Gulftown vs. Sandy Bridge - Benchmark Comparison 

It's here: Sandy Bridge vs. Ivy Bridge!

Instead of the usual "review" style article, today we're going to look at "Gulftown vs. Sandy Bridge" from one particular perspective, and one that I think isn't properly represented on most sites, and that's "multi-tasking". 

Thus, this article will encompass the complete "reviews" of both the Intel Core i7 970, and the Intel Core i5 2500K. The 970 will be tested at 4.2GHz, and the 2500K at stock clocks? Why? A - Because I can, but more importantly B - because it shows that NO amount of overclocking the six core will create an advantage where there would not otherwise be one due to having more cores. That's important. 

Multi-tasking

What does multi-tasking mean in this context? I don't mean the architectural definition so much as the "workflow" definition. 

We didn't really start using the term multi-tasking or even "multi-threading" until the days of the PPC 604e processors and the advent of MacOS X. Lion just came out, and it's the seventh major revision of what is essentially an 11 year old OS. In 2000, when all of that was new stuff, I was a Mac user, and Macs were in fact better all around for workflow if you did a lot of publishing, video, etc. on your computer. Today, Apple struggles to keep up with gimmicks, but that's not the point of this story. I'm talking about how we as the user use a computer. 

Ask yourself: How many things do I do at once on my computer?

While you are reading this article, is your computer doing anything else? Any major background tasks, OTHER than antivirus? Are you downloading stuff? Ok, what about a video? Are you watching YouTube or playing local video files? Sure, you can do all of that on two cores, but how many of you are doing the following simultaneously: Video recording, video encoding, gaming, file transfers, downloads, have Photoshop, Premiere, After Effects, two browsers, 26 tabs, and a few minor background tasks going? All at once? 

Show of hands? Couple of you? 

Ok, how frustrating is it that you can't do these things smoothly on . . . ONLY . . . four cores? It's almost impossible, even at 4.2GHz and above. I'm going to get into this later, but really, no matter how FAST each core is, with our current architectures, there is only so much that can go on at once, before the system screams "HEY!!! Enough! I'm a bit overwhelmed". 

Why is that? A lot of factors: The memory controller, the host controller, the CPU architecture, etc. Mostly, and for the purposes of this comparison, we're going to focus on number of cores. This test takes memory slightly out of the equation by using exactly the same sticks, thus confining memory issues to the memory controller, not the ram itself. 

With enough cores, we could blow up just about anything . . . 

The Gulftown Platform

You know what, I know I'm a little late to this party. Blame my back. Intel has been good to me, so I'm doing a comprehensive set of benchmarks, and hopefully will be in on the next round of CPU seeds, which will get tested and published quicker than this. Don't blame Intel, it was my bad. I'm doing the best I can here. 

That said, you don't need me to publish a big commentary or explanation of the architecture or features of the platform. You know this stuff already. I'm going to get to the point. We're testing the Intel 970 in this test ( I have also had the 980X in the lab, but the difference in performance at stock clocks is virtually non existent. The main advantage to the X is ease of overclocking, which with the right board is made pretty negligible too )

The Sandy Bridge Platform ( H67 )

Sandy Bridge is another story. The main claim to fame with 2nd Generation Core processors is of course on board GPU functionality. In another article, I'll look at that more closely, even though it performed quite well. 

Today, I'm looking at Sandy Bridge from a professional's point of view. Is it an upgrade from last gen's six core CPU's? For multi-tasking, it can come down to a preference about money, but we'll get into that after the benchmarks. 

The Benchmarks

For this and all future sets of benchmarks, we're going to use a benchmark suite consisting of the following: ( all versions as of testing )

PCmark 7

Settings - Check ALL boxes, "Secondary Storage Suite", set to C drive.

Passmark Performance Test 7 

Settings - "Run All Tests", and "Multitasking", both set to 600 seconds. 

