Intel Core i7-3960X Review: Sandy Bridge-E And X79 Express

[Crashman]

[citation][nom]Raidur[/nom]Would be nice if for once they did SLI against all the other CPUs.[/citation]SLI has little CPU overhead, CrossFire tests are far more CPU-bottlenecked. That's why AMD graphics cards work so much better when paired with Intel processors.

 

[monsta]

What an underwhelming outcomet for these new chips , I'm going to wait for Ivy to come out , thanks for the great review

 

[Novuake]

SLI has little CPU overhead, CrossFire tests are far more CPU-bottlenecked. That's why AMD graphics cards work so much better when paired with Intel processors.

So what your saying is that a CrossFire setup is more stressing on the CPU to empliment than SLI?

Please explain? With links if you don't mind...

 

[Crashman]

[citation][nom]Novuake[/nom]So what your saying is that a CrossFire setup is more stressing on the CPU to empliment than SLI?Please explain? With links if you don't mind...[/citation]No links, I said it from memory I really don't have time to do your research. In general I've found CPU scaling to impact AMD graphics more than Nvidia. I did a lot of experimentation before settling on a specific platform for the PCIe scaling articles.

I probably don't need to say more because the forums are full of members who can recall articles well enough to provide links without further research.

 

[gio2vanni86]

Great article, hope to hear more about these and Ivy bridge as the year ends and 2012 begins. This is my one year to build a new rig, thus far the 2500k and 2700k still look like the processors i might end up buying for a Z68 platform. Unless the benchmarks change in the gaming realm i see no reason to pay 500+ for any of those processors. I'm sure my Intel Core 2 Extreme QX6850 is old and possibly holding back some potential in BF3 multiplayer, as far as i see my frames are pretty solid at Ultra Res 1680x1050. As stated before no rush and looking forward to 2012 and more results of these and ivy OC'd in the future which might sway me to buying them.

 

[Guest]

Crashman: So you want to make claims that you won't back up, "from memory", but you don't want to "do Novuake's research"? Doesn't Tom's make their money by doing other's research?

I'm not sure why you're correlating the 2nd half of my post to Bulldozer support, because it had nothing to do with Bulldozer, hence it got it's own paragraph. The first paragraph is irrefutably true, however, unless you are now saying that your numbers are a complete sham, or that they "don't tell the real story".

 

[Crashman]

[citation][nom]2nd_timer[/nom]Crashman: So you want to make claims that you won't back up, "from memory", but you don't want to "do Novuake's research"? Doesn't Tom's make their money by doing other's research?[/citation]I don't make money by doing favors, which is what I did in my reply to Raidur. It was a tidbit, it was free, and it's not being used to prove a point. You're welcome.[citation][nom]2nd_timer[/nom]I'm not sure why you're correlating the 2nd half of my post to Bulldozer support, because it had nothing to do with Bulldozer, hence it got it's own paragraph. The first paragraph is irrefutably true, however, unless you are now saying that your numbers are a complete sham, or that they "don't tell the real story".[/citation]Your first paragraph appeared to intentionally ignore "option three". It stated that BD is a better value than "SB-E", which is true. However, it neglected that the article pointed towards SB being a better value than either BD or SB-E. Now you can read the BD review to find out "the real story", that is, unless your misdirection was intentional. And no, I won't link that either, it's in the CPU's section and you'll find it in 30 seconds. Again, you're welcome!

 

[Guest]

That's seriously is a nice processor!

 

[dragonsqrrl]

[citation][nom]2nd_timer[/nom]Doesn't Tom's make their money by doing other's research?I'm not sure why you're correlating the 2nd half of my post to Bulldozer support, because it had nothing to do with Bulldozer, hence it got it's own paragraph. The first paragraph is irrefutably true, however, unless you are now saying that your numbers are a complete sham, or that they "don't tell the real story".[/citation]
Dude, stop trolling. Your Bulldozer argument is heading down a dead end... better luck next time.

 

[bernardv]

I'd call this "irrelevantly fast", since a vast majority of people need only a fraction of that CPU's computing power.

 

[wiyosaya]

[citation][nom]aldaia[/nom]CPU value: $300Bragging rights: $700[/citation]
Fool factor: $1,000

 

[airborne11b]

Funny, in gaming there are still no improvements over an X58 system. My 2.5 year old i7 920 @4.3ghz is still dominating games, even with tri-sli GPUs (no bottleneck).

