Is it time to jump?

[JLangevin]

Ive built 4 systems this week, 2 of them were AMD, 2 of them were Intel, and 2 of the 4 were top end machines.

The high end Intel was an i7 2700k with dual gtx 580s, 8gb of Dominator RAM, and Sabertooth Z77 board.

3DMark11 scores stock were 11,415
3DMark11 scores with 580s @ 900mhz/1.1v were 12,231
3DMark11 scores with 580s OCd and Turbo on the i7 OCd to 4.6ghz were 13,014

The high end AMD was an FX-8150 with single GTX 680, 8gb Patriot Xtreme RAM and Crosshair V Formula board.

3DMark11 scores stock were 9,110
3DMark11 scores with 680 @ 1275mhz were 9,657
3DMark11 scores with 680 OCd and BD OCd to 4.0ghz were 10,071

For shits and giggles I stuck my own 680 in the rig for SLI on the AMD and the scores jumped from 10,071 to 11,102.

So what does that mean? Stock 580s in SLI on the i7 at factory speeds were faster then OCd FX and OCd 680s.


I am at a crossroads with my rig in my sig and am really thinking it may be time to make that leap to Intel, but damn if I didnt want to wait and give piledriver a chance... Im not a brand loyalist, I want to go with bang for the buck, and with the price of i7s really coming down, its becoming harder and harder to justify staying AMD.

The i7 rig I just mentioned I managed to pick up the Sabertooth Z77 for $189 and the 2700k for 289 as a combo deal at Microcenter...

Thoughts? (This isnt a bash on AMD post, I really like the brand, but I think they have let us down too many times to keep using our pales to save their sinking ship)

 

[JLangevin]

Oh, and I forgot to mention. I stuck both 680s in MY rig in my sig with the 965 Black Edition Phenom II and got a score of 10,003 with my OC at 4.1ghz. What a disappointment that 8150 vs 965 was so marginal.

One think I would like to point out is that the FPS on both the 680s and the 580s between the intel rig and the AMD rig were nearly identical, its the physics tests and combined tests where AMD is scoring half the FPS.

 

[EQWoody]

In my honest opinion. AMD has WAY to much catching up to do. The FX series from AMD is the highest end they make for desktops, the intel chip you tested isnt even top end anymore. It's high end, but it shows you that even as a higher end chip it still stomps AMD's highest end chips. I ran AMD for years. my favorite build was an dual core 939 chip.

My intel i3 2100 has showed benchmarks higher for gaming than most all the quad core amd chips.

/rantoff.

TLDR: Buy intel from now on if you want better performance.

 

[oldpablo]

Honestly, who cares about that score. Play some games, run some applications, don't just use an arbitrary score like 3DMark. Get a feel for which is snappier, which is smoother, which one you actually like using.

 

[SonDa5]

Was virtual mvp turned on the Z77MB? IF so that will boost up 3dMark11 score about 2,000 points.


I don't think your comparison is very good.

Would have been better if you ran single GTX680 in each system. 2xGTX580 in SLI will stomp single GTX680. (expected outcome)

 

[JLangevin]

Was virtual mvp turned on the Z77MB? IF so that will boost up 3dMark11 score about 2,000 points.


I don't think your comparison is very good.

Would have been better if you ran single GTX680 in each system. 2xGTX580 in SLI will stomp single GTX680. (expected outcome)

If you read all of my post you will see where the SLI 580s also stomped the SLI 680s on the AMD.

 

[Tsumi]

Well, 3DMark11 also takes into account how powerful your CPU is, and there is no contest that a 2700k is significantly more powerful than the FX-8150 in 90% of benchmarks/real case scenarios. Multithreaded, it's barely on par, occasionally beating the 2500k.

Real question is in games, and your average/minimum FPS.

 

[JLangevin]

Well BF3 does run pretty amazing on my 965be and 680 when both are overclocked. I havent done any FPS monitoring yet, but I can certainly see a difference when compared to my 580.

 

[SonDa5]

If you read all of my post you will see where the SLI 580s also stomped the SLI 680s on the AMD.

I read your whole post. The SLI GTX680 obviously has a problem.

Would have been much easier to compare single GTX 680 in each system.

