How Much Will I Gain in Gaming When Upgrading from a Phenom II X4 965 to an X6 1090T?

[LOL_Wut_Axel]

IPC is application depended, you cannot take itunes/lame (which are older apps) and conclude that Bulldozer always has lower IPC than Thuban.

Except it does, because it was designed to counteract the low IPC via high clock speeds. This has been common knowledge for, what, almost 2 years now?

Stop trying to sugarcoat it. There's clearly something wrong here. The X6 1100T is faster than the FX-6100, period. If it's not, outside of apps which use new instructions like AVX, then there's something wrong.

iTunes/LAME are the best programs for showing single-threaded performance. It correctly displays differences in architecture speeds in both AMD and Intel processors. You can clearly see the evolution going from the 1st to 2nd gen Core processors. If the excuse is now that you don't like iTunes, here's Cinebench as well:

Exact same thing. And again, this is not taking into account the 10-20% loss from the modular architecture.

 

[NTMBK]

Bulldozer adds support for a lot of new instructions. SSE4.1,SSE4.2,AVX- not to mention all of AMD's ones like FMA4 and XOP. Piledriver then adds FMA3, which is also coming in Haswell. Phenom II has none of those instructions, and it is easily conceivable that modern games will have optional code paths using those instructions.

 

[2is]

Planned obsolescence is not a conspiracy theory. Also, if you don't understand the concept of single-threaded performance we're better off not getting into an argument. Single-threaded performance directly affects multi-threaded performance. And if iTunes doesn't sound fancy enough for you, here's LAME:

Same exact result. There is ZERO reason outside of planned obsolescence for the FX-6100 to be faster than the 1100T.

I understand single threaded concepts clearly. I also understand there's more to what determines application performance (even single threaded) then raw single threaded power. Various instruction sets, memory bandwidth, cache architecture to name a few.

 

[AtenRa]

@LOL_Wut_Axel

All the applications you have quoted (iTunes, Lame, Cinebench) were written before Bulldozer was released. Again IPC is application depended, if the application is written to take advantage of the Bulldozer architecture it will perform faster in BD/PD CPUs than Thubans, is not so hard to understand.

 

[SPBHM]

the trouble with going to 6 cores but keeping the rest unchanged like this (even the same 6MB l3...), is that in many games you are not going to see any improvement, but it's a pretty good upgrade when it comes to more parallelized work (rendering, file compression), and some newer games will also take some advantage... but it seems like the PII is really falling behind AMDs newest CPUs for gaming.... the 6300 is probably a faster CPU for gaming (even the trinity APUs might be in some games), and it uses a comparable (or better) amount of power.

the thing is, for $70 you can't really go wrong with that CPU... it's single module trinity and Pentium territory.
but don't expect a gain in every single game...


Piledriver has a lot of cache, and significantly improved memory performance overall compared to PII, this alone should help for many games, I think...

and it seems AMD really fixed something that helped gaming with Piledriver, if you look at something like Crysis 2

 

[inf64]

Piledriver core is much more than just 1-5% faster than BD as Lol_Wut states. Facts:

http://www.hardware.fr/articles/880-6/bulldozer-vs-piledriver-4-ghz.html Hardware.fr tested both at fixed clock(4GHz),all other things being equal(board,memory,GPU,HDD). They have concluded that in desktop applications(not games), PD has 7.7%(1.3% to 16.7) IPC improvement over BD,both at 4Ghz. In games, PD has even higher IPC advantage,a significant 13.5% (8.1% to 20.8%) ! Average comes down to 10.6% which is very significant IPC jump for a mere "quick fix". Couple this with higher average OC numbers Vishera can reach Vs Bulldozer and overall lower power draw when both are OCed and you get a "tock" in intel's terms . Vishera is much better product than X6,that's a fact supported by numbers.

 

[Rvenger]

Because AMD still manufactures Phenom II processors

I don't know where you got that from but AMD stopped manufacturing 45nm shortly after BD was released. (With the exception of the 960T which I believe is now discontinued)


http://www.tomshardware.com/news/amd-cpu-apu-athlon-phenom-Llano-Bulldozer,14173.html
http://news.softpedia.com/news/AMD-Stopped-Shipping-Phenom-II-and-Athlon-II-CPUs-238017.shtml

 

[LOL_Wut_Axel]

Bulldozer adds support for a lot of new instructions. SSE4.1,SSE4.2,AVX- not to mention all of AMD's ones like FMA4 and XOP. Piledriver then adds FMA3, which is also coming in Haswell. Phenom II has none of those instructions, and it is easily conceivable that modern games will have optional code paths using those instructions.

