TomsCore i7-4960X Preview: Ivy Bridge-E, Benchmarked

[csbin]

http://www.tomshardware.com/reviews/core-i7-4960x-ivy-bridge-e-benchmark,3557.html


CPU performance is improved so little, X79 chipset and nothing changes, upgrades, motivation and desire is too small.:hmm:

 

[ShadowVVL]

yeah no point in upgrading if you already have a 3930X-3970X.

perf/watt looks good.

 

[frozentundra123456]

yeah no point in upgrading if you already have a 3930X-3970X.

perf/watt looks good.

Pretty much what was expected. Since sb to ivy only gained a few percent, not sure why anyone would expect bigger gains in the E series. Nice efficiency gains, but not sure anyone cares that much in this segment except for servers.

 

[Shmee]

I wonder how well they OC. Probably not that much better, but we will have to see.

 

[NTMBK]

Nothing on overclocking? That would be the first thing I would test, see if the solder under the lid unleashes IB.

 

[Jacky60]

I wonder why Intel bother with these tiny incremental improvements?
I hope they overclock better but I'm increasingly tempted to stick with my 920 until I die at this rate.

 

[ShintaiDK]

I wonder why Intel bother with these tiny incremental improvements?
I hope they overclock better but I'm increasingly tempted to stick with my 920 until I die at this rate.

Because its targetted at the workstation and server segment with completely different demands than an "enthutiast".

 

[Sweepr]

Much more power efficient than SB-E, solder instead of TIM, OC results cant come soon enough. They could definitely offer 6C/12T mainstream (LGA1155/1150) chips by now if they wanted to.

 

[BrightCandle]

I care about the overclocking results. Although this time I will pay a lot of attention to whether the reviewers have an "engineering" chip or a retail one. The shenanigans played last time with SB-E were downright dishonest for overclocking.

If IB is anything to go on however I doubt we are going to see anything useful overclocking wise. Even those that delidded and such didn't get exceed SB overclocking.

 

[Idontcare]

Disclaimer: The following preview is based on an early engineering sample Core i7-4960X.

Uh, thanks but no thanks. There is a reason "early engineering samples" are called that and not called "retail sample".

 

[Arachnotronic]

Pretty much what was expected. Since sb to ivy only gained a few percent, not sure why anyone would expect bigger gains in the E series. Nice efficiency gains, but not sure anyone cares that much in this segment except for servers.

These chips are aimed at servers where perf/watt is a critical component of TCO. Also servers get the 12 core versions

 

[UaVaj]

if bf4 (or any other titles) is all that ultra, if bf4 can make use of all 6 core, if ivb-e is solder.

will upgrade ivb-e.

if none of the above holds true. keeping ivb.

 

[exar333]

30% more efficient is pretty awesome. That said, I want to see how much this impacts power/temps when OCing at speeds >4.2ghz.

 

[Rakehellion]

yeah no point in upgrading if you already have a 3930X-3970X.

Or 3930K.

 

[Shmee]

Wait till haswell E maybe I guess

 

[Makaveli]

Because its targetted at the workstation and server segment with completely different demands than an "enthutiast".

Bingo!

people seem to forget the highend desktop is just left over server parts!

They don't really care all that much about the enthutiast market every once in a while they toss us a bone to chew on. The money is in the enterprise market which is what they care about most.

These numbers are pretty much what I expected.

I'm waiting for Haswell E and Sata Express before I look at upgrading anything.

 

[parvadomus]

This is what AMD needs to be able to compete, a shrink to 22nm and a more efficient arch (SR), would put an 8 core part on par with 4770K (according to that power consumption charts) in terms of efficiency. They need TSMC or GF to start to compete with Intel though.

 

[JimmiG]

Ivy Bridge was never meant to improve performance. It's basically a die-shrink of Sandy Bridge, but because of TIM instead of solder we never saw the benefit you normally associate with a die shrink (lower temps, higher OC). If IB-E uses solder that might finally allow the 22nm process to shrine, allowing for higher overclocks than Sandy Bridge. Other than that, there are absolutely no interesting features or enhancements to make IB-E interesting.

