If the K difference in how Intel CPU's benchmark relfected outside of graphics??

[Salt]

I have a Core i5 2300. Its a decent CPU that paired up with a HD 6870 1GB, and 8GB of system ram is fast enough for now for what I'm doing: Games, scientific applications, the occasional bit of video editing, and tons of homework for pharmacy college.

However, I'm looking at upgrading potential in the future. Whenever I see a benchmark for a "K" series processor on Passmark it is always substantially ahead of its non-K variety even at similar clock speeds. Okay, it has a better graphics processing unit *(HD3000 vs HD2000 in the non-K), and I understand passmark reflects that, and it probably should. What I am asking is, does the difference show up in actual system performance otherwise? The only time I'll be using HD2000 or HD3000 graphics is if I'm RMA'ing a graphics card, or maybe trouble shooting a driver. Though I suppose its not inconceivable that I could find a need for 8 monitors ...

So real world, should I be extrapolating results the 2600 (non K) performance to see what say a 2700K can really do, or take those 2700K benchmarks as what they are for running with a graphics card?


Might I do better with the upgrade dollar going to SSD's rather than looking at processors at all down the road? I took the time to put in reasonably quick 7200RPM drives, but its nowhere near SSD level performance.

[David (PassMark)]

Going from a normal HDD to a SSD will make your computer seem a lot faster than upgrading the CPU from the i5 you already have.

Dual monitors will also increase your productivity more than an faster CPU.

As you aren't going to be using the onboard graphics get the non K CPU variant. Unless you are planning on overclocking, in which case get the K CPU.

[Salt]

Definitely on the monitors...

I've been a multi-monitor user since the early Matrox cards came out. (I wish they made something competitive with AMD & NVIDIA, but no dice!) I currently run 3 displays, with plans for 4 displays when I get a bigger desk. That was one of the things which got lead me to the Sapphire Flex 6870 I purchased, it had good multi-monitor support without (some of) the hassles. ATI drivers & software DO make me miss working with Matrox though. Just an example, when my computer resumes from sleep, or even turning off monitors it has a hard time finding all of the DVI displays, even the ones that aren't run through display port. Its probably a catalyst issue, but it is an annoyance to have to push the input buttons on the monitors, and then use hydravision to (usually successfully, but not always) return things to their rightful monitors & positions.



I think you are right about the SSD vs CPU upgrade path for bang for the buck. I'm expecting several of the science applications I run to start using the GPU to help (a few of them already have, but these folks are pretty slow to adapt to things that don't have extensive error checking).

Probably by the time the Core i7 reaches the price point I would buy it, it will be obsolete. 6 years ago I needed a workstation to do the tasks I'm doing now with reasonable snappiness. A lot changes in PC's in a couple years.

Thanks for your advice!

[David (PassMark)]

There will be a whole new collection of reasonably priced "Ivy Bridge" CPUs coming out around March / April 2012. They should wipe the floor with the existing CPUs.

[Seronx]

There will be a whole new collection of reasonably priced "Ivy Bridge" CPUs coming out around March / April 2012. They should wipe the floor with the existing CPUs.

Not really...

Ivy Bridge CPU has the same clocks and the same performance just a mild drop in TDP

The GPU is basically the same GPU with 4 more EUs

and you really have to fix that problem with K processors....

I told everyone about the bug with the Intel Sandy Bridge and they overclocked beyond the point where the CPU reads the multiplier...
Now you have a bentmark on your hands good luck fixing it

http://www.cpubenchmark.net/overclocked_cpus.html
i7 2600 @ averaged: 4.5GHz = i7 2600K @ 3.4GHz

I congratulate you for totally ignoring me
http://www.cpubenchmark.net/high_end_cpus.html

Or at least fix it by Passmark 8 and remeber most transcoding applications nowadays use SSE4.1/4.2 that don't use AVX Int or Xop Int

[David (PassMark)]

All the commendatory I have read is that Ivy Bridge will perform better than Sandy Bridge.

Should be higher turbo boost speeds, architecture improvements, faster larger cache, and much better GPU.

Clock speed doesn't count for all that much nowadays. Same clock speed doesn't mean same performance.

> and you really have to fix that problem with K processors

Benchmark works fine and is accurate.
It just happens that the K processors perform better, on average, in real world systems.. Where there is an issue is in the accurate splitting up and filtering of submitted results that are from a poorly configured system, or from a system with overclocking. Every new range of CPUs creates new issues with O/C detection. A big part of the problem is that CPUs now overclock and underclock themselves during normal operation.

But we are working to getting a more accurate split in the charts. Benchmark itself is OK however.

[Seronx]

All the commendatory I have read is that Ivy Bridge will perform better than Sandy Bridge.

Should be higher turbo boost speeds, architecture improvements, faster larger cache, and much better GPU.

16 EU x 4 32bit ops x 1.15GHz = 73.6 GFlops
vs
12 EU x 4 32bit ops x 1.35GHz = 64.8 GFlops

Isn't a big improvement

i7 2700K = i7 3770K in clock speed and turbo boost

There was architecture improvements none that would speed up passmark that much which has no AVX

The cache will be faster just because of 22nm Tri-gate allowing for more poor with less voltage but they didn't scale up the cache it is still 8MB for the highest end i7 on LGA1155

Clock speed doesn't count for all that much nowadays. Same clock speed doesn't mean same performance.

Clock Speed is a huge factor....

IPC x Clock Rate = IPS(GIPs/MIPs)

Floating Point Operations x Clock Rate = GFlops

They are interchangeable though

IPC for most processors are Floating Point Operations that actually do Integer operations

Clock Rate is a huge factor...

And Intel has found out how to troll your benchmark by releasing a CPU that can't tell if it is overclocked or not...

[David (PassMark)]

Lets wait and see once Ivy Bridge actually comes out. My prediction is that they will be faster than Sandy Bridge in pretty much any benchmark. Feel free to tell how wrong I was in a few months once actual results are available.

While it might be nice to think that Intel designs their CPU with the express aim to fool our overclocking detecting, I don't think we are that important to Intel. For what it is worth we have similar issues with detecting clock speeds and bus speeds on AMD's latest CPUs. Each new CPU family brings a bunch of new (often undocumented) MSRs and different behaviors.

[Seronx]

Lets wait and see once Ivy Bridge actually comes out. My prediction is that they will be faster than Sandy Bridge in pretty much any benchmark. Feel free to tell how wrong I was in a few months once actual results are available.

i7 2600 vs i7 3770

8,970 -> 9,170~9,260

Your prediction is pretty off though

Placing it around where the FX-8150 should be...

You really need to separate 32bit scores from 64bit scores

[David (PassMark)]

Ivy Bridge CPU has the same clocks and the same performance...

Some independent reviews are now out for Ivy Bridge. Like over at Anandtech.com. After running a large number of tests their conclusion was,
"You're looking at a 5 - 15% increase in CPU performance over Sandy Bridge at a similar price point". GPU performance also improved up to 40% and there was a 30W reduction in power usage.

As the price will be the about the same the it would make sense for the OP to wait for Ivy Bridge, if they haven't already purchased.