Sounds nice and dandy, except they do not use those instructions because those are not suited for gaming but for things like encryption and encoding. Games still use SSE3 and its variants.

So... no, that's not the answer.

 

[LOL_Wut_Axel]

I don't know where you got that from but AMD stopped manufacturing 45nm shortly after BD was released. (With the exception of the 960T which I believe is now discontinued)


http://www.tomshardware.com/news/amd-cpu-apu-athlon-phenom-Llano-Bulldozer,14173.html
http://news.softpedia.com/news/AMD-Stopped-Shipping-Phenom-II-and-Athlon-II-CPUs-238017.shtml

Either they have a huge stockpile of them or they're still manufacturing some models. Both online retailers like Amazon and Newegg are still selling them, and physical retailers like TigerDirect. That story was released almost a year and a half ago.

 

[KingFatty]

My personal experience is that Vishera (FX-6300) was better clock-for-clock in a CPU intensive game. I specifically benchmarked Starcraft 2 back-to-back with a Phenom II vs. FX-6300, my results are posted in another thread.

However, I would have to admit that the value of the FX-6300 really showed itself with overclocking, compared to the Phenom II. I would have gone Intel if not for 1) ability to overclock a large amount, and 2) ability to buy just the chip as a simple drop-in upgrade letting me keep my old motherboard/RAM/etc. untouched and just swap the chip and re-apply thermal paste.

But anecdotally, I found my old Phenom II overclocked to its max of near 3.9 GHz with 2.6 GHz CPU-NB did like 45 FPS average in my particular replay. Just the default/stock setting on the FX-6300 (3.5 GHz) roughly matched the Phenom II. Then, the overclocking of the FX-6300 made a big difference, plus it uses less power which I think helps achieve the huge overclocks. But I think Phenom II just hits a wall around 4 GHz overclock, whereas FX goes much higher.

 

[LOL_Wut_Axel]

I understand single threaded concepts clearly. I also understand there's more to what determines application performance (even single threaded) then raw single threaded power. Various instruction sets, memory bandwidth, cache architecture to name a few.

Instruction set we already discussed. As for memory bandwidth, it's a hit or miss with bandwidth itself being higher but latency being higher as well:

Cache latency isn't anything to write home about, either:

And we already know that L2 and L1 cache latency is higher than Thuban.

 

[tweakboy]

I bought a Phenom II X4 965 for my secondary, more portable gaming/multimedia machine (specs in sig). Tomorrow I will buy a second-hand Phenom II X6 1090T for $70 and wanted to know how much of an upgrade I can expect in games given that recent ones have started to take advantage of more than four cores, as you can see comparing the FX-4300 vs the 6300. The graphics card installed in that machine is a GeForce GTX 570, and just like the X4 I'll have the X6 overclocked to 3.6GHz and 2.4GHz CPU-NB.

Maybe 2fps to 5fps more ..... However in desktop multi threaded apps the difference will be a good one.

gl

 

[LOL_Wut_Axel]

Piledriver core is much more than just 1-5% faster than BD as Lol_Wut states. Facts:

http://www.hardware.fr/articles/880-6/bulldozer-vs-piledriver-4-ghz.html Hardware.fr tested both at fixed clock(4GHz),all other things being equal(board,memory,GPU,HDD). They have concluded that in desktop applications(not games), PD has 7.7%(1.3% to 16.7) IPC improvement over BD,both at 4Ghz. In games, PD has even higher IPC advantage,a significant 13.5% (8.1% to 20.8%) ! Average comes down to 10.6% which is very significant IPC jump for a mere "quick fix". Couple this with higher average OC numbers Vishera can reach Vs Bulldozer and overall lower power draw when both are OCed and you get a "tock" in intel's terms . Vishera is much better product than X6,that's a fact supported by numbers.

You can't use games to measure IPC differences. There's other big variables to take into account because the graphics card is a big part of the equation as well. Either way, those numbers look hugely suspect when you look at the AT review and compare with the FX-8320:

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/3

Except for compiling with Visual Studio, it doesn't look to be more than 5% faster per-clock than the 8150. The 8150 is clocked 100MHz higher, but at the same time Turbo is more effective on Piledriver.

 

[Piroko]

Seems to me like developers are simply intentionally crippling the Phenom II processors given the fact that their per-core, per-clock performance is higher, whether it be in integer or floating point. The Phenom II X6 1090T/1100T is faster than the FX-6300, so why would that be different in games?