The IPC improvements will come with Haswell-E. If it also uses solder, combined with a very mature 22nm process, it will be somewhat interesting. Effortless 5+ GHz overclocks combined with the IPC of Haswell would finally move CPU performance forward and give owners of the 2500K a reason to upgrade.

It's pretty sad that Intel's CPUs are so uninteresting these days that people care more about what Intel puts between the die and the heat spreader than about the actual multi-billion transistor chip itself...

 

[Roadrunners]

It's pretty sad that Intel's CPUs are so uninteresting these days that people care more about what Intel puts between the die and the heat spreader than about the actual multi-billion transistor chip itself...

I am sorry to say that I agree.

 

[cytg111]

as we all predicted 6-12 months ago, IVB-E is a close to non-relevant part. Why even put it out, use the $$ on validation, production and marketing when SB-E is allready doing the same job? As it does not make much sense I actually expected an 8 core part even up untill now.. for the puzzle to make sense and all. This is not Intel-like.

"I'm waiting for Haswell E and Sata Express before I look at upgrading anything. "
- If there's an 8 core part in the E, im on that ship too.

 

[NTMBK]

I wonder whether Intel are making a specific 6-core die this time around?

On Sandy Bridge E, there were two dies- the 4 core, and 8 core (which was used for 6 core parts too). It seems mad that Intel would ship a massive die on their latest manufacturing node with half of the die disabled just to make the enthusiast parts.

 

[USER8000]

I wonder whether Intel are making a specific 6-core die this time around?

On Sandy Bridge E, there were two dies- the 4 core, and 8 core (which was used for 6 core parts too). It seems mad that Intel would ship a massive die on their latest manufacturing node with half of the die disabled just to make the enthusiast parts.

Perhaps they are going to 6 core and 12 core this time around??

Its mentioned here:

http://www.guru3d.com/news_story/8_core_sandybridge_e_and_6_12_core_ivybridge_e_in_2013.html

 

[cytg111]

and perhaps they've got production to spare? Even if it sounds nuts to disable half the cores, if the fab would otherwise idle it may still pay dividends to do just that?

 

[Kippa]

Just a noob question. On retail Xeons do you have to buy a cpu fan as an extra or does one come with the cpu? Thinking of getting on Ivy Xeon.

 

[Khato]

30% more efficient is pretty awesome. That said, I want to see how much this impacts power/temps when OCing at speeds >4.2ghz.

Indeed, the primary point of interest with IVB-E is the efficiency. Which is what makes Tom's power numbers rather unfortunate as they don't provide CPU load power numbers, only the graph comparison of the entire benchmark suite run and an average. Now the averages allude to an interesting point - in the benchmarks we see basically the same percentage performance increase going from the i7-3970x to the i7-4960x as we do going from the i7-2700k to the i7-3770k (as should more or less be expected given that all parts have basically the same frequencies) however the i7-3770k uses 85% the average power of the i7-2700k while the i7-4960x uses 80% the average power of the i7-3970x.

As that piqued my curiosity I decided to see if more information could be extracted from the graph - there is. First, we can confirm from the tail end that idle power usage is comparable between the like platforms, so that shouldn't play too much of a part. Second, there are two portions of the graph that look to be pure CPU load, with the one before the major spikes (I'm assuming those are due to GPU loads) offering a relatively clear and stable view of load power consumption. From there it's a simple matter of measuring pixels to derive approximate load power numbers for each of the processors of interest - 109W for the i7-3770k, 137W for the i7-2700k, 161W for the i7-4960x, and 225W for the i7-3970x.

Now for the interesting part - SNB -> IVB dropped load power consumption by roughly 28W (extrapolate to 40.5W if 6 core) while SNB-E -> IVB-E dropped it by 64W. Subtracting idle power usage from each the i7-3770k is using roughly 66% the load power of the i7-2700k while the i7-4960x is using 60% the load power of the i7-3970x. It's more of a difference than I was expecting... though it's still on the edge of what can be explained by high sample variation. aka, probably just wishful thinking that it might be due to a few tweaks to the cores used in IVB-E.