Best Gaming CPUs For The Money: January 2012 (Archive)

[Isaiah4110]

Because an FX 6300 has better performance, even in poorly threaded titles since it still has its L3 cache. The lack of L3 cache affects performance of Trinity/Richland vs FX much like it did with AM3 Athlon II vs Phenom II. Here is a benchmark result of a poorly threaded title. The 3.9ghz FX 6350 is about even with a 4.3ghz 750k. With both at 4.0ghz, the FX would win, even in poorly threaded titles, due to L3 cache.

Again, Damric's point, going back several pages in this discussion, is that this Tomshardware test was using 1866 memory, when the 760k supports 2400. Conventional wisdom has it that memory above 1600 has diminishing returns for most gaming CPUs. However, for the exact fact that the 760K does NOT have L3 cache means that memory scaling does in fact matter.

A few pages up, there are links to a post that shows up to 18% FPS gain just by taking the 760k from 1600 to 2400. Since the TH review is running 1866, that means that the 760K running at 2400 would be up to 12% faster than the FPS numbers in this article (for the CPU-limited titles of course). So for your examples, that would put the 760K@4.3Ghz StarCraft II numbers in the vicinity of 35 FPS and Skyrim in the vicinity of 87 FPS.

Plus, based on posts I've seen since following this discussion, it's entirely reasonable to expect 4.5-4.7GHz out of the 760K with a decent $30 aftermarket cooler, which would put it slightly ahead of the FX CPUs.

Bottom line is I think Damric has brought up some really good supporting evidence that should at least question the conventional wisdom around the FX vs. Athlon in the budget gaming space.

Would love to see a properly staged "budget shootout" - perhaps when the unlocked Pentium G hits the scene that'll prompt some increased attention in the budget space.

I think it is very important to note that Damric has also specified that it is not just overclocking the memory that is necessary to see such improvement gains. The NB chip also needs to be overclocked as high as it will go while still remaining stable or it will bottleneck the transfer of data between CPU and RAM.

 

[peter234]

i would say i5 4670k ive got one my self + a gtx 780 runs titanfall maxed out 60fps no lag at all maybe it drops to 59 but thats it... it pretty much stay on 60 fps all the time

 

[ingtar33]

I think it is very important to note that Damric has also specified that it is not just overclocking the memory that is necessary to see such improvement gains. The NB chip also needs to be overclocked as high as it will go while still remaining stable or it will bottleneck the transfer of data between CPU and RAM.

the NB on piledriver can only be overclocked to a point before you stop seeing gains. Its not like on the PhII... there are different bottlenecks in piledriver... typically, gains from nb overclocking end around 2600mhz... even more interesting piledriver seems to work best the closer the HT and NB speeds are to each-other. So a 2600/2600 setup would be ideal for it. Considering most high end motherboards hit 2200mhz on the NB and 2400mhz on the HT at stock settings, there isn't a lot of overclocking to do there. my own asus motherboard hits 2600/2600 without even touching voltages or other settings.

 

[damric]

You know that the FM2 chips don't have HT link? I was surprised when I discovered that, but it makes sense since the whole Chipset northbridge is on die. BD/PD only need 1:2 as opposed to 1:3 ratio for RAM:CPU-NB on Phenom II, but just like Phenom II, extra doesn't hurt.

 

[ingtar33]

You know that the FM2 chips don't have HT link? I was surprised when I discovered that, but it makes sense since the whole Chipset northbridge is on die. BD/PD only need 1:2 as opposed to 1:3 ratio for RAM:CPU-NB on Phenom II, but just like Phenom II, extra doesn't hurt.

yeah, i know. I was speaking in general, but you're right. the reason why FM2 doesn't use HT link is because AMD moved the northbridge onto the cpu when they released the piledriver based apus in trinity (and of course the athlon II x4 750k is basically a trinity a10).

As for the 1:2 you're mixing up things. Yes, on the PhenomII generally you wanted to keep your NB at x3 the frequency of your ram... so DDR3 1600 ram, ran at a frequency of 800mhz, x3 = 2400mhz... not only was that ideal for the PhII, but it actually made overclocks more stable. The trick of course was many am3 motherboards struggled to reach nb speeds in that range.

On piledriver, your 1:2 ratio sorta works... however you gain no additional overclock stability with matching, furthermore, you get almost no performance gains (small ones but largely irrelivent), furthermore, bumping your nb and ht to 2600/2600 seems to nail down the highest memory and cpu benching marks regardless of the memory AND cpu speed, and speeds over 2600 don't seem to yeild any results.

 

[damric]

@ingtar33:

Have you tried those higher speeds with matching 1:2 ram though? Like 2600MHz CPU-NB With DDR3-2600?

CPU NB: CPU NorthBridge (should not be confused with NorthBridge chipset, such as the AMD 990FX chipset)
– part of the CPU that has its own clock domain and voltage plane. CPU NB clock frequency determines the
Memory controller and L3 cache speed. CPU NB has a notable impact on overall system performance.