Some explanations:
- a lot of Games love L2, BD/PD has at worst twice the amount of L2
- BD/PD has higher bus speed which has at least in Starcraft shown to be a limiting factor
- even with 'similar' ipc you still get higher clocks
- perhaps a slight gain with the revamped Win8 scheduler.

 

[2is]

Instruction set we already discussed. As for memory bandwidth, it's a hit or miss with bandwidth itself being higher but latency being higher as well:

Cache latency isn't anything to write home about, either:

And we already know that L2 and L1 cache latency is higher than Thuban.

Some things benefit with more memory bandwidth some with better latency, but bandwidth usually triumphs or do you think DDR1 is faster than DDR3? and some with simply having MORE of it. im talking both cache and ram here. Instructions have been discussed, which is to say. You're making excuses for every possible scenario and all of the excuses are guess work tweaked to support further guess work of obsolescence.

 

[LOL_Wut_Axel]

My personal experience is that Vishera (FX-6300) was better clock-for-clock in a CPU intensive game. I specifically benchmarked Starcraft 2 back-to-back with a Phenom II vs. FX-6300, my results are posted in another thread.

However, I would have to admit that the value of the FX-6300 really showed itself with overclocking, compared to the Phenom II. I would have gone Intel if not for 1) ability to overclock a large amount, and 2) ability to buy just the chip as a simple drop-in upgrade letting me keep my old motherboard/RAM/etc. untouched and just swap the chip and re-apply thermal paste.

But anecdotally, I found my old Phenom II overclocked to its max of near 3.9 GHz with 2.6 GHz CPU-NB did like 45 FPS average in my particular replay. Just the default/stock setting on the FX-6300 (3.5 GHz) roughly matched the Phenom II. Then, the overclocking of the FX-6300 made a big difference, plus it uses less power which I think helps achieve the huge overclocks. But I think Phenom II just hits a wall around 4 GHz overclock, whereas FX goes much higher.

I'll run some benchmarks when I buy it. I'll be back in around two hours. But so far it just seems to me like Starcraft II is a game that the Phenom IIs were intentionally crippled in. Again, slower than a 1st gen Core i3? Makes no sense.

 

[tweakboy]

Get the K so you wont have any problems and can freely OC.

 

[KingFatty]

I'll run some benchmarks when I buy it. I'll be back in around two hours. But so far it just seems to me like Starcraft II is a game that the Phenom IIs were intentionally crippled in. Again, slower than a 1st gen Core i3? Makes no sense.

Well I think part of the big problem with SC2 is that Phenom II doesn't get to fully shine - it was rather mis-represented in all the benchmarks and even on overclocking sites, because of the universal mistake of failure to *also* overclock the CPU-NB.

So I don't think the Phenom IIs were so much as crippled, but rather all the overclock websites seemed to misunderstand the Phenom II and fail to overclock the CPU-NB when overclocking the CPU alone for testing SC2.

 

[ThePeasant]

You're the one demonstrating a lack of understanding of IPC. It's mostly a function of the platform architecture as well as the code being executed. Key point here being that the other programs you are using to talk about the supposed performance of games are exactly that. Other programs.

It is completely possible for a microarchitecture to have generally weaker IPC while excelling with certain instruction profiles (look no further than the P4). It is completely possible, for example, for a program (ILP bound for whatever reason) to favour higher clock speeds over a larger execution width or for the program's memory access/branch patterns to favour the newer predictors in BD/PD/SB or for the program to favour larger caches over faster ones, or more likely a combination of these and other peculiarities of the platform, coding style, compilers and even instruction sets. The kinds of permutations you get from different platforms and programs are so vast, that we can only really give an expected value for typical programs and perhaps some bounds. To use one or even a hundred programs as proof that BD/PD is always inferior in IPC to K10 is completely fallacious.

I find AMD actively paying developers to cripple their product among the least likely of reasons but I will admit I can see a motive for it and the differences are certainly interesting. However, it would take more than a few GPU bottlenecked benchmarks comparing K10 to prior gen Core to convince me and quite a few other people here it seems.

 

[LOL_Wut_Axel]

You're the one demonstrating a lack of understanding of IPC. It's mostly a function of the platform architecture as well as the code being executed. Key point here being that the other programs you are using to talk about the supposed performance of games are exactly that. Other programs.

