Thursday, April 25th 2013
Core i7-4960X "Ivy Bridge-E" Roughly 10% Faster than i7-3970X: Early Tests
PC enthusiast "Toppc" with the Coolaler.com, with access to a Core i7 "Ivy Bridge-E" sample clocked to match specifications of the Core i7-4960X, wasted no time in comparing the chip to a Core i7-3970X "Sandy Bridge-E." The two chips share a common socket LGA2011 design, and run on motherboards with Intel X79 Express chipset. An MSI X79A-GD45 Plus, with V17.1 BIOS was used to run the two chips. Among the tests Toppc put the chip through, are overclocker favorites SuperPi mod 1.6, CPU Mark '99, WPrime 1.63, Cinebench 11.5, 3DMark Vantage (CPU score), and 3DMark 06 (CPU score).
The Ivy Bridge-E chip outperformed its predecessor by roughly 5-10 percent across the board. In Cinebench, the i7-4960X scored 10.94 points in comparison to the i7-3970X' 10.16; SuperPi 32M was crunched by the i7-4960X in 9m 22.6s compared to the 9m 55.4s of the i7-3970X; CPU Mark scores between the two are 561 vs. 533, respectively; 3DMark Vantage CPU score being 38,644 points vs. 35,804, respectively; and 3DMark 06 scores 8,586 points vs. 8,099 points, respectively. In WPrime, the i7-4960X crunched 32M in 4.601s, compared to its predecessor's 5.01s. Below are the test screenshots, please note that they're high-resolution images, so please open each in a new tab.
Cinebench 11.5SuperPi and CPU Mark3DMark Vantage CPU score3DMark 06 CPU score and WPrime 1.63
Source:
Coolaler.com
The Ivy Bridge-E chip outperformed its predecessor by roughly 5-10 percent across the board. In Cinebench, the i7-4960X scored 10.94 points in comparison to the i7-3970X' 10.16; SuperPi 32M was crunched by the i7-4960X in 9m 22.6s compared to the 9m 55.4s of the i7-3970X; CPU Mark scores between the two are 561 vs. 533, respectively; 3DMark Vantage CPU score being 38,644 points vs. 35,804, respectively; and 3DMark 06 scores 8,586 points vs. 8,099 points, respectively. In WPrime, the i7-4960X crunched 32M in 4.601s, compared to its predecessor's 5.01s. Below are the test screenshots, please note that they're high-resolution images, so please open each in a new tab.
Cinebench 11.5SuperPi and CPU Mark3DMark Vantage CPU score3DMark 06 CPU score and WPrime 1.63
122 Comments on Core i7-4960X "Ivy Bridge-E" Roughly 10% Faster than i7-3970X: Early Tests
So, as far as I'm concerned, I still don't see a selection of GPU based benchmarks being overly relevant when a third-party GPU is being used to render the result.
EDIT: Digging around on the sitewhere the graphics comparison cropped up (note to Ikaruga: a link with the graph you posted would've been good), also presents some CPU benchmarks which are probably more relevant- although I'm wondering how mature the board BIOS is given the memory bandwidth numbers:
If there was a way for INTEL to magically deliver a 30% improvement over Sandy-Bridge-E, that's what we would have today. Sadly this is not possible to achieve for any outfit; ARM, AMD, NVIDIA, TI, SAMSUNG etc. Nobody could deliver these kinds of gains within the constraints that INTEL has. The advancements are incremental and at no point did INTEL ever promise anyone massive gains when moving from one generation to another. That assertion came from us and it has always been incorrect. When investing billions of dollars into R&D and thousands of man hours, lazy isn't the word I'd use to describe the efforts of INTEL, AMD or any outfit for that matter. That word would be better reserved for my own profound and limited understanding of the technology and it's evolution. I'm an enthusiast because of my appreciation for the technology not my expectations of it. :)
We also tend to forget that, as far as expertise at processors and process manufacturing is concerned, no other company on this planet has dedicated as many resources to this as INTEL. I doubt if any one of us can find a single company that has commercial products on a fin-fet process at this node let alone with such massive cores containing so many logic gates. Whatever system you may own that you find to be "good enough" for all your needs, it's precursors had similar miniscule advances between each generation, which culminated in your "good enough" system.
As I am no engineer I find it hard to be disappointed by Ivy-Bridge-E, because whatever disappointment there may be, it stems from my own ignorant expectations rather than INTEL's inability to produce a compelling CPU.
I will be buying the 4960X, with the relevant motherboards when it's released.
Sorry, but that was a utter load of rubbish to me and I can't find any another way to say it (other than not saying it at all, but since everyone around me keeps saying I should speak out more, I'm starting to not hold back any more).
This kind of thinking borders on Stockholm syndrome... but in regards to consumers as opposed to a full-on assailant. And it's kind of sad if the trend will be shifting that way... As long as it's consumerism we're talking about, we, the customers, the paying customers, should have the first say in anything, not the other way around. I'm not buying into Intel's shitty performance increments, and I'm certainly not defending their name.