The CPU NorthBridge (CPU NB) clock speed determines the efficiency and bandwidth capacity of the Memory
controller. L3 cache runs at this frequency as well. CPU NB performance tuning gives a measurable boost for
overall system performance. In particular it can reduce the Memory latency and improve L3 cache bandwidth
and latency. In some scenarios the CPU NB should be tuned in order to take the full advantage of overclocked
Memory frequency and bandwidth. In general the CPU NB clock should be at least two times the Memory clock
(example in case of DDR3-1600: 2 x 800MHz = 1600MHz NB clock).

That's a excerpt from:

http://sites.amd.com/us/Documents/AMD_FX_Performance_Tuning_Guide.pdf

I see that you are running only DDR3-1600, so it doesn't make as much difference going abve that on your CPU-NB, at least according to AMD engineers ^^

 

[ingtar33]

@ingtar33:

Have you tried those higher speeds with matching 1:2 ram though? Like 2600MHz CPU-NB With DDR3-2600?

CPU NB: CPU NorthBridge (should not be confused with NorthBridge chipset, such as the AMD 990FX chipset)
– part of the CPU that has its own clock domain and voltage plane. CPU NB clock frequency determines the
Memory controller and L3 cache speed. CPU NB has a notable impact on overall system performance.

The CPU NorthBridge (CPU NB) clock speed determines the efficiency and bandwidth capacity of the Memory
controller. L3 cache runs at this frequency as well. CPU NB performance tuning gives a measurable boost for
overall system performance. In particular it can reduce the Memory latency and improve L3 cache bandwidth
and latency. In some scenarios the CPU NB should be tuned in order to take the full advantage of overclocked
Memory frequency and bandwidth. In general the CPU NB clock should be at least two times the Memory clock
(example in case of DDR3-1600: 2 x 800MHz = 1600MHz NB clock).

That's a excerpt from:

http://sites.amd.com/us/Documents/AMD_FX_Performance_Tuning_Guide.pdf

I see that you are running only DDR3-1600, so it doesn't make as much difference going abve that on your CPU-NB, at least according to AMD engineers ^^

you're right, i am rolling with ddr3 1600 9-9-9-24 timed ram; 4 sticks too, to maximize bandwidth. But my opinion is based on playing around with better clocks on the HT/NB and better ram with other systems.

 

[damric]

So in the other systems did you make it to DDR3-2600 with 2600MHz NB on any AMD CPUs? I'm curious because I can only get to about DDR3-2500 stable since my highest RAM divider is for 2400MT/s, and you know how base clock tweaking goes on these PCI-tied chips.

I had an FX-4100 not long ago that had a rock solid OC of 5GHz CPU 3GHz NB and DDR3-2000CL8. It had one of the highest Maxmem scores for AMD.

 

[ingtar33]

So in the other systems did you make it to DDR3-2600 with 2600MHz NB on any AMD CPUs? I'm curious because I can only get to about DDR3-2500 stable since my highest RAM divider is for 2400MT/s, and you know how base clock tweaking goes on these PCI-tied chips.

I had an FX-4100 not long ago that had a rock solid OC of 5GHz CPU 3GHz NB and DDR3-2000CL8. It had one of the highest Maxmem scores for AMD.

nope. the best i had rolling was 2133 9-10-10-27... but i found almost no difference between 2100mhz and 2600mhz on the nb (actually i got slightly higher scores on Sandra Memory benches with 2600mhz)

 

[damric]

So what you what your saying is, you really don't have any experience matching high speed ram with high speed northbridge on BD/PD?

 

[ingtar33]

So what you what your saying is, you really don't have any experience matching high speed ram with high speed northbridge on BD/PD?

not personally. But i do have a whole extremely active overclocking forum i'm an active member in who back up my own limited finding. Its one of those things, where if you read a dozen people's online tests verifying the lack of performance gain with nb/ram on piledriver, and you do your own tests a few times, and see the same thing, you sorta just let it lay where it is. Someone else already did all the heavy lifting and did the extensive testing. All I've done is verify their work on my own limited scale.

 

[damric]

So what you what your saying is, you really don't have any experience matching high speed ram with high speed northbridge on BD/PD?

not personally. But i do have a whole extremely active overclocking forum i'm an active member in who back up my own limited finding. Its one of those things, where if you read a dozen people's online tests verifying the lack of performance gain with nb/ram on piledriver, and you do your own tests a few times, and see the same thing, you sorta just let it lay where it is. Someone else already did all the heavy lifting and did the extensive testing. All I've done is verify their work on my own limited scale.

Well I do

 

[Wisecracker]

4770K 10% FPS+ In BF4 with R9 290X & Mantle over DX
(59% increase for Kaveri CUs)

From Tech Report

A10-7850K = A10-6700 = A10-5700 = Athlon X4 750K in BF4
(before over-clocking)

The 'UNB' speed should always exceed the clock speed of your RAMs. That's the fact, Jack.