 

[Pieter3dnow]

To be honest unless you get a "Bulldozer compiled" executable you are not going to see lightning fast benchmarks. I can only refer to the well known compiler crap where if you have an SSE extension for AMD it always goes back to default instead of fastest possible.

If you check phoronix (linux website) you can see that the difference (and yes sometimes able to get higher scores) is there...

 

[pelo]

Buy the Intel chip. [H] did a review on crossfire/SLi scaling and an OC'd 8150 will run into its performance ceiling way before an OC'd i7 SB. Not a big deal if you're doing single GPU gaming (unless it's a GTX690 or 7990) but it also showed that the 8150 ran out of steam in GTX580 SLi configurations.

If you're going to spend that much cash on SLi GTX680s then don't cap your frame rates with a weak processor. Piledriver will probably do better but the i7 Sandy can be had now and it will almost certainly still provide better FPS.

http://www.hardocp.com/article/2011/11/03/amd_fx8150_multigpu_gameplay_performance_review/

Some games are closer but nowhere does the 8150 actually win. Considering that comparison used a 2500K and civ5 can utilize more than 4 threads, a 2600K/2700K would absolutely slaughter BD. If you're doing any sort of gaming with high end GPUs you're almost always served better by an Intel CPU.

 

[buttons]

I just built a new AMD system and couldnt be happier. AMD 1045T, gigabyte 760g, 8gb of ram, 7970

3dmark2011: P8857
7970 1220/1575
1045T@3.99ghz

I think its tremendous bang for the buck. My motherboard/cpu were $99 combined after a $10 mail in rebate from microcenter.

 

[fullvietFX]

SLI 580s vs a single 680?

 

[MacLeod]

Intel is faster, no doubt. If you're building from scratch its the best way to go. You can save a few bucks going with a $160 8120 vs the $220 2500K (Newegg prices) and the $190 990FX Sabertooth vs the $250 Z77 Sabertooth but if you're not worried about a tight budget, go Intel.

That being said, aside from Civilization V, Arma and Starcraft 2, the differences in the rest of the games out there won't be enough to notice in real world game play. So while Intel is faster, you can still have a great performing gaming rig with an AMD proc at the helm especially if you overclock it.

If your current rig is getting the job done, you might as well wait a couple more months til we will hopefully be seeing some info on Vishera and how well it performs. It wont be an Intel killer but it might be a sizeable enough jump in performance that it would be a viable alternative. Especially considering you've already got a superb AM3+ motherboard. You're basically in the same boat I'm in and I'm gonna wait. If Piledriver is a flop, I'm jumping ship. If its a solid performer, Ill drop one on my beloved Sabertooth, overclock its nads off and enjoy.

 

[psy81]

I can definitely see where you are coming from. I am in a similar situation, stuck with my Phenom II 965 on an AM3+ board with no real upgrade path. If you feel the need for the best of the best, make the switch to intel since they will remain ahead of AMD for the foreseeable future. I was contemplating it for the longest time. The price of 2500K and a nice motherboard is extremely tempting. Personally, I don't need the best of the best but I do want something that is competitive and has a decent lifespan. I am going to hold off to see what piledriver has to offer (giving AMD one more chance) and then I will decide.

 

[Dan_D]

Honestly, who cares about that score. Play some games, run some applications, don't just use an arbitrary score like 3DMark. Get a feel for which is snappier, which is smoother, which one you actually like using.

This. 3D Mark scores are arbitrary. They are meaningless. They aren't even useful for comparisons between system configurations because the factors that determine the scores are not the same factors that determine the performance of games and other applications. CPU clock speeds can have a drastic effect on 3D Mark, yet mean very little in actual games. Sure in games a massive change in CPU clock speed makes a lot of difference, but a 400MHz overclock when you are already at 4.6GHz probably will need you nothing but 1-3FPS on average in games if you are lucky. Yet in 3D Mark, you could gain well over a thousand points by increasing the CPU clock speed.

It isn't as if 3D Mark scores of 11,000 points translate to being able to run Crysis at 640FPS or something. The number is literally meaningless. I have no idea why anyone considers 3D Mark a relevant application. It's not even useful for stability and stress testing. There are plenty of better ways of stressing a machine than that.

