I think two 7950s would bottleneck that CPU but one probably won't and I'm thinking one would be a good match and balance for that CPU. Those are my thoughts but I'm not a gamer so hopefully, you will get some more input.
With the one video card you should be fine especially with the NB running that high. I wouldn't worry about it that much, at that point you shouldn't really have a bottleneck, just a theoretical limitation.
A 4ghz CPU speed on a phenom II doesn't state anything about the north bridge frequency storm-chaser. If fsb is left at default, you can run 4ghz on the CPU with nb at default speeds by only changing CPU multi.
In some games the CPU could bottleneck crossfire. Most games are only going to be utilizing four of those cores anyway, except for frostbite engine games.
CF would be likely to improve your gaming performance tho. But to get the full potential, it would be better matched with a stronger CPU.
A 4ghz CPU speed on a phenom II doesn't state anything about the north bridge frequency storm-chaser. If fsb is left at default, you can run 4ghz on the CPU with nb at default speeds by only changing CPU multi.
In some games the CPU could bottleneck crossfire. Most games are only going to be utilizing four of those cores anyway, except for frostbite engine games.
CF would be likely to improve your gaming performance tho. But to get the full potential, it would be better matched with a stronger CPU.
This leads to the next problem. That's cpunb frequency which you are focused on, and only really relevant to memory related operations. Cpunb frequency doesn't make a meaningful impact for 3d performance which is cpu limited... This isn't a question of available data or communication bandwidth, it is a matter of core performance and efficiency.
The performance of the CPU will somewhat limit the maximum potential of the GPUs he is considering when used in crossfire. Cpu-nb does nothing to change that.
This leads to the next problem. That's cpunb frequency which you are focused on, and only really relevant to memory related operations. Cpunb frequency doesn't make a meaningful impact for 3d performance which is cpu limited... This isn't a question of available data or communication bandwidth, it is a matter of core performance and efficiency.
The performance of the CPU will somewhat limit the maximum potential of the GPUs he is considering when used in crossfire. Cpu-nb does nothing to change that.
That sounds pretty close to what I see when really digging for answers in many places. He likely might benefit some by RE-locking the CPU and pushing for a higher total cpu clock speed, since each of the four cores has the full complement of FPU and Integer units unlike the FX series of processors.
RGone...
I have been searching for the article that covers this, but the bottom line is, the Intel CPU's will scale SLI/CFx better than the AMD's do with high end cards... I remember that.
This leads to the next problem. That's cpunb frequency which you are focused on, and only really relevant to memory related operations. Cpunb frequency doesn't make a meaningful impact for 3d performance which is cpu limited... This isn't a question of available data or communication bandwidth, it is a matter of core performance and efficiency.
The performance of the CPU will somewhat limit the maximum potential of the GPUs he is considering when used in crossfire. Cpu-nb does nothing to change that.
His NB overclock of 3000mhz will vastly improve the CPU L3 cache performance, in fact nearly doubling the L3 cache write speed, and reducing L3 latency from ~8.2ns to 5.4ns. The result is improved CPU throughput and better overall system performance, including graphics, not just in the memory department, as you claim.
FACT: NB overclocking equates to real world performance gains with a Zosma or Thuban based CPU: ie higher frame rates and better game play. Not miniscule gains, but serious improvements. Some games yield a 15% increase in FPS, or higher. I have a very similar rig with nearly identical specs, so I can personally attest to the fact that CPU NB overclocking will improve the capabilities of his system and thereby lessen the imposed CPU bottleneck.
If you have a minute, can you post up some some results of the testing you did showing that it will improve FPS in games (canned benchmark is best) or even synthetic benchmarks (focusing on the GPU score). I've seen both sides of this discussion before but not really any benchmarks.
Lets get real about this for a second. Whether or not the CPU is going to be a bottleneck or not really doesn't matter here. Adding a second GPU is going to be the largest gain in FPS he can get for a given amount of $.
Could he get a faster CPU/MOBO/RAM combo and raise his FPS some? Absolutely, but even if his current combo only gets say 85% of what 79XX in crossfire is fully capable of he is still going to get a massive jump in performance over a single card. Far more than spending the same amount of cash on new CPU/MOBO/RAM could do.
