Upgrading home system - Core i5 or i7?

Guy S. wrote on 7/5/2012, 2:51 PM
My home editing system is built around a first generation Core i5 750 and it's been OK but I would like to move to a faster CPU to improve the frame rate and quality of previews while editing.

The main difference between Core i5 and i7 appears to be hyperthreading and I've seen CPU charts that show this makes a difference in apps that can take advantage of HT. BUT... My work system is a 6-core Xeon-based HP workstation (pre-Sandy Bridge) and I've seen no noticeable difference in timeline performance with hyperthreading turned ON or OFF.

So... I've looked at LGA 2011 systems but they are Sandy Bridge parts that support pcie 2.0 whereas the newer Ivy Bridge CPUs support pcie 3.0. The newer Ivy Bridge parts also have a higher memory bandwidth - even though they use 2-channel memory vs. 4-ch for the LGA 2011 systems: 25 GB/S vs. 20GB/s.

My take on all this is that Ivy Bridge parts are likely to yield a better timeline editing experience vs. Sandy Bridge and that the Core i5 3570k is the best bang for the buck. However, I don't mind spending an extra $100 on the Core i7 3770k if it will yield a noticeable improvement my timeline editing experience. I would also consider spending a little more on a LGA 2011 system if the expense were warranted.

I plan to just swap out the MB, CPU, RAM and cooling system and hold off on replacing my 550 ti graphics card until a comparable PCIe 3.0 card is released. I am also considering whether or not to add a small SSHD for swap and temp files.

At home I will mostly be editing interviews shot with a GH1 (AVCHD 720P) or GH2 (1080 24P). I typically use very few effects but I will be using Magic Bullet Looks. This one effect cuts the framerate in half - or more - on my 6-core work computer so I suspect it would be next to unusable on my home system.

Ideas, Comments, & Feedback appreciated!


NicolSD wrote on 7/5/2012, 3:35 PM

The Premiere Pro users have a very, very fancy test file that they run to find out how their systems perform. Just about anything that can influence the results is taken into account: RAM size, disk speed, RAID, etc.


Take a look at it and look for the best Ivy Bridge CPUs and how high they rate. The last time I looked, it was 82nd. Next, take a look at the Sandy Bridge machines. An I7-3930K is in second place, right behind a dual XEON X5680.

I do believe it says something about the platform.
AtomicGreymon wrote on 7/5/2012, 4:07 PM
That's Sandy Bridge-E. No point in comparing that until Ivy Bridge-E shows up. The 3770K and 3930K are also at totally different price points. The only SB-E CPU you might compare to a 2600k or 3770K is the 3820, and the highest spot it occupies on that chart is #164. It's also the only 3820 on that list, so do you think #164 is a fair representation of what the CPU could do with overclocks similar to the higher rated systems, and in combination with superior components and more RAM? Probably not. Similarly, there are only five systems on the chart using 3770K or 3770 CPUs compared to a whopping 179 using an i7-26xx; so again it's not really a balanced representation. How could it be? Ivy Bridge hasn't been out for long enough. There's been a lot more time for overclockers to build the best possible Sandy Bridge machines and overclock the heck out of them.

The highest normal Sandy Bridge machine is #9, an overclocked 2600K at 5.2Ghz. Most reviews of the Ivy Bridge CPUs, while disappointed that they aren't as much of an improvement as everyone hoped they'd be (and not worth the price of upgrading from Sandy Bridge if you already have a SB system), seem to indicate they perform no worse, and perhaps modestly (but probably imperceptibly) better, albeit a bit hotter. So with similarly good components and a high overclock, I really see no reason why an Ivy Bridge CPU couldn't match any of the Sandy Bridge ones on that chart. You'd probably want a slightly better cooling solution, but they could do it.
NicolSD wrote on 7/5/2012, 4:25 PM
The 3930K has not been around for that long either. But right from the start, it busted its way straight to the top. Even an overclocked 3770K will not beat a similarly overclocked 3930K. You are comparing 6 cores/12 threads with 4 cores/8 threads.

Even out of the box, not overclocked, the 3930K gives you much better results in Premiere Pro CS5.5. I haven't found any benchmarks using CS6. All data I have found shows an improvement over the 2600K and 2700K. But not over the 3930K and 3960X.
john_dennis wrote on 7/5/2012, 5:29 PM
You might look at these synthetic benchmarks. In my mind, it is either a 3770 (base or k) or a 3930k even for a home workstation. Though there is a difference in price between the i5 and i7, over a few years it becomes moot if you can pay for it in the first place.

