to be more focused on their GPUs. Integrating that tech into their APU lineup will probably beat intels APU options even though they cant do it with cores customers will get a more "even" usability experience with a proper balance between computing power and HD graphics. Atleast in theory. So they could very well pick themselves up on sales of APU units and GPUs while affording some time at intel's CPU heels for a bit longer.

    I might be onto something

    or completely talking out my ass. This is my interpretation of AMDs current scenario.
     
    HD64G says thanks.
  8. Steevo

    Steevo

    Joined:
    Nov 4, 2005
    Messages:
    9,000 (2.55/day)
    Thanks Received:
    1,621
    http://www.techpowerup.com/forums/t...ache-and-memory-benchmark-here.186338/page-10

    http://en.wikipedia.org/wiki/Branch_predictor



    AMD
    L2 10ns = 10 billionths of a second = 10 CPU cycles of stall for any instruction not found in the L1 cache minimum. If this was a 1Ghz processor and you had 100% cache miss it would be as effective as a 100Mhz processor that was 100% effective. ** Not exact math, and since the instruction may be found in the L3 cache even a 100% failure of the L2 will not result in a failure to process data, it will occur at a very slow rate, like one in 33 CPU cycles give or take a few.**
    L3 33ns = 33 billionths of a second = 33 CPU cycles of stall.


    VS

    Intel

    L2 3ns
    L3 10ns

    Intel HT uses pipeline stalls to feed the "core" instructions waiting in the cache while the transfer from the first thread occurs. So instead of an actual 10ns penalty, the core still may use 6 of those cycles on another thread to improve total instruction per clock.

    In short, AMD processors spend most of the time waiting for instructions when they don't meet the IPC that Intel does. However the longer pipeline sometimes (server applications with large data sets and small common requests) is better, as the when the data is found in the cache or there is no possible way to have all the data in the cache, they perform on par with Intel.
     
    Last edited: Jan 18, 2014
    qubit says thanks.
    10 Million points folded for TPU
  9. TRWOV

    TRWOV

    Joined:
    Aug 11, 2011
    Messages:
    4,105 (2.88/day)
    Thanks Received:
    2,837
    Location:
    Mexico
    AMD has the right ideas but they can't realize them at the right time.

    Kaveri should have been here like 3 years ago, not this week. Right now only a handful of apps are capable of taking advantage of HSA and not even to its fullest potential, it'll take at least a couple of years to have a software environment that matches the processor capabilities. AMD nerfed their CPUs expecting the GPU to take on the load but it took 3 years for their plans to come into fruition.

    As for the Athlon 64, AMD had the idea of northbridge integration early on and made it big with the A64 IMC but once Intel put all their resources into that avenue not only did they catch up but speed past AMD. AMD engineers could be extremely talented but Intel has like 100x the engineers AMD does. You just can't take on those numbers with talent alone.
     
    Crunching for Team TPU
  10. suraswami

    suraswami

    Joined:
    Apr 5, 2005
    Messages:
    6,328 (1.69/day)
    Thanks Received:
    877
    Location:
    Republic of Asia (a.k.a Irvine), CA
    I have a I7 based desktop and laptop at work and at home have FX 8320. I don't miss the I7 at home and don't miss the FX at work. At home FX gives me good gaming experience for cheap and thats what matters. You can keep comparing 200FPS to 300FPS and get giddy about it or stop worrying about e-penis and use your existing one for a glorious pee happiness!

    Future is GPU and multi-threaded computing and thats where AMD has their bite. Intel will only follow that (as always).

    For comparison sake lets talk about Gaming Consoles, AMD literally wiped out Intel from it. Will anyone talk about it?
     
    minx says thanks.
  11. qubit

    qubit Overclocked quantum bit

    Joined:
    Dec 6, 2007
    Messages:
    10,864 (3.92/day)
    Thanks Received:
    4,212
    Location:
    Quantum well (UK)
    TRWOV, yeah, I know Intel is much bigger and has much more resources, but this problem seems to be as much about vision, or lack of it, as competing with the bigger guy. I remember all the big layoffs AMD had a few years ago as the fallout from the Bulldozer disaster bit. That wouldn't happen at a company that was well run and had a slight hiccup.