And as far as AMD goes, Bulldozer sucks. I hate to say that but there it is. I am not saying it doesn't have it's uses or potential in the server market, but in the desktop arena they run too hot and use too much power for what little performance they offer. It's pretty sad when a lot of the lineup gets trounced by the Core i3 2100 in games. There best CPU only keeps pace with the 2500K some of them time and when overclocked, the latter can absolutely leave it in the dust. The 2600K, 2700K, 3770K and so on are simply unmatched. Hell the 8 core Bulldozer doesn't touch my Core i7 980X on it's dated X58 platform. Speaking of which, AMD's fallen way behind on that as well. No PCI-Express 3.0 support. Considerably less memory bandwidth. No native USB 3.0 support. About the only things their platform has going for it is 6x6 SATA III 6Gb/s ports vs. Intel's 2 ports. Though companies like ASUS and even others are using the ASMedia 1061, which provides good performance and in some instances even beats out the Intel C600 series controllers. The SB950 also has a good amount of PCI-Express lanes, but Intel's X79 platform does just as well and the Z77 chipset gives us enough flexibility to cover for that deficiency.

The platform has further limitations with regard to poor thermal solutions. The mounting pattern for heat sinks is still basically the same crap we've had since the socket 939 days in terms of clearances and incompatibility with Intel's mounting hole patterns. This means that any heatsink designed for both will have to use a complex adapter to work on the AMD side, or be stuck to being mounted a certain way in order to fit the hole pattern AMD must use. Many heat sinks follow the same basic design as the Thermalright Ultra 120 Extreme and suffer the same drawback it has on the AMD platform which is not being able to be mounted with the fans facing the memory and the back of the case, but rather top or bottom of the case. Now heat sink makers could easily find ways around that and I believe there are some choices which allow for better mounting but the problem is that any heat sink made to combat AMD's crappy mounting hole pattern will probably end up virtually incompatible on the Intel side. Any heat sink maker would do well from a business standpoint to cater to the platform which sells the most units and that's Intel. This puts AMD at yet another disadvantage.

They had an opportunity to redesign their platform twice now. Once for Bulldozer and once for the FM1 socket. Instead they basically left all the limitations of their physical designs in place. Their desire to maintain compatibility with older sockets has bitten them in the ass too. They do not have the flexibility to adopt triple or quad channel memory to increase memory bandwidth, and they are also limited by getting stuck with VRD / VRM compatibility with older sockets and boards which boxes them in with the power design. As a gaming platform, AMD only started licensing SLI recently. I think the late adoption of that bit them in the ass. When NVIDIA has the better performing hardware, AMD had a real chance to continue to sell CPUs and boards even if people hated their GPUs. But instead they practically shoved people into Intel's camp in order to gain SLI compatibility.

With AMD it's been one bad decision after another. Sure it makes AMD look like the good guy when they take the extra time to ensure backwards compatibility with older platforms but it also kills them as it forces them to work within the confines of existing platform architectures rather than think outside the box and really innovate. It also bites people in the ass when they drop a shiny new processor into an older system and they don't end up getting the performance they think they should get, or some features are only partially enabled, or unsupported completely. AM2 users should recall the nightmarish headaches that came with the territory of CPU upgrades. With regard to Bulldozer you get older platforms that don't support the CPUs additional C-state and don't take full advantage of their power savings capabilitiies, as laughable as those are.

Older platforms also won't overclock as well. In speaking with people at various motherboard manufacturers they explained the additional challenges of adding the extra CPU microcode to the BIOS / UEFI and the types of problems it causes with validation and the trade offs that must be made in order for boards to be compatible with very different CPUs.

AMD is getting left further and further behind and they seem to be doing very little to change that.

 

[ImperfectLink]

If the app isn't making good use of many threads, then the intel will usually come out on top. AMD didn't make a crappy processor, intel just made a really good one.

SandyBridge is exceptional on all points especially timing. High IPC, turbo clocking for lightly threaded apps and simply put, the world is prepared to take advantage of hyperthreading.

Had AMD come out with BD a few years prior, by now it would be mature and competitive. Don't feel bad, if you're in the market for a new system, buy what's good at the time and what suits your price/performance ratio.

 

[drescherjm]

If the app isn't making good use of many threads, then the intel will usually come out on top.

The big problem is that in a lot of cases when an application can use more than 4 threads the i5 2500K or i7 2600K still comes on top or very close to the top.

Had AMD come out with BD a few years prior, by now it would be mature and competitive.