Beyond that, adding a second GPU now and knowing that Intel will have a new platform out in a few months, and AMD should have there next iteration out by next spring he will have the option to carry those over and get yet another big gain by replacing only part of the system.
While you are right that increased CPUNB improves overall system performance, and at no point have I challenged that, you are completely out to lunch on the level of impact this has. While you can be theoretically on the right track, without actual testing data or any specific real world numbers to properly frame your opinion, you are expressing a stance which is coo-coo for coco nuts. CPU-NB just doesn't have that meaningful of an impact on 3D performance, and certainly not in the way you suggest it does for improving CF/SLI performance when the limiting factor is the computational power of the CPU and its ability to crunch numbers.
What you are missing here, in a CF situation, the bottleneck is the CPU's computational ability - that is improved in a very moderate way by increasing CPU-NB speeds, but it doesn't eliminate the problem. I've continued this discussion with you up to this point because statement is factually inaccurate:
With the one video card you should be fine especially with the NB running that high. I wouldn't worry about it that much, at that point you shouldn't really have a bottleneck, just a theoretical limitation.
By your own admission, CPU-NB increases L3 cache writes and latency, and it also greatly improves bandwidth and performance in the system memory department. Increasing CPU-NB, in the best possibly circumstance for reflecting the best case scenario for the influence of CPU-NB - memory performance - has a MINIMAL impact. Here's some actual testing of the impact of CPU-NB frequency in this best case scenario, to give your stance the most fair treatment possible.
By increasing CPU-NB frequency by 276MHz, AS WELL AS tightening ram timings from CL8 to CL7, I cut 11 seconds off my PhII 965 superpi32M time. Those two improvements yielded less than a 2% improvement in my overall score, from 15m 14s, to 15m 3s.
Slower CPU-NB and RAM:
Faster CPU-NB and RAM:
Why only a 2% improvement? Because this is an example of a test that is CPU bottlenecked - the biggest limiting factor is the CPU frequency, not the memory frequency. Memory performance is important to overall performance, but its relatively minor in comparison to the CPU performance.
On the same processor, I cut 3.5 minutes off my time by increasing the CPU frequency from 4.5GHz to 6.3GHz. Then by going to 6.7GHz, I cut 45 seconds more off my time... My best time on the processor for SP32M was 10m 42s, and it was helped by running the CPU-NB at 4.4GHz, however that only improved my score by a few percentage points, and the main driver was increased CPU frequency.
10m 42s (faster core, memory, cpu-nb):
11m 28s (slower core, memory, cpu-nb):
So just as in SP32M, when considering crossfire on the CPU in this thread, CPU-NB frequency is the least important factor in the equation. CPU frequency is number one, cas latency is number two, and CPU-NB is somewhere in the list after that... Extra bandwidth and improved latency is of minimal help, when the core engine itself isn't able to do the work quickly enough.
With crossfire, performance scales more closely to an increase in CPU power, not memory bandwidth. CPU-NB will help a tiny bit, but you have thus far completely overstated its impact, and over extended your perceived knowledge of the situation.
Lets get real about this for a second. Whether or not the CPU is going to be a bottleneck or not really doesn't matter here. Adding a second GPU is going to be the largest gain in FPS he can get for a given amount of $.
Could he get a faster CPU/MOBO/RAM combo and raise his FPS some? Absolutely, but even if his current combo only gets say 85% of what 79XX in crossfire is fully capable of he is still going to get a massive jump in performance over a single card.
Lets get real about this for a second. Whether or not the CPU is going to be a bottleneck or not really doesn't matter here. Adding a second GPU is going to be the largest gain in FPS he can get for a given amount of $.
Could he get a faster CPU/MOBO/RAM combo and raise his FPS some? Absolutely, but even if his current combo only gets say 85% of what 79XX in crossfire is fully capable of he is still going to get a massive jump in performance over a single card. Far more than spending the same amount of cash on new CPU/MOBO/RAM could do.
I'd say, absolutely yes. In crossfire, the cards would just stretch their legs even more with a newer CPU behind it... But as an upgrade to the platform you have right now, without changing anything else, another 7950 in crossfire would be a noticeable improvement in performance.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