3Dmark 11

Settings - Basic, and Extreme

Heaven DX11 Benchmark 2.5

Settings - 1920x1080, others default

Call of Pripyat Benchmark

Settings - "4x, DX10.1 Style, HDAO, Ultra(CS), DX10.1, tessellation, and shadows”

Furmark 1.9

Settigns - "Benchmark Preset 1080"

Crystal Diskmark 

Settings - Default

Batman Arkham Asylum

Settings - "1920x1080" - vsync to "NO" - MSAA to none or "ATI" - Detail to "Very High" - everything else "Yes" and physics "Off" even if on Nvidia GPU

Just Cause 2

Settings - 1080, everything else to highest settings, all three benchmarks

Half Life 2 "Lost Coast" Benchmark

Settings - 1080, everything max, color correction and vsync off

The Test Platforms:

For the Intel 970, we have the usual ultra high end build, consisting this time of:

Gulftown Test Platform: 

  • CPU - Intel Core i7 970 at 4.2GHz
  • CPU Cooling - Coolit System E.C.O. 240 with Push/Pull Fans
  • GPU – AMD reference 6870 at stock clocks
  • Motherboard - Asus Rampage III Extreme with 1208 bios
  • RAM – G.Skill 6GB kit ( Temporary until a new 12+ kit can be had ) using only 4GB for this test. 
  • PSU - Corsair AX1200 1200w
  • OS - Windows 7 x64 
No doubt about it, this is an ULTRA HIGH end system. Full reviews of key components are forthcoming, but for now, what you need to know is that this system is not a bottleneck for any card/device.

The Test Platform for the Sandy Bridge:

  • CPU - Intel Core i5 2500K at stock clocks
  • GPU – AMD reference 6870 at stock clocks
  • Motherboard - Intel DH67BL microATX
  • RAM – G.Skill 6GB kit ( Temporary until a new 12+ kit can be had ) using only 4GB for this test. 
  • PSU - OCZ OCZ700MXSP 700 watt
  • OS - Windows 7 x64 
Not a bad little build, and with the same exact GPU and RAM, this is a fair test of both CPUs pushed as hard as they can be on their respective boards. I did do the test with the 970 overclocked, but to be honest, what this test is more about is multitasking, and as we're going to see in a moment, for that, all the clocks to be had wouldn't save the 2500K.

The Benchmarks

First, to establish some baseline, we'll look at the default run of: 
 

PCmark 7

An appreciable, but not huge difference. So far, architectural and clock differences don't mean much. What about Passmark? Passmark contains both a default run, similar in scope to PCmark ( I like redundancy in testing, it keeps things honest ), that shows us a similar picture. 

Passmark Performance Test 7

Virtually no difference whatsoever. So what about that "multi-tasking" thing I've been talking so much about? Let's take a look at a special set of benchmarks in Passmark, each of which is run SIMULTANEOUSLY with all of the others:

With benchmarks now running simultaneously, we're seeing a HUGE difference in performance. Prime numbers? Crush. 

Sorting random strings? Route. 

Compression? Compressed. 

Encryption? Embarassing. 

So on, 

and so on, 

Even the memory controller can't keep up, and I was curious about this, so I ran it several times, on different days. Same results. 

Remember, this is the SAME exact ram sticks, so to be trounced like this can only mean that the memory controller on the boards is that different. I'm anticipating new Sandy Bridge boards soon, so we'll reserve judgment on this particular aspect of these tests. 

Whew! So, right there, you can see it. In multi-tasking, Gulftown is still KING of the road. But how much does that matter in every day bench testing? Let's continue before we get to the conclusion. 

3Dmark 11

Remembering that the GPU is identical, this isn't shocking, but it's nice to see "test validity" holding up here. The CPU makes NO impact here. 

Or here. 

On the physics score, it does show that more cores is better though. Does this matter in most games? I can't feel a difference, but that's ok, it just shows how good this little 2500K CPU is. 

Once weighted together, the overall score is odd. I've run it a few times, and there is room for fluctuation, but I publish the numbers that I have most recently run, and I have screen shots to back them up, instead of averaging them. 

The Extreme setting shows pretty much the same story. 

Identical, holding up. 

Same physics difference. 

Once again, on this run, the Sandy came out ahead, but it sometimes comes out behind. There isn't enough difference here to care. 

Heaven DX11

Just to be sure . . . but no difference at all. 

Same here. 

Furmark

Furmark is a stress test for GPUs, but I like to have it in each bench run because if something is not stable in the entire build, we'll find out here. 