The big problem is AMD needed to have a working 32nm process in Q1 or Q2 2010. Not Q3 2011. An 8 core FX would have been a disaster on the 45nm process. Although it could have worked if they had 4 to 6 cores and did not design it to run at 4+GHz.

 

[Dan_D]

If the app isn't making good use of many threads, then the intel will usually come out on top. AMD didn't make a crappy processor, intel just made a really good one.

SandyBridge is exceptional on all points especially timing. High IPC, turbo clocking for lightly threaded apps and simply put, the world is prepared to take advantage of hyperthreading.

Had AMD come out with BD a few years prior, by now it would be mature and competitive. Don't feel bad, if you're in the market for a new system, buy what's good at the time and what suits your price/performance ratio.

Even multithreaded applications which show significant improvements on CPUs with more than 4 cores, and show gains with Hyperthreading enabled still run faster on Intel than AMD.

If you want to call the Bulldozer good, I'd disagree, but I think it's still probably more accurate to say that it is entirely wrong for the desktop market. This market requires high clocks and strong IPC performance. These are neither of Bulldozer's strong suits. It may pay off in the long run. I'm not going to say that it couldn't. Some aspects of the Pentium 4 were ill timed and paid off for Intel in the long run. So it certainly could happen. But Bulldozer at present is a bad chip. Whether by inferior design or just bad timing it isn't what AMD needs to succeed today. I haven't looked at server numbers in awhile but with low enough clock speeds the thermals and power envelope may get under control but poor IPC performance can hurt in any market. So Bulldozer is certainly a better move in the server market space where parallelism is key, but also thermal considerations and power consumption are huge in that market. The desktop market is far more tolerant of high thermals and high power consumption because a single machine, even one that uses a lot of power won't impact the end user's electric bill too much.

In datacenters where you may have dozens to thousands of machines, 20 watts per CPU is a HUGE deal and impacts operating costs exponentially as the amount of machines increases.

 

[ImperfectLink]

That's true. They bet on the wrong horse. The most they can hope for is similar performance (to quad core) in higher threaded apps. It's a combination of having gone the wrong direction and the competition going in the right direction for this time.

Edit: I agree with your point Dan. Mine was just that AMD has it all stacked against them at this time and for this design. The CELL processor may have been a good chip if it was utilized in the way the designers thought it would be.

 

[pelo]

Don't dismiss the reviews and the negative perception of the BD architecture as Intel fanboys and AMD haters. For gaming there's no question that Intel made a far better chip. If you're planning on sticking dual GTX680s in your rig then you really have no choice but to go Intel as well. That's really how far behind AMD is.

In some games AMD might perform a few FPS lower than Intel with SLi 580s/680s, but when it does perform worse (~50% of the time?) it performs significantly worse. We're talking 20-30FPS and more.

Unless you're okay with waiting until late Q3 early Q4 for Piledriver then buy the Intel chip and platform now. Piledriver still won't match the 2600K/2500K but at least it will get closer.

Dan, it fares better on the server end than it does it does on desktop but it isn't all good. The problem with its server performance is that it still lags behind Intel chips and is really only an upgrade to the Valencia chips if you have highly parallel tasks where optimizations and compilers take advantage of BD's uArch. Perf-per-watt there too is highly workload dependent and as a result Interlagos doesn't bode well against Intel alternatives. Despite Intel costing more, those costs are outweighted by 24/7 runtime power consumption delta between the two that favors Intel again. So in short it's only an upgrade to current platforms due to socket compatibility but suffers from the same ailments it does on the desktop. So, much like the desktop, unless you've got a very specific need for a cheap high core count chip it's a pass.

 

[J Macker]

Honestly, who cares about that score. Play some games, run some applications, don't just use an arbitrary score like 3DMark. Get a feel for which is snappier, which is smoother, which one you actually like using.

QFT. 3DMark has no merit.

 

[Dan_D]

That's true. They bet on the wrong horse. The most they can hope for is similar performance (to quad core) in higher threaded apps. It's a combination of having gone the wrong direction and the competition going in the right direction for this time.

When Intel went the wrong direction it took them time to change. Their new tick tock strategy allows them more flexibility but being able to maintain larger and more numerous design teams is ultimately their strong suit. You can't beat the R&D budget Intel has to work with. AMD can design an architecture, but if it sucks in the here and now it's far more difficult for them to recover or change direction. Also long term pay offs in evolutionary design are far more risky.