    Sura: while framerate always matters, even if it's very high as you "can't have too much processing power" for games (one for a different thread probably) I think it's a good point about shutting out Intel from consoles. Excellent win there, however they did it.
     
    Last edited: Jan 19, 2014
    suraswami says thanks.
  12. TRWOV

    TRWOV

    Joined:
    Aug 11, 2011
    Messages:
    4,105 (2.88/day)
    Thanks Received:
    2,837
    Location:
    Mexico
    I don't think Intel worries about the videogame consoles. Those kind of large volume deals only help to pay the bills and keep the machine going but you hardly get a profit because of the razor thin margins. Intel isn't on the business of making even.
     
    Crunching for Team TPU
  13. newtekie1

    newtekie1 Semi-Retired Folder

    Joined:
    Nov 22, 2005
    Messages:
    21,323 (6.07/day)
    Thanks Received:
    7,253
    R&Ding a CPU architecture takes a large amount of time and a large amount of money. Intel has been improving on the same Nehalem architecture since 2009! And Intel has way more money than AMD does to R&D a new architecture.

    That is a great point, and AMD has to dedicate a large amount of FAB time to producing those console CPUs. My guess is that is the largest reason they aren't planning any new top tier PC processors. The FABs they would be using to produce those high end desktop parts are now dedicated to console CPUs.
     
    suraswami and qubit say thanks.
    Crunching for Team TPU
  14. FordGT90Concept

    FordGT90Concept "I go fast!1!11!1!"

    Joined:
    Oct 13, 2008
    Messages:
    15,123 (6.15/day)
    Thanks Received:
    4,735
    Location:
    IA, USA
    AMD has been in distant second place since 2006 and the Core 2 launch.

    They can't afford to develop a new microarchitecture nor can they afford to compete with Intel on the fabrication processes either.

    Put bluntly, AMD's odds of catching up with Intel again in the next two decades are virtually nil.
     
    qubit says thanks.
    Crunching for Team TPU
  15. TRWOV

    TRWOV

    Joined:
    Aug 11, 2011
    Messages:
    4,105 (2.88/day)
    Thanks Received:
    2,837
    Location:
    Mexico
    It wouldn't be so bad if IPC lagged by two generations or so but Steamroller is currently at what? Clarkdale/Lynnfield levels? AMD is basically 4 years behind Intel in terms of IPC.

    As some have mentioned, anything this side of Conroe is good enough (I know lots of people still on Q6600s) but for us enthusiasts and for some business that performance delta is a big issue.
     
    Crunching for Team TPU
  16. qubit

    qubit Overclocked quantum bit

    Joined:
    Dec 6, 2007
    Messages:
    10,864 (3.92/day)
    Thanks Received:
    4,212
    Location:
    Quantum well (UK)
    Ford, I remember how long it took them to develop this architecture - 4 years or something? Development hell for certain and likely the reason it came out the way it did, especially with poor managers at the top leading the company. A flawed design that they couldn't let go of and it cost them dear.

    While I agree that AMD is probably less likely to catch up with Intel in the real world than winning the national lottery at 13 million to 1, I still think that in principle they could make something competitive in performance with Intel, if they had the right leadership in place. That's why I said about taking a leaf from Intel's book to keep development costs down while being careful to avoid a patent lawsuit. Probably just getting rid of that stupid siamese core would do it. I seem to remember that their old Phenom processors still offer good performance today and that they even beat Bulldozer in certain benchmarks. I imagine that refining the Phenom design could really help them put out a decent processor.

    Note that by "competitive", I don't mean that it necessarily has to beat Intel, just offer really good performance, coming close to their products in each price range and put the hurt on Intel. Just look at what they did with their latest graphics cards: slap a decent cooler on them and they're an absolute corker! It's certainly not a one horse race in the graphics market which is as it should be.
     
