(2) Xeon x5670 or (1) Xeon E5-1660

[Malabalu]

Having a tough time deciding between these two. Yes, I know they're old, but the performance for the comparative price is too good to pass up. I'll be using Premiere and AE, mostly for 1080p video, but occasionally low bit-rate 4k, and then some minor gaming on the side - think Fallout 4 as for all I need. The system will have 24gb of ram and an RX480 8gb to start out. Nothing incredibly fancy, but I'm getting the full build for under $400, so I don't care too much.

Which would be the better route? I know that having 12 cores over 6 has its benefits, but does the fact that the e5-1660 is a little newer and has a higher clock speed overcome that? Obviously the e5-1660 will be better for gaming, but the bulk of my work will be video editing. Would it still beat out the dual x5670's?

 

[Ecky]

I imagine dual x5670's would be faster for editing. Sandy Bridge was a significant bump in single-threaded performance, but it's not going to make up the difference.

 

[Malabalu]

Any idea what kind of advantage it would have?

If it's about a 5-10% difference between the two in multi-core applications, I'd probably go with the e5-1660 for its single threaded performance. But if it's 10-20% difference in multi-core, I'll go with the x5670's without a doubt.

 

[Ecky]

I expect it'll be a lot more than that in multi-core. Off the top of my head, I'm going to guess SB was a 10-15% bump over Nehalem, and there's a 12.5% clockspeed advantage. So, ~25% better single-threaded performance vs 100% more cores.

 

[Malabalu]

I expect it'll be a lot more than that in multi-core. Off the top of my head, I'm going to guess SB was a 10-15% bump over Nehalem, and there's a 12.5% clockspeed advantage. So, ~25% better single-threaded performance vs 100% more cores.

Thanks for the additional info! This question may be a stretch, but do you know of any sites with hard data I could dig through related to this?

The best I've found so far were these few bits of info:

An After Effects CPU crunch benchmark between the dual 5670's and a W3680 3.3ghz 6 core (I figure the 1660 would be at least 10% up from this, considering the same core count and clock speed); ultimately the W3680 was about 25% slower.

Cinebench 15 scores -
e5-1660 is about 6% lower render score than dual 5670's
but is almost double the 5670's in the single score.

Geekbench 4 - rough averages
3,600 single 16,000 multi for the e5-1660
2,600 single 17,000 multi for the 5670's

Passmark -
e5-1660 = 12,587
dual x5670 = 12,724

Based on all of this, I'm just as conflicted, because what you said makes 100% sense that more cores should greatly outweigh it; but I'm struggling to find decent benchmarks to show it. Most are coming out fairly equal on the multi-core tasks, and significantly better for the 1660 on single core. Or maybe I'm reading the data wrong?

 

[Ecky]

You could try Anandtech's Bench. You may have to compare semi-equivalent products there.

It's also possibly a factor that there is a relevant bottleneck somewhere in the dual 5670 system. Generally multi-CPU doesn't scale as well as multi-core within the same CPU; moving a thread from one chip to the other has far worse latency than moving from core to core within the same chip.

I'd also advise caution about benches for Xeons as I've noticed they don't always include the generation. A 1660 is a very different CPU from a 1660v4, which you probably already know.

Just speculation on my part, really.

 

[Malabalu]

You could try Anandtech's Bench. You may have to compare semi-equivalent products there.

It's also possibly a factor that there is a relevant bottleneck somewhere in the dual 5670 system. Generally multi-CPU doesn't scale as well as multi-core within the same CPU; moving a thread from one chip to the other has far worse latency than moving from core to core within the same chip.

I'd also advise caution about benches for Xeons as I've noticed they don't always include the generation. A 1660 is a very different CPU from a 1660v4, which you probably already know.

Just speculation on my part, really.

Thanks! Gave them a try, they showed substantial performance gains on the i7 equivalents between the two, but unfortunately, no dual cpu comparison.

And thanks for the bench advice, I definitely noticed that, and made sure to only go off of benches that included the generation and clock speed to verify.

But I think you might've hit the nail on the head with there being a bottleneck elsewhere in the system. I'm thinking the different socket and board may be playing a significant role in it; it's looking like the dual 5670's are definitely a hair above the 1660 in all multi-core tests, but the single core for 1660 being so much higher might edge me over to it. Unless I'm able to find some solid results showing a 20%+ benefit in multi-core applications, I think the 1660 might be the best blend of editing Premiere/ AE and being able to play most modern games at decent fps.