 

[vertexx]

Why would you not go with the 4430, which can be had for $177 right now, as the entry level i5?

 

[envy14tpe]

Wow! First time I've seen the fx-6300 recommended over the i3. Just wow. It's about time.

 

[ubercake]

C'mon AMD??? My trusty 4 year old i7-960 is in the same tier as your current best processors?

We need some competition in this marketplace. Is it just too late for a 3rd player to enter the picture?

 

[Sakkura]

Wow! First time I've seen the fx-6300 recommended over the i3. Just wow.

It's not recommended over the Core i3. The FX-6300 is recommended at $120, the Core i3-4130 is recommended at $125. Pretty silly though, because the Amazon links actually have them both at a little less than $120.

 

[Sakkura]

Why would you not go with the 4430, which can be had for $177 right now, as the entry level i5?

Because the 4430 costs $179.99 while the faster 4440 costs $177.

With the 4570 at $193 and the 4440 at $177, you're paying an extra $16 (9%) for another 100 MHz basic clock, and 300 MHz at max. Turbo (3.2% and 9.1% respectively). Meh, they're pretty close in value. I'd probably recommend both of them, but picking the faster 4570 is okay by me since the price gap is not unreasonable.

The 4670 is all the way up at $220, so compared to the 4570 that's an extra $27 (14%) for 200 MHz across the board (5.6% to 6.3%), clearly a bad deal.

 

[vertexx]

Why would you not go with the 4430, which can be had for $177 right now, as the entry level i5?

Because the 4430 costs $179.99 while the faster 4440 costs $177.

With the 4570 at $193 and the 4440 at $177, you're paying an extra $16 (9%) for another 100 MHz basic clock, and 300 MHz at max. Turbo (3.2% and 9.1% respectively). Meh, they're pretty close in value. I'd probably recommend both of them, but picking the faster 4570 is okay by me since the price gap is not unreasonable.

The 4670 is all the way up at $220, so compared to the 4570 that's an extra $27 (14%) for 200 MHz across the board (5.6% to 6.3%), clearly a bad deal.

Yeah, I had the 4430, then switched to 4440 then back - prices are fluctuating by ~$10 either way with these CPUs. Amazon charges sales tax in my state so, I can't get the 4440 for $177. And $193 right now for the 4570 is probably the better deal.

 

[caamsa]

Are you going to do a full review on the Kibini desktops? esp the 5350 and drop in a low and high end discrete graphics card so we can compare the performance to the mainstream processors? I saw one review where they used a 750TI with the 5350 and it did rather well for gaming.

 

[Thedarklordx64]

@ TomsHardWare , PLEASE UPDATE THE CHARTS TO INCLUDE CPU's FROM 2014!!! There is no Ivy Bridge E, how long has that one been out, there is no AMD Kaviri, etc. Can you PLEASE update the chart and not just repost the same thing month after month.

 

[Onus]

Are you going to do a full review on the Kabini desktops? esp the 5350 and drop in a low and high end discrete graphics card so we can compare the performance to the mainstream processors? I saw one review where they used a 750TI with the 5350 and it did rather well for gaming.

This please. The review (on PCPER) was very promising. It looks like those satisfied with "High" (rather than UltraMaxOhWOW!) settings might be very happy with this cheap little chip. It also needs to be added to the CPU charts.

 

[John Philips]

Where would top of the APU´s Kaveri A10 stand on this list?

 

[de5_Roy]

Where would top of the APU´s Kaveri A10 stand on this list?

in the hierarchy chart, against haswell core i3. in the recommendations, probably never, as a10s are currently priced. $185 a10 is barely cheaper than the top fx8350 cpu, pricier than fx8320 and is hugely undercut by fx6300.

 

[RedJaron]

Why would you not go with the 4430, which can be had for $177 right now, as the entry level i5?

Because the 4430 costs $179.99 while the faster 4440 costs $177.

With the 4570 at $193 and the 4440 at $177, you're paying an extra $16 (9%) for another 100 MHz basic clock, and 300 MHz at max. Turbo (3.2% and 9.1% respectively). Meh, they're pretty close in value. I'd probably recommend both of them, but picking the faster 4570 is okay by me since the price gap is not unreasonable.

The 4670 is all the way up at $220, so compared to the 4570 that's an extra $27 (14%) for 200 MHz across the board (5.6% to 6.3%), clearly a bad deal.

Yeah, I had the 4430, then switched to 4440 then back - prices are fluctuating by ~$10 either way with these CPUs. Amazon charges sales tax in my state so, I can't get the 4440 for $177. And $193 right now for the 4570 is probably the better deal.

I still think the 4440 needs to be mentioned here. A $70 gap is pretty sizable and that small clock boost doesn't mean much of anything when gaming as the3.1 GHz is more than beefy enough. The real story of stepping from the i3 to i5 is the four physical cores and larger L3. I'll pay $50 for that, but I don't need to spend ~$20 extra for a performance boost I won't even notice outside benchmarks.