Even when Intel's Netburst architecture failed to pay off in the sense which it was intended (5.0GHz+ CPUs) Intel was still selling them in droves and making money hand over fist. AMD having a CPU which falls below expectations which isn't selling is a much larger concern.

 

[Dan_D]

Don't dismiss the reviews and the negative perception of the BD architecture as Intel fanboys and AMD haters. For gaming there's no question that Intel made a far better chip. If you're planning on sticking dual GTX680s in your rig then you really have no choice but to go Intel as well. That's really how far behind AMD is.

In some games AMD might perform a few FPS lower than Intel with SLi 580s/680s, but when it does perform worse (~50% of the time?) it performs significantly worse. We're talking 20-30FPS and more.

Unless you're okay with waiting until late Q3 early Q4 for Piledriver then buy the Intel chip and platform now. Piledriver still won't match the 2600K/2500K but at least it will get closer.

Well Intel's lead comes about from stronger IPC, and typically higher overclocking headroom. When the overclocks are so easily achieved and at least match AMD's and usually exceed them, the CPU with the IPC disadvantage always looses. At 1080P I'd say that much of the time the difference wouldn't be noticed in a blind test. But at 2560x1400 and beyond, especially in multimonitor configurations and where SLI, 3-Way SLI, Quad-SLI, 4-Way SLI come into play, the landscape changes and CPU performance becomes important again.

Vega did some benchmarks showing the strengths of PCI-Express Gen3.0 support in such configurations. I have to take his word for it as I've not done such testing myself, so I can only refer to the limited data available to me. But if that's true then AMD's platform becomes even more disadvantaged for gaming systems.

Dan, it fares better on the server end than it does it does on desktop but it isn't all good. The problem with its server performance is that it still lags behind Intel chips and is really only an upgrade to the Valencia chips if you have highly parallel tasks where optimizations and compilers take advantage of BD's uArch. Perf-per-watt there too is highly workload dependent and as a result Interlagos doesn't bode well against Intel alternatives. Despite Intel costing more, those costs are outweighted by 24/7 runtime power consumption delta between the two that favors Intel again. So in short it's only an upgrade to current platforms due to socket compatibility but suffers from the same ailments it does on the desktop. So, much like the desktop, unless you've got a very specific need for a cheap high core count chip it's a pass.

This is what I suspected. Intel has slowly been removing all the advantages AMD had over their Xeon line for some time now. Integrating the memory controller, ditching the FSB, higher IPC, better thermals, power consumption etc. were all steps in this process. Intel's CPUs are good at high clocks. A Xeon with lower clocks still has strong IPC and with the platform reaching parity if not suprassing AMD's in some ways, Their power consumption is also much better in the desktop market, and with Intel's ability to make lower voltage parts at high yields comparatively, I figured this is how things would end up. I know some people still swear AMD has advantages in CPU scaling, and that may or may not be true, but I have to think that when your IPC is significantly worse, the scaling advantage becomes less and less of an advantage.

 

[Pieter3dnow]

And as far as AMD goes, Bulldozer sucks. I hate to say that but there it is. I am not saying it doesn't have it's uses or potential in the server market, but in the desktop arena they run too hot and use too much power for what little performance they offer. It's pretty sad when a lot of the lineup gets trounced by the Core i3 2100 in games. There best CPU only keeps pace with the 2500K some of them time and when overclocked, the latter can absolutely leave it in the dust. The 2600K, 2700K, 3770K and so on are simply unmatched. Hell the 8 core Bulldozer doesn't touch my Core i7 980X on it's dated X58 platform.

In reality AMD had price going for them since Bulldozer that has faded away. I still frown upon some of your statements , gaming wise it is rather obvious that if the program doesn't do multi thread well it isn't going anywhere, but is that something you could ask AMD to fix all these shuvleware programmers ?

 

[Zarathustra[H]]

IMHO, 3DMark is completely irrelevant as a benchmark. In fact, benchmarks are completely irrelevant in general, other than for showoff and bragging rights.

As has been mentioned before, a system is only insufficient if it can't play the games you want to play at the settings you want to play it.

There are a handful of games for which no AMD CPU is sufficient regardless of overclocking, but they are few, and if you don't play them then you have no reason to abandon what you are using.

