[TOM'S] AMD inter-generational CPU shootout

[BallaTheFeared]

Not to mention there are no Vishera products on that review like FX8350, FX6300 etc

 

[AtenRa]

My thoughts exactly, after benching my HD4600 in Dota 2 I know for a fact Ra was at best laning and none of his 3 minute strip is with a late game 5v5 team fight.

My run was with every IQ setting ON except for AA at 1080p. You can always turn off a few IQ settings if you believe that will help. The fact is that you can play DOTA 2 with HD4000. Perhaps some times performance will dip bellow 30fps but on average you can setup the game to lower IQ to compensate.

Same happens with Marvel Heroes, you can play the game at high IQ but when 5-6 heroes start to fight all together, fps drops bellow 30. You have to turn IQ settings to Low in order to compensate for that performance loss.

 

[BallaTheFeared]

That's the nice thing about GPUs, you can turn down settings and reduce their load a hundred fold, there is no setting that does the same for cpu requirements.

Apples > Oranges

 

[el etro]

No question about it, Intel is better in gaming but the FX 6350 is a better all around cpu than i3 3225, it even matches the intel chip in gaming. If you calculate the averages, both high and minimums across all games they're evenly matched. $140 and below there's real competition guys, make no mistake about it. For anything higher than $140, Intel just trashes AMD, no doubt.

What baffles me is the constant AMD bash feast even when they offer a compelling alternative such as the FX 6350, a cpu that is priced the same as the Intel, performs similarly in games and better in most of the applications out there.

Sure the 6300/6350 are better value than core i3 today(On everything).

So to sum up, FX6300 OCed to 4GHz is within 10-15% of the Core i5 3570K at HALF the COST. You can even buy a $30 heat-sink and OC to 4.6GHz and have almost the same performance at again a lower price.

That $100 price difference can get you an HD7950 over HD7790. No matter how fast the Core i5 is, any gamer will be way better with FX6300 + HD7950 than core i5 + HD7790.

Big deal here. Really.


------------------


anyways AMD needs to boost single-threaded performance(like they promised to us back in 2010), and of course improve their multicore scaling(intel is much better here since Core2)(AMD needs so much this in order to fight Intel with more cores) and memory controller(Will be awesome for improve future IGP performance) to make their CPUs more competitive in the market. Piledriver, by this time, is decent but only fixed the flaws of the original Bulldozer design.

 

[Slomo4shO]

That's the nice thing about GPUs, you can turn down settings and reduce their load a hundred fold, there is no setting that does the same for cpu requirements.

Apples > Oranges

So you are arguing that turning down the settings have no effect on minimum FPS?

Big deal here. Really.

It is alright, most individuals would rather argue about individual components rather than builds with set budgets.

 

[BallaTheFeared]

So you are arguing that turning down the settings have no effect on minimum FPS?

Not if those settings (most don't) have nothing to do with cpu load and the min fps is a direct result of the CPU.


Also a 6300 wouldn't touch an overclocked i5 in gaming, not even close. An overclocked 6300 still won't beat a stock i3 in many cpu limited titles.

 

[Slomo4shO]

An overclocked 6300 still won't beat a stock i3 in many cpu limited titles.

I fail to see the point of this "argument". The argument is true solely on single thread games and the inverse is true for multi-threaded games so what exactly is the purpose of declaring that "the i3 is better for X games" when the 6300 is better for Y games. Wouldn't the logical conclusion be that each has their own niche at a specific budget? Blanket statements, much like other assumptions, are typically wrong. I guess logic and fanboyism doesn't mix.

 

[BallaTheFeared]

Single threaded is very rare, dual threaded is not, especially in non console port games.


That is exactly my point, both the i3, and AMD have their own shortcomings.

Which is where I see the 6300 vs i3 debate, on one side you can see when a game can use more threads the 6300 makes a compelling choice, then all of a sudden the i3 guy posts the cpu limited games and the 6300 looks bad. I try to stay out of these discussions, because of the price point and the needs of others and yada yada blah blah, but my personal feeling is to save and get the best overall gaming processor for another few hours of work at min wage.

Here is my issue with both.

Say you budget $130 for the cpu, $200 for the gpu. When each is bottlenecking, limiting performance, you're losing $50, $70 worth of gpu performance. Because you skimped on a vital portion of a gaming platform you're now tossing money out the window on a graphics card for overall performance not performance when you actually need it. Now compound that, 3 years out we have at least one, possibly two generations, you go buy another $200 gpu... You're talking about a GPU that is 2 to 4 times faster than the $200 gpu you're buying today. Do either one of those processors make any sense anymore? Not to me.

