Picking A Sub-$200 Gaming CPU: FX, An APU, Or A Pentium?

[hanskey]

Intel Pentium G620 plus Radeon 6670. It gets about 50% higher frame rates than the A8s do (according to Tom's). There are viable Intel entry lever gaming solutions. The G620 plus 6670 is also usually slightly cheaper than the A8s and uses far less power, saving even more money over time. It's only sacrifice is in moderately/highly threaded performance, however, at this low of a gaming level, that does not effect gaming performance. It only reduces the G620's productivity performance.

Intel's Larrabee is extremely different from AMD's and Nvidia's technology. It would not be easy for them to move over to working for Intel. Intel didn't need them anyway when it made Larrabee and then Knight's Corner and it's family of products. It's not speculation that Intel can improve just like any other manufacturer. Intel has proven that they can with Knight's Corner and it's family of products, as well as with their CPU divisions and more.

Otherwise, yes, we seem to agree on the rest of this right now.

So, first, I want to apologize for any nerd-snark I've committed, the Internet makes a$$%#$s of us all.

However, I still feel that Intel is champ mainly because of memory performance and as evidence for this, I'm posting a link for you. I know this is a synthetic benchmark, and therefore insufficient for telling the full performance picture, but synthetic tests do allow us to isolate the performance of a CPU under very specific operations and give us a view of various subsystems' performance that we can't easily extract from the framerate of X game. When you look at the various synthetic tests, the memory tests are pretty much the only subsystem tests where Intel is always on top, although the highest-end Intel parts win many other subsystem tests. Plus, these test are backed up when you also look at the number of clocks for cache access and we know Intel uses fewer cycles at every level of cache than AMD, although I think this is somewhat mitigated by AMD's higher average clock rate (Sorry I don't have a link on that, but google is your friend).

http://www.tomshardware.com/charts/desktop-cpu-charts-2010/Memory-Bandwitdh-SiSoftware-Sandra-2010-Pro-Mem,2413.html
These results are consistent between Tom's and other sites' various synthetic benchmarks. The first thing you might note is that none of the processors actually deliver as much bandwidth as they claim in their specifications and AMD delivers far less than it claims and far less than Intel on a consistent basis. To me, this kind of subsystem isolation testing would indicate that Intel's chip are faster largely thanks to memory throughput, and when you examine other applications, those that have the largest memory footprint, usually give Intel the biggest advantage over AMD, like the WinRAR, and 7zip tests. I don't know about the games tested, but I would be surprised if those that favor Intel don't have a much larger memory footprint, or at least much more memory operations, which is usually a function of the running applications memory footprint, but also memory access patterns and frequency. After all, the less memory an application accesses, the faster it performs, because it stalls the CPU to retrieve data from RAM and the caches less often.

Maybe I'm reading test results wrong, but as I read the tests here and elsewhere, Intel appears to dominate mainly through much better memory subsystems.

 

[hanskey]

Intel Pentium G620 plus Radeon 6670. It gets about 50% higher frame rates than the A8s do (according to Tom's). There are viable Intel entry lever gaming solutions. The G620 plus 6670 is also usually slightly cheaper than the A8s and uses far less power, saving even more money over time. It's only sacrifice is in moderately/highly threaded performance, however, at this low of a gaming level, that does not effect gaming performance. It only reduces the G620's productivity performance.

Intel's Larrabee is extremely different from AMD's and Nvidia's technology. It would not be easy for them to move over to working for Intel. Intel didn't need them anyway when it made Larrabee and then Knight's Corner and it's family of products. It's not speculation that Intel can improve just like any other manufacturer. Intel has proven that they can with Knight's Corner and it's family of products, as well as with their CPU divisions and more.

Otherwise, yes, we seem to agree on the rest of this right now.