I happen to play one of them a lot (Red Orchestra 2) which is why I made the switch from AMD to Intel. Another one is Starcraft 2, which appears to be unusually CPU dependent.

Most games - however - are simply not that CPU dependent, and you are more likely to hit GPU dependence WAY before you have CPU issues.

 

[pelo]

Even when Intel's Netburst architecture failed to pay off in the sense which it was intended (5.0GHz+ CPUs) Intel was still selling them in droves and making money hand over fist. AMD having a CPU which falls below expectations which isn't selling is a much larger concern.

They were making money hand over fist because of some illegal payoffs What's worse is that 5ghz number is actually only half of what Intel had hoped for with their clock speed goals. Yes, Intel was actually aiming for 10ghz clock speeds with their Netburst architecture.

AMD is also adopting the tick-tock cadence but because they're relying on others for fabs this means that both their tick and tock are revolved around architectural tweaks whereas Intel's tick-tock is more fab process maturity > architecture. Considering Haswell and Ivy, though, it seems Intel is applying both annual changes and devoting them to graphical performance, at least on the desktop.

AMD knows they messed up, but you're right in that it affects them far more than it would and had affected Intel with Netburst so it's difficult for them to recover. Apparently Steamroller is abandoning the basic tenets of Bulldozer and should be a complete turnaround so they had something in the works (architectures take several years from design > implementation). I still don't expect it to do wonders and maybe we'll get something like the Denebs/Thubans where they performed very well for price-to-performance.

Now that's IF they decide to keep going with this x86 desktop thing at all. I wouldn't be surprised to see AMD abandon desktops in pursuit of tablets/laptops and server/HPC chips. Mobile and HPC is where the money is and they're both growing at a higher rate than desktops are. If you look at AMD's powerpoint presentations you'll notice the lack of a Steamroller desktop chip yet they mentioned a H1 2013 Kaveri Steamroller derived/GCN powered APU. But what about the desktop? Well, that's still undecided. If we do get a desktop chip it's likely to be another server oriented chip. Past 2013 they really haven't mentioned anything at all, and whatever they have planned isn't concrete. I think what happens to the company on the desktop will rely almost entirely on how HSA works out. If they can leverage HSA in such a manner that CPU performance benefits and they get the boost they're looking for then they'll keep at it. If the HSA gamble doesn't pay off then they'll fragment or go ARM.

 

[Dan_D]

In reality AMD had price going for them since Bulldozer that has faded away. I still frown upon some of your statements , gaming wise it is rather obvious that if the program doesn't do multi thread well it isn't going anywhere, but is that something you could ask AMD to fix all these shuvleware programmers ?

My point is that AMD doesn't fare much if any better in games and applications where multithreading is executed well. That refutes the entire argument that it's the application's fault and not AMD's with regard to poor performance. The gap closes in situations where the application or game handles multithreading better. I never said it didn't, but the AMD CPUs still fall short of Intel's even under the best of circumstances.

 

[MacLeod]

As a long time AMD fanboy, I have to admit its getting harder and harder to resist the call of Intel. My rig has been acting a little screwy since I reinstalled my OS after a mobo swap a few months back and Im thinking I should just do a clean install. Then I start thinking, might as well just go 3570K/Maximus V.......and maybe a nice SSD while I'm at it.

Even tho I wont notice much if any difference at my resolution and single card configuration......but then I think why not just buy a GTX670 while I'm at it. DAMMIT I've got to stop reading threads like this or else I'm going to go broke.....or get divorced......or both.

So sorry Dan and Pelo but for the sake of my marriage, I'm blocking your posts from here on out.

 

[Zarathustra[H]]

but is that something you could ask AMD to fix all these shuvleware programmers ?

Really doesn't matter whose fault it is. At the end of the line, when you have spent your money did you get something that works for you, or did you not.

If it doesn't, then the cause is really rather irrelevant.

That being said, multithreaded coding is not straight forward, and even when done right, has a lot of issues (thread locking, infinite wait states, etc.), which in some cases can make it slower than single threaded code, and in all cases, limits its scaling (2 threads shows more added benefit over one thread than 4 threads shows over 2, and by the time you hit 8 threads you are already showing SIGNIFICANT diminishing returns.

gaming is not a good application of multi threaded code (where encoding and rendering is), and since only a small portion of the code winds up being truly multithreaded, you see small benefits from core scaling, per Amdahls law.