 

[Slomo4shO]

Here is my issue with both.

Say you budget $130 for the cpu, $200 for the gpu.

So for a $330 CPU & GPU budget, lets go a step further and include the motherboard into the mix. Let us just assume $70 for the motherboard to bring the balance to $400 for the three components.

CPU: Intel Core i5-4430 3.0GHz Quad-Core Processor ($174.99 @ Amazon)
Motherboard: MSI B85M-G43 Micro ATX LGA1150 Motherboard ($74.24 @ Amazon)
Video Card: Sapphire Radeon HD 7850 2GB Video Card ($153.98 @ Newegg)
Total: $403.21

CPU: AMD FX-6300 3.5GHz 6-Core Processor ($117.98 @ Outlet PC)
Motherboard: Biostar TA970 ATX AM3+ Motherboard ($69.99 @ Newegg)
Video Card: Sapphire Radeon HD 7950 3GB Video Card ($199.99 @ Newegg)
Total: $387.96

CPU: Intel Core i3-3220 3.3GHz Dual-Core Processor ($118.99 @ Amazon)
Motherboard: Gigabyte GA-B75M-HD3 Micro ATX LGA1155 Motherboard ($64.99 @ Microcenter)
Video Card: Sapphire Radeon HD 7950 3GB Video Card ($199.99 @ Newegg)
Total: $383.97

Each of these 3 builds would have their benefits and weaknesses. One could argue that a particular bundle may be superior at a particular task but arguing that any one of these three are hands down superior to the other two would be a fallacy. Systems are built based on the need of the user. The argument for the general performance of a single component is typically moot in a system build that has a budget and goal in mind.

 

[frozentundra123456]

That's the nice thing about GPUs, you can turn down settings and reduce their load a hundred fold, there is no setting that does the same for cpu requirements.

Apples > Oranges

That is pretty much why I finally upgraded my E4500/9800GT setup. I was running at only 900p, and was willing to turn down some settings. But I was finding that I was becoming more and more often CPU limited and turning down the graphical settings did not help. Even replacing the power supply and adding a powerful gpu would not have helped much.

So I do agree that picking a system is not as simple as some say-- put as much of the budget as possible towards the gpu. As in most everything you do, one has to compromise and get a balanced system. In fact, personally, I tend to put more emphasis on the CPU. It is much easier to upgrade the graphics card down the road, than to replace the CPU and/or motherboard. Not to mention that gpu performance probably will continue increasing, while CPU performance has stagnated, meaning that if you get a good CPU now, you should be set for a very long time with just gpu upgrades.

 

[KingFatty]

What in the heck? Instead of getting worse and out of control, this discussion is actually turning out very interesting and helpful. Nice job everyone.

My decision is way different when I can just drop-in a new CPU to replace an older CPU on an existing system. So my choice for CPU is different when I'm upgrading just the CPU vs. when buying a new system.

 

[Phynaz]

Not to mention there are no Vishera products on that review like FX8350, FX6300 etc

Yeah, because it's really tough to add 10% onto the 8150 score

 

[alexruiz]

Also a 6300 wouldn't touch an overclocked i5 in gaming, not even close. An overclocked 6300 still won't beat a stock i3 in many cpu limited titles.

CPU limited titles? A lot of you guys need to understand "cpu limited" and learn how to clarify and into Poorly CPU threaded, and well CPU threaded titles. Furthermore, everyone claiming "cpu limited" should always clarify the type of game we are talking about, older and unoptimized for more threads, or newer and optimized for more threads. Tom's review itselfs proves my point, the Propus Athlon II X4 640 was slower in many newer titles that are considered "gpu limited". Why? Aren't those games "gpu limited"?

Same for those pushing for Pentiums G1220, and thinking it will play crysis 3. The game is so called "gpu limited", but then, if so, why isn't the G1220 scoring as the i5s in the game, if it is not "cpu limited"?

All games, one way or another, are CPU limited, the difference is how well they can use the CPU threads available. Some are more graphically intensive than others, highlighting the lack of GPU power more abruptly, hence why many people so call them "gpu limited". But again, if crysis 3 is not "CPU limited", why isn't the "mighty" G1220 scoring as the i5?

Let's break the broad "cpu limit" concept games into "poorly threaded" and "well threaded". So what you meant was that an i3 will be faster than a FX6300 in poorly threaded games?