  17. Zen_

    Zen_

    Joined:
    Apr 18, 2010
    Messages:
    496 (0.26/day)
    Thanks Received:
    113
    The market for high end desktops has pretty much shrunk to enthusiast and professionals, and even we / they don't need to upgrade every few years. It makes no sense for AMD to spend what little resources they have trying to compete in that sector when growth is in cloud servers, console APU's, and maybe mobile x64 if microsoft gets their act together. They need to expend resources on growth sectors, and try to make low end x64 and discrete video cards more profitable to survive as a company.
     
    Last edited: Jan 18, 2014
    suraswami and HD64G say thanks.
  18. qubit

    qubit Overclocked quantum bit

    Joined:
    Dec 6, 2007
    Messages:
    10,864 (3.92/day)
    Thanks Received:
    4,212
    Location:
    Quantum well (UK)
    Yeah, a fair point, but data centres will always want high performance processors and lots of them. They just seem to have given up and gone in a different direction, when I don't think they should.

    EDIT: I think it's a fallacy to think workstations won't need more CPU horsepower. More power leads to better kinds of apps and new capabilities. Otherwise with a "good enough" mentality, we just get stagnation.
    Yes, ARM is a very definite threat all right and Intel could be hit really hard too. It's inherently a better architecture than x86/64 and with similar "go faster" design tweaks as has been applied to x86/64 will really fly. Would it be a good idea for AMD to transition to this architecture over time as it gains traction in the high processing performance space, do you think?
     
  19. newtekie1

    newtekie1 Semi-Retired Folder

    Joined:
    Nov 22, 2005
    Messages:
    21,323 (6.07/day)
    Thanks Received:
    7,253
    AMD is shining in data centers right now. Data centers want great multi-threading at reasonable prices, and that is what AMD delivers. It is a lot cheaper to throw together a 32-Core machine, capable of supporting 30 Single-core or 15 Dual-Core virtual machines using AMD parts than Intel. Basically for the cost of a single 10-Core Intel Xeon data centers can buy two 16-Core AMD Opterons. Data centers are where the multi-core strategy pays off.
     
    HD64G and qubit say thanks.
    Crunching for Team TPU
  20. HD64G

    HD64G

    Joined:
    Apr 30, 2011
    Messages:
    190 (0.12/day)
    Thanks Received:
    84
    Intel couldn't give the console companys a nice solution because they haven't ant GPU to show them that it would compete 7850's power that PS4's iGPU has. So, they lost long before game started. On the other hand nVidia hadn't any CPU to show them. Only AMD had a solution for both. And with PS4's praphic power, the games we are seeing in the near future both on PCs and consoles will be outstanding. Not terrible ports anymore. Thanks to AMD mind you.
     
  21. Bones

    Bones

    Joined:
    Dec 14, 2013
    Messages:
    62 (0.11/day)
    Thanks Received:
    29
    Location:
    Alabama
    This goes along with what I was thinking.

    Just stating some thoughts here.

    Intel can do CPU's well, Nvidia can do GPU's well but neither can do both well. While AMD can't do either as well, they can do them well enough - This has them in a position to gain ground on Intel related to having the $$ to do more R&D. What I mean here is to increase the $$ they would have to do it in the first place and if they play their cards right, make something of it.

    One thing that's helped them is back when they bought out ATI.
    AMD made a good move when they aquired ATI and while during that time it was labeled a "Bad Move", it's paying off now - The APU's are the best of that kind you can get, their GPU's while not as powerful as Nvidia's are good enough and they clearly beat anything Nvidia has for Crypto coin mining - maybe it hasn't been noticed how their high end GPU's are selling out everywhere (But of course it has). The combined resources with the AMD/ATI allowed them to develop the APU and it's now being used for consoles and desktops/laptops.
    Remember some time ago when Sony was expressing an interest in making CPUs? Sony made a fortune with the Playstation and PS2 consoles with suceeding console models just adding to the pile and AMD has this basic approach in mind but reversed towards consoles. They don't have to compete directly against Intel in the desktop market there, instead they can concentrate on that market to regain financial ground. With the anticipated changes to come in the future with smaller, more powerful components in our setups, smartphones, consoles, ect, they have at least made a sideways move to be successful.
    I mean why go through them if you can go around them instead?