Yeah, Larrabee was a totally different approach to graphics, but it was proven to be an inferior one as well. Vector processors (SIMD), like modern GPUS, massively outperform the Larrabee architecture (MIC), and that's why Intel isn't still pursuing it for competing with AMD and Nvidia in the graphics space. The main reason people want to abandon the current GPU SIMD architecture for the Larrabee model is that vector processing is incredibly difficult to write in the currently available languages, not because MIC has any performance benefits over SIMD processors. Larrabee was a complete failure because it's performance was pathetic, not for any other reason, and that is primarily thanks to a weak architecture, when compared to SIMD. Intel managed to sort of salvage the architecture for the HPC market, but KC is still not up to the task, by most people's analysis. I'm getting that from a now-retired IBM developer, who writes an excellent blog about parallel processing, which I recommend you read in it's entirety at some point. Here's a fairly recent article on MIC and KC.

http://perilsofparallel.blogspot.com/2011/10/mic-and-knights.html

Note that I'm not arguing that Intel is incapable of improvement, just that Intel will almost certainly never find it practical to compete with Nvidia and AMD in graphics, and that the Larrabee architecture is fundamentally less good at graphics than a SIMD architecture. In fact Larrabee, in it's KC form, get's a fair performance boost compared to it's initial form from some new SIMD-type instructions, and it still can't compete with SIMD solutions, even though SIMD is so hard to program.

I don't think it is practical, because if they switched to a SIMD architecture so they could actually compete in graphics, then they'd also have to write a new SIMD micro-architecture and make sure that code can be compiled to that architecture, and then would end up having to write numerous APIs and SDKs, just like NVidia and AMD. That's a whole lot of expense, with no guarantee that people would choose an Intel GPU over AMD and NVidia, and is the opposite direction Intel wants to go in. Intel wants to entice people to adopt KC (Larrabee) with "easier programming, but similar performance" and basically kill the add-on graphic card industry in favor of on-die Larrabee graphics, but there's not a lot to suggest that MIC will ever be capable of that. Plus, AMD already beat them to the punch by delivering cheap good built-in graphics on the chip with Llano, so the low cost market that Intel targeted for Larrabee/MIC is already being taken over by AMD, leaving Intel nothing but the HPC market and prayers for MIC.

 

[hanskey]

I also just read that AMD plans on using one of the Trinity processors in Ultra-thin notebooks, delivering the same performance as the current A8-3850 in a 17 Watt part. This mean that for the first time, really good and useful ultra-thins will be available at less than $1000. In fact, AMD has reportedly told manufacturers their AMD ultra-thin notebooks should go for between $400 - $900.

To me, this could easily be as big of a revolution in computing as the iPad. Who needs a tablet when you can get an ultra-portable Windows 7 box (64-bit) for $500, that is super light, 18mm thick, that you can game on, with better battery life than the iPad3? Personally, I feel that with such ultra-portable laptops, AMD has a good chance of stealing a lot of tablet market from Apple, etc., because they'll be selling for about the same price with far better hardware and software than a tablet can manage, and Windows is still the application king. Also, having a built-inkeyboard and a screen you don't have to strain your arms or neck to view straight on, is a big plus for AMD ultra-thin notebooks over tablets.

I also read that AMD doesn't plan on getting back into the high-end of the market until 2014 with Steamroller, so that's a little sad, but it's a good strategic plan and a better use of AMD's R&D money than trying to beat the i7, IMO.

 

[blazorthon]

With all this talk of memory controllers. from tech-powerup
http://www.techpowerup.com/tags.php?tag=Llano

AMD's upcoming "Trinity" family of desktop and mobile accelerated processing units (APUs) will use up to four x86-64 cores based on the company's newest CPU architecture, codenamed "Piledriver". AMD conservatively estimated performance/clock improvements over current-generation "Bulldozer" architecture, with Piledriver. Citavia put next-generation A10-5800K, and A8-4500M "Trinity" desktop and notebook APUs, and pitted them against several currently-launched processors, from both AMD and Intel.

It found integer and floating-point performance increases clock-for-clock, against Bulldozer-based FX-8150. The benchmark is not multi-threaded, and hence gives us a fair idea of the per core performance. On a rather disturbing note, the performance-per-GHz figures of Piledriver are trailing far behind K12 architecture (Llano, A8-3850), let alone competitive architectures from Intel.

It's only disturbing if the Piledriver based CPUs aren't more energy efficient than Llano and Bulldozer based FXs at the same frequency. IE, if Bulldozer was as energy efficient at 5GHz as Sandy Bridge is at 3GHz despite the IPC difference, then they would still have similar performance and similar performance per watt and similar power usage.