Intel's ***** performance increments? What has lead you to believe they could or were supposed to be better? What information do you have, that nobody else apparently has, leads you to make this statement with such zeal? In your expertise, what could INTEL have done from a technical POV to ensure the double digit percentage gains you so desire?
"You, the paying customers", aren't entitled to anything. INTEL above all else is a self serving business, they owe you nothing other than what they have provided to you when you purchased your most recent CPU form them. You are not entitled to anything past that, hence there's no discourse needed between INTEL and you on what they release and how they go about it.
I'm not an engineer or anything of the kind (and I'll certainly not attempt to come off as one from behind my keyboard), but from what I gather it is truly difficult to design a CPU given many of the limitations in TDP and such. Your annoyance stems from your ignorance, not from an understanding of what it is INTEL was trying to achieve here.
The question is where have you been done wrong by Ivy-Bridge-E for the CPU to warrant such disdain from you? I re-iterate, you're disappointed entirely because of your expectations. When did INTEL promise you or anyone else massive performance gains with Ivy-Bridge-E?
But in a sense, I guess what you said is right, it's not Intel's fault, it's the consumer's fault for buying into nonsense every damn tick and tock they shelve out, and to an extent, to their competition. But I won't go into how wrong thinking (external) competition is the only thing pushing anything forward in this world. Very few people/organizations strive to better themselves because they can these days and for the most part, just stagnate slightly above the lowest common denominator. I challenge you to find me a nVidia GPU-based laptop with a AMD processor. Even more so, specifically with the setup I listed.
There might be a batch of laptops with those GPU/GPUs w/ something like i7-3630QMs, but seeing as ix-4xxx will be priced pretty much the same, I see no point in paying for older hardware, no matter if there are any real improvements or not.
As it stands Intel is subpar and that is why myself and others haven't upgraded as often the last few years.
More often than not if you're getting skt2011, you want either multi-threaded performance, support for more memory, the ability to have VT-d and overclock at the same time, or to have a lot of expansion opportunities with the 40 PCI-E lanes on the CPU, which is certainly no slouch. Clocks will only go so high but clocks alone is not a reason to go skt2011 and anyone who thinks that has no idea what they're doing.
Dunno you guys but I'd drool over a dual 3960X/3930K setup.
All in all, most processors do everything it needs to now a days and more often than not overclocking is merely for fun, but the fact still stands that performance along is not a reason to invest in skt2011. Also the changes from skt1366 and skt2011 don't highlight performance. The big change was eliminating the IOH and putting PCI-E on the CPU (40 whole lanes of it,) and practically making QPI useless on most X79 boards.
Simplifying the motherboard design and adding more functionality to the CPU would be considered an improvement, despite not being directly performance related, wouldn't you say? 95% of CPU owners don't own a skt2011 or skt1366 rig either.
Bloomfield was launched in Q4 2008, the 3820 was launched in Q1 2012, over 3 years later.
Okay Tatty, let's forget performance and age of the CPU for a moment.
skt1366 has an IOH using QPI for PCI-E and DMI, skt2011 does not, it's all built into the CPU.
skt1366 has a triple channel memory controller, skt2011 has a quad.
skt1366 used to keep the uncore and core frequency segregated so the uncore can run slower than the core, skt2011 does not do this, the uncore is directly tied to the core frequency iirc.
skt1366 quad cores had 8Mb of L3, skt2011 so far has had at least 10Mb for a quad.
This all becomes moot though if you have something like a 970 or a 980, but as far as a quad-core is concerned, the 3820 and skt2011 is a contender and I'm not sure how you can say that it isn't. I'm not saying that every 930 and 920 owner should go out and invest in a 3820 or better though. I'm just saying that significant changes have been made to the platform since skt1366, regardless of weather or not they addressed performance or something else.
I am not suggesting anyone is getting ripped off or that Intel are very naughty boys, simply that some statements can be misleading, unless of course 8% across one generation, or 22% across 3 generations really is HUGE :D Big and huge are somewhat subjective but my interpretation is perhaps skewed, so I suppose my point is....... that additional time you refer to is not providing enough improvements IMO.
So I think Intel has done a good amount in 3 years. Could they have done more? Absolutely, I won't deny that they could have done more, but would it be the same quality as revisions in the past if they used the same amount of time? I doubt it. Less care would be put into each change so I have a strange feeling that it will negatively impact QA.
While raw IPC is hard to improve I think they are just sandbagging on the core count, I mean bloody Westmere-EX (LGA 1567) had 10 cores.
What on earth is blocking them on making available a bloody 12core CPU for socket 2011 other than themselves?
Just because you can cram 10 or 12 cores on to a CPU doesn't mean you can get them to run fast and even if you can, power consumption and heat dissipation is a huge issue.