No difference at all. 

Batman Arkham Asylum

I hope that the new Batman will have a benchmark too, because I'll add it to the fray if it does. For now:

I ran this test a number of times, and every time, the Sandy Bridge came out a bit ahead. I can't really figure it out myself, but we can assume that the engine has some optimizations for the new CPU. 

Just Cause 2

It's pretty much what we expect for a gaming benchmark, no difference. 

Half Life 2 - Lost Coast

Just to drive the point home, it's not that the CPUs are much different in that they allow you to do ONE thing so much faster . . . 

Call of Pripyat Benchmark 

But we do see some odd anomalies sometimes. Again, I'll redo all of these benches when I get a more performance based Sandy board in, and then we'll really see where we're at. Maybe I'll have new Sandy CPUs by then too. 

Issues?

None. 

Gulftown: The Advantage of Six Cores

So what did we learn today? 

Six cores are good. Four cores are good too, just not as good for pure multi-tasking. Again, we're talking about multi-tasking not so much from an architectural point of view, which is important, but from a workflow view. Right now, there aren't many benchmarks dedicated to doing repeatable, simultaneous benches, but what we do have shows the story clearly. 

In my work, I use FRAPS to record video of games. I record at 1920x1080, full frame, full RGB, and full 5.1 sound. I do this with NO discernable drop in performance for the game, and I do it WHILE my encoding program is encoding in the background. Can I do this with four cores? No. Just plain no. Six cores is the cost of entry for a workflow like that. 

But what is that worth?

To me, a lot. 

In general terms, the six core performance generally costs around a grand. Right now, it's showing at $550 on Amazon, which is a no brainer. 

The 2500K is showing at $220, or roughly half of that. 

If you require your computer to do more than a few things at once, and do them fast, it's not a matter of GHz, it's a matter of CORES. Most articles and benchmarks focus SOLELY on HOW fast a computer can do EACH task, and many include Photoshop filters, etc. in that equation. I don't care about that so much as I care if it can do those things AND some other task at the same time in a reasonable time frame. If it can, that's worth TWO computers to me, and two high performance computers cost a total of a LOT MORE than the difference of these two CPUs. That's the core of my argument here. 

Also, not having to move entire hard drives or RAIDS from one system to another in order to compress the video is worth a lot. 

In terms of TIME, the Gulftown CPU is worlds faster. Is it worth $550? Every penny. It's worth it at twice the price. 

Sandy Bridge: The Advantage of Cost vs. Performance 

But what if you aren't a hardcore multi-tasker? The choice is ALSO simple, but in the opposite way. I cannot comment on anything not in my lab for testing, so I can't tell you if a 2600K is worth the extra money over a 2500k, or whether it's too close to a 970 to bother with, but at less than half the cost of a 970, if all you do is ONE major thing at once, or you are looking for the best of the best mainstream CPU, you can't go WRONG with the 2500K. 

COMPANY

Intel has been a staunch supporter of mine this entire time. I'm late coming about to these tests because they are some of the most intensive work I've yet done ( and I have a extremely bad back ), and because I was originally hoping for more platforms. 
 
The bottom line is that I have zero issue with Intel, unlike several other vendors, sadly. They are a shining example of how to do PR, and I am looking forward to more CPUs, boards, and SSDs from them as their stock allows. 

Support

Top notch. 
 
I had some issues with a board previously, the details are not important, and Intel was on top of it, that's all I can say. 

Conclusion

For multi-tasking like I do, you want more cores, not more Ghz. Investing in a four core CPU and then a couple of hundred dollars into liquid cooling is silly. Six cores and liquid cooling is beastly. 
 
Eight cores next? Maybe? I don't deal in rumors, but the latest "intel" suggests that "not this round" and that we can expect another set of six cores. Should I be given one of those, you can expect THAT comparison to be interesting! 
 
If you're looking for the best value right now, and you either don't need a dedicated GPU, or you just want the best four core out, get the 2500K. It will go toe to toe with a massively overclocked six core with liquid cooling, and only falls flat when asked to do more than a few things at once. 
 

Get updates on new projects and images

Subscribe to my newsletter for regular updates on new projects and new image sets.