Of course, Bulldozer isn't that energy efficient and is really very inefficient, but the logic still stands. Until we see power usage benchmarks for Trinity and Piledriver based CPUs, then we have too little information to make valid opinions and decisions related to Trinity and Piledriver based CPUs.

Considering the improved performance with Piledriver and knowing what went wrong with Bulldozer, Piledriver could be more power efficient and not just have more IPC than Bulldozer.

Also, those benchmarks are all synthetics, so we don't know how well they will translate into realistic performance.

 

[hanskey]

I can agree with you on that Blazerthon. The comment on fx trailing liano is what i found interesting.

It makes sense when you really sit down and think about it though, because Bulldozer has fewer FP resources than Athlon II, and Llano is Athlon II + HD6550 (from what I've read, but correct me if I'm wrong). Trinity is Bulldozer cores + 7000 series Radeon, instead of Athlon II + 6000 series Radeon. As second-gen Bulldozer design it's expected that it improves on current Bulldozer, but not super surprising that it still lags behind STARS architecture in the A8, since STARS is about as perfected as a CPU line gets.

 

[hanskey]

I am very weak on architecture knowledge but i understand it this way
fx has a 256 bit fpu that is split into two 128 bit fpu to handle work loads all requests are handled in front of the bus like a holding area waiting to execute. Now because of amd's weak branch prediction and out of order execution algorithm,, latency is added waiting to execute down the pipe, intel has the same 256bit fpu but is capable of 8 32's or 4 64 bit fpu's depending on the workload. Amd needs more memory bandwidth and more clock speed just to equal intel. This is not to be taken as fact and i am sure i screwed it up somewhere but liano has a lower latency on the fpu's than fx.

Oh man. Over on the Best Gaming CPUs for the money review thread, blazorthorn threw down some seriously good articles related to all this architecture conversation. I'm going to repost his links and ask that he posts more, because this is some great reading and totally right on, in terms of understanding WTF has happened to AMD:

I have some links to show some of these problems, if you're interested in them:

http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/2

http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/6

http://hardforum.com/showpost.php?p=1037482638&postcount=88

That last one is one of the best reads on how AMD's ex engineers disliked AMD's management's choice to abandon being competitive with Intel in performance and power efficiency for seemingly no good reason.

I don't think you are wrong, but there are so many more ways that AMD could improve than we could have imagined. Roughly: 50% performance benefit by having human hardware designers redo the processors, and %50 improvement with other obvious architectural improvements. Before I got all this additional info on what is causing AMD's performance dive, I was skeptical that Steamroller would be competitive with Intel in 2014 for the top of the charts, but if AMD only remedies the known issues, we could be looking at another real race for the top in very short order.

 

[blazorthon]

[citation][nom]hanskey[/nom]It makes sense when you really sit down and think about it though, because Bulldozer has fewer FP resources than Athlon II, and Llano is Athlon II + HD6550 (from what I've read, but correct me if I'm wrong). Trinity is Bulldozer cores + 7000 series Radeon, instead of Athlon II + 6000 series Radeon. As second-gen Bulldozer design it's expected that it improves on current Bulldozer, but not super surprising that it still lags behind STARS architecture in the A8, since STARS is about as perfected as a CPU line gets.[/citation]

Actually, Llano has modified Stars core plus a VLIW5 GPU (only the Llano A8s have the 6550D, the A6s have the 6530D and the A4s have weaker). Llano has Husky cores which are modified Stars cores and are slightly faster than Stars cores in the Athlon IIs at the same frequency (although they are only slightly faster). Trinity should be Piledriver (modified Bulldozer) plus a GCN GPU (Radeon 7000 has both GCN and older VLIW5 GPUs, VLIW5 is used in all of the low end, IE the 7600 and below families of cards).

Llano's 6550D/6530D is actually a modified Redwood GPU from Radeon 5000. It's called the 6550D, 6530D, etc because it has the added Blu-Ray support that differentiates the VLIW5 Radeon 6000 cards from the Radeon 5000 cards, all of which are VLIW5.