 

[frozentundra123456]

CPU limited titles? A lot of you guys need to understand "cpu limited" and learn how to clarify and into Poorly CPU threaded, and well CPU threaded titles. Furthermore, everyone claiming "cpu limited" should always clarify the type of game we are talking about, older and unoptimized for more threads, or newer and optimized for more threads. Tom's review itselfs proves my point, the Propus Athlon II X4 640 was slower in many newer titles that are considered "gpu limited". Why? Aren't those games "gpu limited"?

Same for those pushing for Pentiums G1220, and thinking it will play crysis 3. The game is so called "gpu limited", but then, if so, why isn't the G1220 scoring as the i5s in the game, if it is not "cpu limited"?

All games, one way or another, are CPU limited, the difference is how well they can use the CPU threads available. Some are more graphically intensive than others, highlighting the lack of GPU power more abruptly, hence why many people so call them "gpu limited". But again, if crysis 3 is not "CPU limited", why isn't the "mighty" G1220 scoring as the i5?

Let's break the broad "cpu limit" concept games into "poorly threaded" and "well threaded". So what you meant was that an i3 will be faster than a FX6300 in poorly threaded games?

One must live with the games as they are. If the cpu is the limiting factor, it is by definition, cpu limited. Whether it is poorly threaded or not is irrelevant really. If what you are implying is that poorly threaded games penalize AMD hardware because it has more threads and lower single threaded performance, that may be true, but the performance as they say "is what it is". Calling it "poorly threaded" instead of "cpu limited" does not change anything. If you wish to blame the programmers for AMDs poor performance in certain games, you certainly can do so. One could just as easily blame AMD for designing cpus with low single threaded performance. In the end it doesnt matter, because blaming either or both does not change the performance.

 

[Abwx]

Yeah, because it's really tough to add 10% onto the 8150 score

10% is 74% of Vishera improvement over bulldozer at equal frequency
for games on average.


http://www.hardware.fr/articles/880-6/bulldozer-vs-piledriver-4-ghz.html

 

[Phynaz]

10% is 74% of Vishera improvement over bulldozer at equal frequency
for games tested by hardware.fr on average.


http://www.hardware.fr/articles/880-6/bulldozer-vs-piledriver-4-ghz.html

Ok, then add 13%. And I also fixed your post for you.

 

[Slomo4shO]

One must live with the games as they are. If the cpu is the limiting factor, it is by definition, cpu limited. Whether it is poorly threaded or not is irrelevant really. If what you are implying is that poorly threaded games penalize AMD hardware because it has more threads and lower single threaded performance, that may be true, but the performance as they say "is what it is". Calling it "poorly threaded" instead of "cpu limited" does not change anything. If you wish to blame the programmers for AMDs poor performance in certain games, you certainly can do so. One could just as easily blame AMD for designing cpus with low single threaded performance. In the end it doesnt matter, because blaming either or both does not change the performance.

Thread limited seems more appropriate. When only 1 or 2 cores/threads are being utilized by a program and you have the capacity for 4/8 cores/thread, that is not a hardware based bottleneck. This is a software induced bottleneck. Call it for what it is. The CPU is not the limiting factor, it is the poor coding that inhibits better performance.

Even a 5GHz 4670K is considered a "bottleneck" by your existing definition when a program only uses 1 or 2 thread and caps CPU usage on the 1 or 2 cores. However, the reality is that the CPU is not being utilized at 100% capacity. Artificial limiters caused by the software does not mean that the the hardware resources have been depleted and a specific piece of hardware has become a "bottleneck". On the contrary, there are plenty of CPU power available that is not being utilized. Hence, the bottleneck exists elsewhere.

 

[BallaTheFeared]

However you want to term it, doesn't change what occurs with it.

AMD fails to deliver, Intel doesn't.

You can't change software, you can however pick a CPU.

 

[erunion]

On the contrary, there are plenty of CPU power available that is not being utilized. Hence, the bottleneck exists elsewhere.

Are you suggesting the software is the bottleneck? The program can't be a bottleneck. Its an analogy used for hardware. You've gone pretty far off the path with that one.

 

[frozentundra123456]

Thread limited seems more appropriate. When only 1 or 2 cores/threads are being utilized by a program and you have the capacity for 4/8 cores/thread, that is not a hardware based bottleneck. This is a software induced bottleneck. Call it for what it is. The CPU is not the limiting factor, it is the poor coding that inhibits better performance.

Even a 5GHz 4670K is considered a "bottleneck" by your existing definition when a program only uses 1 or 2 thread and caps CPU usage on the 1 or 2 cores. However, the reality is that the CPU is not being utilized at 100% capacity. Artificial limiters caused by the software does not mean that the the hardware resources have been depleted and a specific piece of hardware has become a "bottleneck". On the contrary, there are plenty of CPU power available that is not being utilized. Hence, the bottleneck exists elsewhere.