It is completely possible for a microarchitecture to have generally weaker IPC while excelling with certain instruction profiles (look no further than the P4). It is completely possible, for example, for a program (ILP bound for whatever reason) to favour higher clock speeds over a larger execution width or for the program's memory access/branch patterns to favour the newer predictors in BD/PD/SB or for the program to favour larger caches over faster ones, or more likely a combination of these and other peculiarities of the platform, coding style, compilers and even instruction sets. The kinds of permutations you get from different platforms and programs are so vast, that we can only really give an expected value for typical programs and perhaps some bounds. To use one or even a hundred programs as proof that BD/PD is always inferior in IPC to K10 is completely fallacious.

I find AMD actively paying developers to cripple their product among the least likely of reasons but I will admit I can see a motive for it and the differences are certainly interesting. However, it would take more than a few GPU bottlenecked benchmarks comparing K10 to prior gen Core to convince me and quite a few other people here it seems.

The xbit review I posted ran games at 1024x768. I think that would make it more CPU bound.

And it is true that overclocking the CPU-NB makes the same if not more difference than overclocking the Core in games.

But again, games do not use the new instruction sets in BD or PD and we know that IPC is lower in both than in K10.5. And even IF IPC were higher, how are games now running faster on the 4100 when it has 2M/4T and doesn't run at that high a clock speed? It was much slower before, and now its faster even though newer instruction sets have not been used in games. Not only are each of its cores slower, but it can only address 2-4 threads while the X6 can address 6 while not having a CMT penalty.

I purchased it already, and he confused the name. It was actually an 1100T. Not bad at all for $70. Benchmarks to come later.

 

[inf64]

You can't use games to measure IPC differences. There's other big variables to take into account because the graphics card is a big part of the equation as well. Either way, those numbers look hugely suspect when you look at the AT review and compare with the FX-8320:

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/3

Except for compiling with Visual Studio, it doesn't look to be more than 5% faster per-clock than the 8150. The 8150 is clocked 100MHz higher, but at the same time Turbo is more effective on Piledriver.

Nothing is fishy in that test. The problem of other reviews is that they failed to do tests at equal clock and with disabled Turbo(like Hardware.fr did). What is confusing you is that 8150 in actuality runs (almost always) at ~3.9Ghz no matter if workload is very well threaded or not(if it's not than it's north of 3.9Ghz). 8350 on the other hand has not so much room to boost,just a lowly 5%. In most workloads,8350 actually runs at 4.1Ghz and fluctuates the clock on all cores between 4 and 4.1Ghz. So in reality,8150 and 8350 really do run at very similar clocks(both have same top single core turbo of 4.2Ghz) and big differences in performance between the two are really core optimizations/tweaks. The results of hardware.fr speak for themselves : IPC in applications increased (on average) by 7.7% and in low resolution gaming by 13.5%. There is no other way around this,sorry.

 

[Durvelle27]

this thread makes me laugh

 

[ThePeasant]

But again, games do not use the new instruction sets in BD or PD and we know that IPC is lower in both than in K10.5. And even IF IPC were higher, how are games now running faster on the 4100 when it has 2M/4T and doesn't run at that high a clock speed? It was much slower before, and now its faster even though newer instruction sets have not been used in games. Not only are each of its cores slower, but it can only address 2-4 threads while the X6 can address 6 while not having a CMT penalty.

Once again you are assuming things you shouldn't; IPC is absolutely worse in all workloads wrt K10 and the "CMT penalty" is significant for all workloads.

Turbo and better scheduling could be one explanation why the 4100 performs the way it does (I don't think its faster btw). During the critical (supposedly single threaded) portions of computation it can turbo up, and then when the dependent parallel portions are being executed its higher base clock and four threads limits the deficit.

 

[LOL_Wut_Axel]

Once again you are assuming things you shouldn't; IPC is absolutely worse in all workloads wrt K10 and the "CMT penalty" is significant for all workloads.

Turbo and better scheduling could be one explanation why the 4100 performs the way it does (I don't think its faster btw). During the critical (supposedly single threaded) portions of computation it can turbo up, and then when the dependent parallel portions are being executed its higher base clock and four threads limits the deficit.

No, it's not. Did you even look at CineBench? Or the fact that JFAMD never appeared again because he lied saying IPC would be higher when it wasn't? In terms of absolute IPC K10.5 is faster. The only way Bulldozer is faster is with new instructions like AVX and AES, and games DO NOT use those. Also, the 960T and 1100T have Turbo as well.

Tests were also ran on Windows 8 and all this scheduling hoopla did for performance was bring it up by less than 5%. Again, there is zero reason for the FX-4100 to be faster than the 1090/1100T in games.

Anyway, I'll get the new CPU running now.

 

[BallaTheFeared]

JFAMD was fired from AMD after bulldozer, didn't you get the memo LOL?