As for which GCN GPU the Trinity IGP will be based on, I have to guess that it will either be a it's own GPU, or a Cape Verde GPU (Radeon 7700 family).

[citation][nom]tourist[/nom]I am very weak on architecture knowledge but i understand it this way fx has a 256 bit fpu that is split into two 128 bit fpu to handle work loads all requests are handled in front of the bus like a holding area waiting to execute. Now because of amd's weak branch prediction and out of order execution algorithm,, latency is added waiting to execute down the pipe, intel has the same 256bit fpu but is capable of 8 32's or 4 64 bit fpu's depending on the workload. Amd needs more memory bandwidth and more clock speed just to equal intel. This is not to be taken as fact and i am sure i screwed it up somewhere but liano has a lower latency on the fpu's than fx.[/citation]

Each Bulldozer module shares dual 128 bit FPUs that can work together as a single 256 bit FPU. It is not the same as Intel's 256 bit FPU in Sandy Bridge, nor any other Intel FPU.

AMD doesn't need more memory bandwidth than Intel. What AMD needs is higher memory frequencies to get the same bandwidth as Intel. For example, the Bulldozer and Llano memory controller has about 25% less bandwidth than Intel's Sandy Bridge controller when they have the same number of channels occupied (max that any of them support is two) at the same frequency. AMD's memory controllers are too inefficient. That is why AMD needs 1866MHz or so to rival Intel's bandwidth at 1600MHz or so.

A quad core Bulldozer (dual module) has half of the 256 bit FP resources of it's predecessors and of the quad core Intel processors because instead of 256 bit per core, it's 256 bit (or dual 128 bit) per module. However, for work that involves AVX, SSE 4.x, and XOP instructions, Bulldozer performs far better because it's predecessors (Thlon II, Phenom II, Llano) don't support these instructions. This makes Bulldozer about four times faster per module than it's predecessors are per core, so the quad core FXs are still over twice as fast as it's predecessors for these types of floating point work.

Oh man. Over on the Best Gaming CPUs for the money review thread, blazorthorn threw down some seriously good articles related to all this architecture conversation. I'm going to repost his links and ask that he posts more, because this is some great reading and totally right on, in terms of understanding WTF has happened to AMD:



I don't think you are wrong, but there are so many more ways that AMD could improve than we could have imagined. Roughly: 50% performance benefit by having human hardware designers redo the processors, and %50 improvement with other obvious architectural improvements. Before I got all this additional info on what is causing AMD's performance dive, I was skeptical that Steamroller would be competitive with Intel in 2014 for the top of the charts, but if AMD only remedies the known issues, we could be looking at another real race for the top in very short order.

I hope that this was what you were looking for from me. Sorry about it being a week late. I'm still looking for other links, but haven't found much yet that wasn't cited in the links I've already found and provided.

 

[Guest]

it's a bit idiotic comparing apples to oranges as it's idiotic to compare data from dualcore to quadcore without saying how much cores do the software that is providing the data actually uses

 

[blazorthon]

[citation][nom]wx[/nom]it's a bit idiotic comparing apples to oranges as it's idiotic to compare data from dualcore to quadcore without saying how much cores do the software that is providing the data actually uses[/citation]

Would you mind reorganizing that so it actually makes sense to people reading it and then re-reading the article because it tells you how well-threaded each benchmark is.

 

[Nibhor]

According to this article, a 20% increase of PhenomII's NB frequency translate in a 8-9% fps gain in starcraft 2!

http://www.anandtech.com/show/3877/...investigation-of-thuban-performance-scaling/7

If true, that's an overclocked hierarchy chart's game changer.

 

[kzahrai]

Is there any validation of the same if not similar GPU are being used?

 

[spookyman]

Lets see...

Sandy Bridge line

I5 2500k at $159.99 at Microcenter

or

Ivy Bridge line

i5 3750k at $189 at Microcenter.

Oh AMD can take a flying leap now.

 

[blazorthon]

[citation][nom]spookyman[/nom]Lets see...Sandy Bridge lineI5 2500k at $159.99 at MicrocenterorIvy Bridge linei5 3750k at $189 at Microcenter.Oh AMD can take a flying leap now.[/citation]

Not everyone has Microcenter available. Not even nearly everyone within the USA. Furthermore, they have AMD CPUs at similar discounts too. AMD has even dropped prices again recently.