I agree, a 5ghz 4670K could be a cpu bottleneck. If you want to call it a "software bottleneck", go ahead. However, how many threads does it take before you stop blaming the software.. 2,4, 8, 16, 50??? Point is, fast single core performance is still an advantage in many games, blaming the software does not change that.

The performance "is what it is" no matter what the reason.

 

[Slomo4shO]

Are you suggesting the software is the bottleneck? The program can't be a bottleneck. Its an analogy used for hardware. You've gone pretty far off the path with that one.

A bottleneck limits the throughput of associated resources, the term is not reserved for hardware. Even budget allocation can be considered a bottleneck in a system. Consider that you spend extra on a PSU with a higher capacity but you don't utilize the extra capacity. The extra $$$ spent on the PSU could have been used to improve performance by spending the funds elsewhere such as faster memory or better CPU.

The performance "is what it is" no matter what the reason.

This is true but there is also a built in fallacy with this conclusion. When we fail to acknowledge the root cause of the lack of performance, we continue to place the burdens of better performance on the wrong party. It is true that AMDs single thread performance is poor. It is also true that there are limitations to single core performance. It is further true that multi-core scaling can be improved drastically(more true for AMD). AMD and Intel are able to address single thread performance and multi-core scaling. The performance limitations of a singe core/thread are boundaries that may potentially be extended with new technology but the the current logical steps to circumvent this boundary is to incorporate multiple cores. At this point, any software developer that creates a product that caps out 1 or 2 threads is creating an artificial bottleneck and the blame, thus, should be directed at the poor coding and not the lack of resources. The other caveat is that the enthusiast market is not the primary target for game developers and, therefore ,they will not code for multiple threads until their target audience starts to feel the performance issues. On the other hand, observation of the professional software development show a greater utilization of multiple threads. Newer generation of productivity software has been increasingly coded for multiple threads as competition is present and the consumer will spend their $$$ on the product that can save the most time and $$$.

 

[frozentundra123456]

A bottleneck limits the throughput of associated resources, the term is not reserved for hardware. Even budget allocation can be considered a bottleneck in a system. Consider that you spend extra on a PSU with a higher capacity but you don't utilize the extra capacity. The extra $$$ spent on the PSU could have been used to improve performance by spending the funds elsewhere such as faster memory or better CPU.




This is true but there is also a built in fallacy with this conclusion. When we fail to acknowledge the root cause of the lack of performance, we continue to place the burdens of better performance on the wrong party. It is true that AMDs single thread performance is poor. It is also true that there are limitations to single core performance. It is further true that multi-core scaling can be improved drastically(more true for AMD). AMD and Intel are able to address single thread performance and multi-core scaling. The performance limitations of a singe core/thread are boundaries that may potentially be extended with new technology but the the current logical steps to circumvent this boundary is to incorporate multiple cores. At this point, any software developer that creates a product that caps out 1 or 2 threads is creating an artificial bottleneck and the blame, thus, should be directed at the poor coding and not the lack of resources. The other caveat is that the enthusiast market is not the primary target for game developers and, therefore ,they will not code for multiple threads until their target audience starts to feel the performance issues. On the other hand, observation of the professional software development show a greater utilization of multiple threads. Newer generation of productivity software has been increasingly coded for multiple threads as competition is present and the consumer will spend their $$$ on the product that can save the most time and $$$.

No matter what you feel about how games utilize cpu cores, what you wish will not change things for games that are already published. The best one can hope for is a patch that will utilize cpu resources more efficiently. Can you wish for more multithreaded games, yes. Can you encourage via comments on forums, etc, publishers to make more multithreaded games, yes. Can you change the games that are already published, not likely unless you work for a developer or are a super programmer with access to the source code for a game.

Still, you havent answered the question of how many cores a developer should program for to not introduce an "artificial bottleneck". Obviously you dont think 2 is sufficient. How about 4, or 8, or 16? Still, the point is, for games that use fewer cores, you can complain about it, or you can live with it and get a cpu with fast single threaded performance that gets the best out of what is available. Its kind of like football. I live in the city where the Minnesota Vikings play. We have a very mediocre quarterback at best. Does the coach plan the game wishing he had Tom Brady or Peyton Manning at quarterback? No he has to make the best of what resources he has, not wish for something that doesnt exist.

 

[crashtech]

So for a $330 CPU & GPU budget, lets go a step further and include the motherboard into the mix. Let us just assume $70 for the motherboard to bring the balance to $400 for the three components.

