Value of extra 2mb L3 in 2600K?

[geofelt]

Are there any tests to measure the performance value of the extra 2mb of L3 cache that you get in a 2600?

In part, I got a 2600K instead of a 2500K because it had 8mb L3 instead of 6mb L3 cache. I have no need for the hyperthreading on the 2600k, so I disabled it. My temps under load are 5c. lower now. I expect that the upper limits of overclocking would be similar, between the 2500K and the 2600K, excluding the possibility of better binning for the 2600K.

What I would like to see would be a test between the 2500K and the 2600K where hyperthreading was turned off, and the multipliers were identical. How much performance difference is there, and in what kind of workloads does it appear.

 

[Diogenes2]

You can compare the difference here..

http://www.bit-tech.net/hardware/cpus/2011/01/03/intel-sandy-bridge-review/7

See additional pages for other tests..

 

[geofelt]

The ^ tests were done with hyperthreading enabled, which distorts the measurements for those workloads that are multi thread capable. To measure the value of extra L3, I am looking for a series of tests with hyperthreading disabled.

I expect that a comprehensive test might show that different workloads get more or less benefit from added L3 cache. For my part, the question is academic, since I already have a 2600K, but for new users, such as gamers, it would be useful to know how much they should value a 2600K vs. a 2500K. I suspect not much.

 

[Ben90]

There won't be all that much difference between 8mb and 6mb of L3$. I haven't seen a review that does the apples to apples comparison that you and I want. I can however give you a great idea of the improvements by comparing two older processors. Note that the differences of cache in those graphs is a whopping 6MB difference; although, the Nehalem architecture improved the caching structure a decent amount.

Also, by having more cache you retain IPC better from overclocking. Not alot, but stalls waiting on memory waste more cycles at 5ghz than 2ghz.

 

[IGemini]

Going from past example, I'd say it amounts to an average of 5% in games and more for file compression or office programs. The best parallel I can find so far is comparing a Q9400 to a Q8400 which had a difference of 2MB L2 cache.

 

[dlamb2471]

Yeah I mean how much difference do we think the HT makes in all but the most thread heavy benchmarks? Since you turned off HT on yours, you could always try replicating some of the benchmarks out there and comparing your results.

 

[BD231]

Cache is great but from the sounds of it you won't be doing much of anything that would have you noticing the extra 2megs. 6mb's is a pretty generous amount to begin with.

 

[hamunaptra]

Yeah if you think 7MB of cache is a lot, just wait till BD comes w/ 16MB ROFL! insane.

 

[`br4dz-]

The differences between 2500k and 2600k isn't just the cache values. The 2500k is 4 core/4 thread, while the 2600k is 4 core/8 thread since the 2600k has Hyper-Threading. Most differences you see in benchmarks are due to the extra threads, not cache.

 

[BD231]

Yeah if you think 7MB of cache is a lot, just wait till BD comes w/ 16MB ROFL! insane.

In a sense. With a die shrink and added cores they're basically slapping core on top of core so the cache is needed. The ratio of cache available to each core will probably be about the same even though it's darn near twice the size.

Once apps get more threaded it'll help, hence the extra cache on the 2600k (HT).

 

[IGemini]

Yeah if you think 7MB of cache is a lot, just wait till BD comes w/ 16MB ROFL! insane.

In a sense. With a die shrink and added cores they're basically slapping core on top of core so the cache is needed. The ratio of cache available to each core will probably be about the same even though it's darn near twice the size.

Yeah, cache is made on a per-core basis. The 16MB cache will be resultant of two quad-core dies on BD, so it really makes my Q9650 with 12MB of cache (3MB/core) more "insane."

 

[beginner99]

Yeah, cache is made on a per-core basis. The 16MB cache will be resultant of two quad-core dies on BD, so it really makes my Q9650 with 12MB of cache (3MB/core) more "insane."

Yeah but they don't have the memory controller included and hence the big amount of cache to hide memory latency.

 

[IGemini]

Check your sarcasm detector. Yes, I'm aware.

 

[Twsmit]

any updates on this? I'm interested in an answer for the OP's question.