 

[Pieter3dnow]

My point is that AMD doesn't fare much if any better in games and applications where multithreading is executed well. That refutes the entire argument that it's the application's fault and not AMD's with regard to poor performance. The gap closes in situations where the application or game handles multithreading better. I never said it didn't, but the AMD CPUs still fall short of Intel's even under the best of circumstances.

Really I have seen some phoronix linux benchmarks and some streetfighter IV ones which contradict that ?

** edit
And you know what Dan , I don't like AMD attitude with the pepsi taste test fiasco they had. That upsets me more then then anything else ...

 

[Zarathustra[H]]

Really I have seen some phoronix linux benchmarks and some streetfighter IV ones which contradict that ?

I'd be interested in seeing the Phoronix benchmarks you are referring to.

Streetfighter IV? Really?

What, are we talking a difference of 400fps and 450fps? Completely and utteruly irrelevant in any way shape or form.

As long as you can hit at least 60fps consistently in a game, any improvement over that no longer matters. 800fps is no better than 65fps.

 

[ssnyder28]

I had a 965 BE like you do @ 4.0 ghz, I just upgraded to a 2500k and its a substantial increase in gaming. Especially in the "normal" cpu intensive games. You will see increases across the board before even OCing it. Pick up sandy bridge cheap!

 

[Dan_D]

They were making money hand over fist because of some illegal payoffs What's worse is that 5ghz number is actually only half of what Intel had hoped for with their clock speed goals. Yes, Intel was actually aiming for 10ghz clock speeds with their Netburst architecture.

And I really didn't want to bring that up as it isn't relevant to the discussion. How Intel got the sales isn't important. I won't debate the morality of what they did or did not do. My point was that taking gambles and losing with regard to performance isn't as big a deal when you are moving product than when you aren't.

AMD is also adopting the tick-tock cadence but because they're relying on others for fabs this means that both their tick and tock are revolved around architectural tweaks whereas Intel's tick-tock is more fab process maturity > architecture. Considering Haswell and Ivy, though, it seems Intel is applying both annual changes and devoting them to graphical performance, at least on the desktop.

AMD didn't do better when they had their own fabs. Spinning off their fabs was nothing more than a stunt to make AMD look more profitable from an operational standpoint as the books won't show all the money being sunk into fabrication plant updates, and costs. The problem with the fabs themselves and process maturity is that everyone lags behind Intel. Anyone involved in the semi-conductor field or manufacturing will tell you this. So yeah, AMD must relay on architecture to some degree, but it isn't as if AMD's fab partners don't make strides as Intel does. Even once in awhile they'll leap frog Intel for a short time in one specific area. Though these incremental leads never seem to last.

AMD knows them messed up, but you're right in that it affects them far more than it would and had affected Intel with Netburst so it's difficult for them to recover. Apparently Steamroller is abandoning the basic tenets of Bulldozer and should be a complete turnaround so they had something in the works (architectures take several years of planning from design > implementation). I still don't expect it to do wonders and maybe we'll get something like the Denebs/Thubans where they performed very well for price-to-performance.

This is the best case scenario AMD can hope for in the near future I think. I don't expect to see AMD really compete with Intel at the same level they did with the Athlon 64 in the next five years.

Now that's IF they decide to keep going with this x86 desktop thing at all. I wouldn't be surprised to see AMD abandon desktops in pursuit of tablets/laptops and server/HPC chips. Mobile and HPC is where the money is and they're both growing at a higher rate than desktops are.

Yeah, I have a feeling that everything is going this way. Even Intel will eventually devote more resources to such markets. Most likely with them developing desktop CPUs with these other applications in mind and then scaling them down in order to fill those roles and market segments. I still think that desktop and server markets as we know them now will still be the pinnacle of performance and the first place we'll see technologies, but these will quickly be adapted and scaled down as mobile market demands increase.

 

[Dan_D]

Really I have seen some phoronix linux benchmarks and some streetfighter IV ones which contradict that ?

** edit
And you know what Dan , I don't like AMD attitude with the pepsi taste test fiasco they had. That upsets me more then then anything else ...