CPU: Intel Core i5-4430 3.0GHz Quad-Core Processor ($174.99 @ Amazon)
Motherboard: MSI B85M-G43 Micro ATX LGA1150 Motherboard ($74.24 @ Amazon)
Video Card: Sapphire Radeon HD 7850 2GB Video Card ($153.98 @ Newegg)
Total: $403.21

CPU: AMD FX-6300 3.5GHz 6-Core Processor ($117.98 @ Outlet PC)
Motherboard: Biostar TA970 ATX AM3+ Motherboard ($69.99 @ Newegg)
Video Card: Sapphire Radeon HD 7950 3GB Video Card ($199.99 @ Newegg)
Total: $387.96

CPU: Intel Core i3-3220 3.3GHz Dual-Core Processor ($118.99 @ Amazon)
Motherboard: Gigabyte GA-B75M-HD3 Micro ATX LGA1155 Motherboard ($64.99 @ Microcenter)
Video Card: Sapphire Radeon HD 7950 3GB Video Card ($199.99 @ Newegg)
Total: $383.97

Each of these 3 builds would have their benefits and weaknesses. One could argue that a particular bundle may be superior at a particular task but arguing that any one of these three are hands down superior to the other two would be a fallacy. Systems are built based on the need of the user. The argument for the general performance of a single component is typically moot in a system build that has a budget and goal in mind.

This is a very useful example, thanks. Even though I throw out the excuse making and finger pointing that accompanies any discussion of AMD's single thread performance, as occurred after the above quoted post, it is useful to point out that when allocating limited resources for a gaming build, it's more beneficial to spend approximately twice the dollars on the GPU than the CPU.

I believe the advantage would be with AMD in the above example in gaming overall, though there will obviously be specific contrary scenarios.

 

[Idontcare]

This is true but there is also a built in fallacy with this conclusion. When we fail to acknowledge the root cause of the lack of performance, we continue to place the burdens of better performance on the wrong party. It is true that AMDs single thread performance is poor. It is also true that there are limitations to single core performance. It is further true that multi-core scaling can be improved drastically(more true for AMD). AMD and Intel are able to address single thread performance and multi-core scaling. The performance limitations of a singe core/thread are boundaries that may potentially be extended with new technology but the the current logical steps to circumvent this boundary is to incorporate multiple cores. At this point, any software developer that creates a product that caps out 1 or 2 threads is creating an artificial bottleneck and the blame, thus, should be directed at the poor coding and not the lack of resources. The other caveat is that the enthusiast market is not the primary target for game developers and, therefore ,they will not code for multiple threads until their target audience starts to feel the performance issues. On the other hand, observation of the professional software development show a greater utilization of multiple threads. Newer generation of productivity software has been increasingly coded for multiple threads as competition is present and the consumer will spend their $$$ on the product that can save the most time and $$$.

I think you are doing exactly that here.

What is the motivation for the creation of these products, be they software or hardware?

Why does Intel seek to boost ILP for their cores?

Why does AMD seek to boost TLP for their CPUs?

It is all to make money, profits are the motivation. And the customer buys the product that the customer views as best serving their needs.

If Intel and AMD were not convinced they would make money from improving their products then they wouldn't spend the money improving their products.

So lets come back to software. What motivation is there for a software developer to increase the TLP of their code, given the added expense that will come from hiring more programmers and/or paying for an even deeper and longer app development timeline?

Do they make more money, more profit?

Are people making purchasing decisions regarding the software based on the threading of the program itself?

When AMD decided to go dual-core with their X2 cpus they saw fortunes coming to them. That convinced Intel to follow suit, and fortune came to them (and away from AMD).

So we know the market works.

Now why isn't it working for software? Why do you suppose software might seem "stuck" at 1-2 threading instead of 8-10 threading? (or some arbitrarily higher level of multi-threading)

The obvious conclusion here, IMO, is that the market does not value the higher-threading performance to the degree that is necessary to justify the associated cost increase that comes with developing a more threaded program.

So, in this reality, a reality where people are willing to pay a premium for hardware improvements that result in performance improvements but are not willing to pay a premium for software improvements that result in performance improvements, who's fault is it when the hardware fails to deliver the performance improvements that the market is willing to pay for?

It is not the fault of the software companies that people are not willing to pay a premium to buy software which costs more because it is more multi-threaded. That is a market decision and the software companies have to operate within the limits of what the market is willing to value and pay for.

 

[Slomo4shO]

Did you bother to read past what you quoted in bold?