 

[Guest]

wait, so what graphics card was used in the testings?

 

[army_ant7]

[citation][nom]jihidiwhite[/nom]wait, so what graphics card was used in the testings?[/citation]

AMD Radeon HD 7970
925 MHz GPU, 3 GB GDDR5 at 1375 MHz

 

[army_ant7]

I haven't read this yet, but I'm excited to. Just in case someone stumbles upon this and hasn't seen this yet.
http://www.tomshardware.com/reviews/amd-llano-opencl,3284.html

 

[oxford373]

almost in all benchmarks A8-3870k outperformed athlone II X4 645 ,so why at conclusion page X4 645 is 3% faster than 3870K there should be something wrong
skyrim/battlefield/just cause2/starcraft2/dirt3 /metro 2033
A8-3870K 27.2/ 52.3 /52.7 /30.4 /72.9 /48.4
x4-645 26.2 / 53.1 / 51.4 /30.0 /68.4 /45.4

 

[cleeve]

[citation][nom]oxford373[/nom]almost in all benchmarks A8-3870k outperformed athlone II X4 645 ,so why at conclusion page X4 645 is 3% faster than 3870K there should be something wrong[/citation]\

There's nothing wrong, you're just not reading the chart properly.

The 3870K *is* faster at average FPS, but the 645 is faster at minimum FPS.

 

[oxford373]

http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120-10.html
A8-3870K average FPS 133.3% -min fps 153.5%
X4 645 average FPS 136.7% -min fps 142.7%

 

[cleeve]

[citation][nom]oxford373[/nom]almost in all benchmarks A8-3870k outperformed athlon II X4 645 ,so why at conclusion page X4 645 is 3% faster than 3870K there should be something wrong skyrim/battlefield/just cause2/starcraft2/dirt3 /metro 2033A8-3870K 27.2/ 52.3 /52.7 /30.4 /72.9 /48.9x4-645 26.2 / 53.1 / 51.4 /30.0 /68.4 /45.4[/citation]

You know what? I checked the excel spreadsheet and there were problems with the A8-3870K entries. You're right!

The weird part is, the 3870 still beats the 645 in the minimums on the aggregate chart, but the margin is much smaller now:

3870K: 145.8 min, 141.8 avg
645: 142.7 min, 136.7 avg

Anyway, good eye! I'm uploading the corrected chart now.

 

[Guest]

how much RAM was used in the testing?

 

[army_ant7]

Trinity seems to have some nice prices. When Vishera comes out (or when prices drop further) maybe a remake of this will be in order? 😀

 

[Guest]

i just checked and found out the a10 5800k at 3.8ghz only has a wei of 7.3, the i3 3220 scores only .1 less at 7.2 and its only a dual core processor, costs cheaper, and will definately score better in benchmarks, come on trinity, you put a quad core cpu at 3.8ghz with 4mb cache and cant even get past a 7.3 wei, the i5 3570 at 3.4 ghz scores a 7.6 wei lol.

 

[army_ant7]

i just checked and found out the a10 5800k at 3.8ghz only has a wei of 7.3, the i3 3220 scores only .1 less at 7.2 and its only a dual core processor, costs cheaper, and will definately score better in benchmarks, come on trinity, you put a quad core cpu at 3.8ghz with 4mb cache and cant even get past a 7.3 wei, the i5 3570 at 3.4 ghz scores a 7.6 wei lol.

I don't think the Windows Experience Index (WEI) is the most accurate way of determining hardware performance. For one thing, it isn't specific as to the kinds of workloads tested. Another thing to note is that even if you don't have the highest-end card, it could top out in score already (i.e. even some mid-range cards could I think, not sure).

Look for benchmarks on them (this article having examples of such) to know how good Trinity really is doing, where it excels and doesn't. 🙂

 

[cleeve]

I don't think the Windows Experience Index (WEI) is the most accurate way of determining hardware performance...

...Look for benchmarks on them (this article having examples of such) to know how good Trinity really is doing, where it excels and doesn't. 🙂

+1 for TRVTH