Capcom's engines do seem to benefit from multithreading to a heavy degree. Streetfighter IV, Resident Evil, Lost Planet, and Lost Planet 2 all prove that. You know what else they do? They become GPU bound VERY quickly. And remain so as their resolution increases. At 7680x1600 as an example, the high end Intel based 4-Way SLI rig will always crush the AMD rig. I've done this type of testing more than once. You need the full context of the situation. The benchmarks you saw obviously didn't provide that.

Benchmarking at 1024x768 (or less) and using extremely low settings is relevant how?

 

[Pieter3dnow]

I'd be interested in seeing the Phoronix benchmarks you are referring to.

Streetfighter IV? Really?

What, are we talking a difference of 400fps and 450fps? Completely and utteruly irrelevant in any way shape or form.

As long as you can hit at least 60fps consistently in a game, any improvement over that no longer matters. 800fps is no better than 65fps.

Well the thing is that the game isn't optimized for any set amount of threads it just scales .
there was a pcgamer interview posted here a good while ago with the programmer which explained how they approached the game.

http://www.ocaholic.ch/xoops/html/modules/smartsection/item.php?itemid=614&page=13

SLI is another Issue which i should say i care little about that it doesn't work for AMD i know that but were talking about multithread code.

 

[ImperfectLink]

All form factors will be developed for and offered for sale. It's just a question of what most people want in their household.

My perspective is from a CG artist that does a lot more rendering than gaming. My only beef with Bulldozer when it came out considering the lower price was they should have let me put two in a system instead of just the one. For all my basic modeling and texturing work (single threaded) there's been enough performance for that years ago. The rendering portion needs more threaded performance and it needs to be cost effective. $2k Xeons ain't gonna cut it for a freelancer.

 

[Dan_D]

All form factors will be developed for and offered for sale. It's just a question of what most people want in their household.

My perspective is from a CG artist that does a lot more rendering than gaming. My only beef with Bulldozer when it came out considering the lower price was they should have let me put two in a system instead of just the one. For all my basic modeling and texturing work (single threaded) there's been enough performance for that years ago. The rendering portion needs more threaded performance and it needs to be cost effective. $2k Xeons ain't gonna cut it for a freelancer.

No, but for rendering, two lower clocked Xeons with 6 or 8 physical cores + Hyperthreading will out perform any dual CPU option you'll get on the AMD side.

 

[pelo]

Yeah, I have a feeling that everything is going this way. Even Intel will eventually devote more resources to such markets. Most likely with them developing desktop CPUs with these other applications in mind and then scaling them down in order to fill those roles and market segments. I still think that desktop and server markets as we know them now will still be the pinnacle of performance and the first place we'll see technologies, but these will quickly be adapted and scaled down as mobile market demands increase.

While you're right in that the desktop role is diminishing, if you look at how Intel designs chips you'll see they do mobile > desktop and not the other way around. This has become more clear with Sandy Bridge and Ivy Bridge where graphics have become a major design goal and have trumped CPU performance. Why bother with graphics on an unlocked enthusiast processor? Well, because it's not meant for you, silly If Intel was concerned about giving us the best possible chips for desktops/enthusiasts then the 2011 platform would be the 1155 one and not relegated to workstation status. But because it performs very well and is a smaller chip with less cache that costs less to produce, Intel has marketed them as desktop chips instead of their server-class high TDP workstation monsters. AMD is doing the same thing with their APUs. Build for laptop and then grow it into a desktop chip. Bulldozer was a server chip that made it to desktop (granted it sucked ass but that's not the point). It's also the reason for the first time in a very long time we're seeing decreases in power consumption and essentially a stalling of CPU performance improvements in favor of graphics + perf-per-watt.

Desktops will remain but their role has already undergone changes and the definition of the desktop too will have to change. Desktops will resemble workstations//gaming stations rather than your typical sit down and check your email, facebook,etc. Basically, Intel will transition their 2011 platform to mainstream desktop and their 1155 mainstream desktop will be strictly mobile. With that prices will go up for desktop users and there will be even fewer of us. Err, fewer of you. When I say "us" I shouldn't include myself as I'm already looking to switch to a laptop to replace my desktop entirely. The majority of former desktop owners have already begun that transition or are heading that way and that's the reason why chip design has changed so drastically over the past couple of years.

 

[Dan_D]

Fair enough.