Yesterday, I ordered the parts to build an i7 3770k system even though I had been configuring systems with the 3930k. When it came down to the final decision, the Ivy Bridge is more efficient than the Sandy Bridge E in terms of power and cooling. I came to the conclusion that I can live with the difference in throughput when you consider the difference in power usage as shown in one of the bottom graphs, Load Power Consumption x264 HD 3.0 (203 vs 128 Watts as configured).

In the coming weeks, I'll see just how happy I am with my decision. All the 3930k advocates, don't direct your responses about memory channel bandwidth and PCI Express lanes at me. I recognize there is a place for these attributes, but in my lifetime, I've bought a lot of potential in motherboards that went unused. The most notable example was the top-of-the-line 975 chipset board (ASUS P5W DH Deluxe) that had a very short life in my service.
AtomicGreymon wrote on 7/5/2012, 5:47 PM
Yes, but there's no point comparing Ivy Bridge to Sandy Bridge-E; the two are at different price points, and meant for different types of users (except for the half-hearted i7-3820). I only said that there's no reason a 3770K shouldn't be able to match the performance of the best overclocked 2600K on that chart, given time for overclockers to buy the things and work their magic. When I say "Sandy Bridge", I'm only referring to the consumer version. I'm intentionally excluding LGA2011, unless I've included the "-E".

The OP never explicitly mentioned interest in the 3930K. He said he wouldn't mind spending "a little more" on LGA2011. Neither of the two upper-end LGA2011 CPUs are just a little more. Ignoring the 3960X, which runs $1000 on its own, the cheaper 3930K is currently going for about $570, whereas the two Ivy Bridge CPUs the OP says he's considering cost closer to $235 and $315, respectively. Taking into account the higher price of X79 motherboards, a 3820 might only end up costing a little bit more than Ivy Bridge; but whether the 3820 is anywhere near as worthwhile as its two big brothers is questionable, IMO.

So, which LGA2011 CPU is it you'd be considering, Guy? Based on your post, I'd assumed it was 3820, which led me to favour a high-end Ivy Bridge, instead. If you're willing to shell out $200-300 more for a 3930K, and also a bit more for an X79 motherboard, then yes, I'd agree with NicolSD and say go with LGA2011. If you don't intend to upgrade again at least until Ivy Bridge-E, anyway.
Guy S. wrote on 7/9/2012, 2:57 PM
You guessed correctly, the 3820 is the LGA2011 CPU I'm considering.

Also, I appreciate the link to the PremPro test results, thank you!

WRT the test, I found the results quasi-helpful because the results - even for comparable systems - were all over the map. The lone 3820 did reasonably well and the 3930's and 3770's did very well in most cases. In general, it seemed like the 39xx parts were mostly in the top 200 with scores from 131 to the low 300's.

The lone 3820 was among the top 150, scoring ~220. There were three Core i5's (one of them an OC's IB) within a few points of the 3820, but the SB and IB i5's were mostly grouped in the 400 - 500 range. My i5 was, in most cases, significantly lower.

But here's the interesting part: Once you get past the top 200 systems you just see this huge diversity where highly spec'd Core i7's (OC'd, lots of RAM and a SSD) scored in the same range as the majority of the Core i5's.

The lone 3820 scored a little lower than a couple of the IB 3770 systems, and higher than several others. If I understand it correctly, the point of quad channel memory is faster data transfer. But if an IB CPU with dual-channel memory can transfer 5GB/s more data, then in theory the IB-based system would be a better choice.

I'm also a little leery of the 3820 even though it was a top performer because only one was listed, so there's no way to develop an average result.

Finally, I'm wondering if IB's support for PCIe 3.0 would help me realize the performance potential of the new Keppler-based GPU's - because I sure didn't experience any performance increase on my PCIe 2.0 system.

If you see any flaws in my understanding or reasoning, please point them out!
videoITguy wrote on 7/9/2012, 4:11 PM
To Guy S:
I think you have greatly simplified the results of the Premiere Users Test matrix, but you have surmised greatly the truth in this business- ie, whatever you configure you are likely to end up in the middle of the curve as the rest of us do.

You may want to overreason your upgrade - I wouldn't suggest it...and be careful your choosing to lock yourself down hard with the video card if you are trying to install VegasPro11. Again I wouldn't encourage that approach at this time.
john_dennis wrote on 7/9/2012, 5:12 PM
"...realize the performance potential of the new Keppler-based GPU's"

You may have to wait to see the performance of the 6x0 cards. See this

Guy S. wrote on 7/10/2012, 12:24 PM
I appreciate the feedback, thanks!