    Unless Intel has success with either creating GPU type hardware or even goes as far as to aquire a GPU manufacturer such as Nvidia. I don't see them catching up with their own brand of APU anytime soon and I also don't believe they are putting as much into it as AMD has been, or even worried about it so much ATM. You must also remember that performance desktop chips are only a small portion of business for both companies, AMD has far more to gain by going this route than Intel does overall, Intel would have to do one of the above to get going with them and as always it costs $$ to buyout a GPU company/develop GPU tech, do the R&D to make it work with their CPU designs, develop the fab/process for making them AND bring them to market as a sellable product.... AMD has already done all of this.

    It's water under the bridge for them, bought and paid for except to continue improving their APU's.

    AMD might not if ever catch Intel with their desktop chips but maybe they don't have to. With the recent release of consoles equipped with their APU's, trying to compete against them could actually be bad business for them vs the gains they stand to make by going with consoles as a large part fo their business. What's left is to see how the AMD equipped consoles do and how sucessful they are in the future...... And if they try again to run with Intel as they did once before.
    We'll just have to wait and see what the future holds.
     
  22. repman244

    repman244

    Joined:
    Apr 7, 2011
    Messages:
    1,135 (0.73/day)
    Thanks Received:
    463
    People have been saying this for years, but I don't see ARM being a big threat for a few years to come. I think I even saw a sort of "review" which compared the x86 to ARM (I think it was some Atom vs Krait and A15) and the x86 delivered equal performance. The reason IMO is that Intel or AMD never really focused on ultra low power segments - only recently they showed signs of interest.
    Another thing is that once you start implementing features into ARM processors for "desktop" use you won't see it as a low power part anymore - x86 have so many things integrated into these days and let's not even start with the instruction sets which ARM lacks.
     
  23. HD64G

    HD64G

    Joined:
    Apr 30, 2011
    Messages:
    190 (0.12/day)
    Thanks Received:
    84
    To add some thought, modules is the best way to keep efficiency high in server business which needs parallel calculation and execution. Desktop isn't up there yet as the software hasn't evolved as much but with HSA arriving it goes there faster than before. So, in long term planning ATI aquisition is a very clever plan. Now it is software devs turn to make it work. Some recent samples of HSA capabilities using Kaveri are VERY promising. Interesting times ahead.
     
  24. HammerON

    HammerON The Watchful Moderator Staff Member

    Joined:
    Mar 2, 2009
    Messages:
    6,757 (2.92/day)
    Thanks Received:
    3,877
    "Why doesn't AMD create a better CPU architecture to compete with Intel again?"

    Simple in my opinion, AMD does not have the resources to compete with Intel and so they do not try.
     
    Crunching for Team TPU
  25. Dent1

    Joined:
    May 18, 2010
    Messages:
    3,365 (1.79/day)
    Thanks Received:
    1,035
    They've created faster architectures than Intel in the past with less resources before.


    Because it isn't profitable. The idea is to increase market share, thus increase shareholder revenue. The best way to do this is to get the masses using their CPUs or APUs, rather than compete in a petty performance war.



    2nd place, but not distant. They had their moments of catch up. The Athlon X2 architecture speicficially the "Kuma" revision and the original Phenom I almost closed the conroe gap. The gap widened again when Intel moved to 45nm Wolfdale, with time AMD closed the gap again with their Deneb, Callisto and Regor. (Phenom II/Athlon II)

    Between 2006-2009, IPC wise AMD was only 1 step behind, but always caught up briefly. Since Intel's I-series they've been 2-3 steps behind IPC wise with no IPC catch up.


    They won't have to. In a decade IPC will be less important, multi threading support will increase and AMD's multi core or multi module approach should pay off.
     
    Last edited: Jan 18, 2014
    HD64G says thanks.

Currently Active Users Viewing This Thread: 2 (0 members and 2 guests)

Share This Page