Still waiting for a worthy CPU to upgrade too.

Maybe the next one.

 

[Guest]

dragongrrl; if you take offense to supposed facts being regurgitated from an article, learn that it shouldn't

Let's have a little AMD vs. Intel showdown on the high end, taking value out of the equation altogether:

The complete list of what SB-E has to do to beat Bulldozer by a mere 33%:

SB-E is 33% bigger than Bulldozer, but has roughly the same number of transistors
SB-E only comes in a harvested CPU, presumably because it's too big and unmanufacturable
SB-E needs 50% more threads to achieve 33% more performance

From this, we can infer that GloFlo has a vastly superior 32nm process. If Intel is lucky, they may be able to achieve a similar density with 22nm, but the next gen 28nm Bulldozer CPUs will be right back to a density lead again. This, combined with an apparent inability to produce any fully functioning 8 core parts would imply that AMD/GloFlo have superior process technology.

The fact that Intel needs 12 threads to beat AMD's 8 threads means that AMD is faster per-thread. If AMD put out an equal sized chip, it would also be 12 threads, but it would outperform Intel by nearly 20%.

Or were these talking points only meant to be applied from Sandy Bridge to Bulldozer, but not going the other way? Aren't double standards fun?

 

[Noworldorder]

Not at THAT price....

 

[eddieroolz]

Not much of an improvement really, but hopeful for Ivy Bridge.

 

[dragonsqrrl]

[citation][nom]3rd_timer[/nom]Let's have a little AMD vs. Intel showdown on the high end, taking value out of the equation altogether:[/citation]
So you're trying to argue that the Sandy Bridge architecture is less efficient then Bulldozer's?... again, I'm afriad you're fighting an uphill battle.

http://www.guru3d.com/news/former-amd-engineer-talks-about-bulldozer-fiasco/

And just to preface with a rebuttal of pretty much the only point you make in your argument, die size in relation to transistor count is completely unrelated to the efficiency of an architecture, or the superiority of one foundry's manufacturing process over another.
[citation][nom]3rd_timer[/nom]The complete list of what SB-E has to do to beat Bulldozer by a mere 33%:SB-E is 33% bigger than Bulldozer, but has roughly the same number of transistors[/citation]

And you're point being? The size of an individual transistor as well as their relative density in relation to each other have nothing to do with the efficiency of an architecture, or its performance. I think any Bulldozer review you read would support this conclusion, but especially this one:

http://www.tomshardware.com/reviews/fx-power-consumption-efficiency,3060.html

[citation][nom]3rd_timer[/nom]SB-E only comes in a harvested CPU, presumably because it's too big and unmanufacturable[/citation]

No... I can assure you, SB-E is perfectly capable of having all 8 cores enabled on Intel's current 32nm process.

http://www.cpu-world.com/news_2011/2011082601_Intel_to_launch_Xeon_E5-2600_series_in_Q4_2011.html

[citation][nom]3rd_timer[/nom]SB-E needs 50% more threads to achieve 33% more performanceFrom this, we can infer that GloFlo has a vastly superior 32nm process. If Intel is lucky, they may be able to achieve a similar density with 22nm, but the next gen 28nm Bulldozer CPUs will be right back to a density lead again. This, combined with an apparent inability to produce any fully functioning 8 core parts would imply that AMD/GloFlo have superior process technology.[/citation]

Honestly, what the hell are you talking about?

You sound like saturnus, "And it's also obvious that Global Foundries 32nm SOI process is far more advanced than Intels 32nm bulk process. 2B transistors in 315mm2 in the FX vs. 1B transistors in 216mm2 in the i7, that's a 37% advantage in transistor packing density."

This is completely false... where are you guys pulling this information from? The size and density of transistors on a die has far more to do with the type of transistor used, guided by the design decisions of engineers for a given architecture and process. It has nothing to do with how successful a foundry's manufacturing process is or isn't. The irony in your argument is that GF is actually having issues with yields on its 32nm process, not Intel.
[citation][nom]3rd_timer[/nom]The fact that Intel needs 12 threads to beat AMD's 8 threads means that AMD is faster per-thread. If AMD put out an equal sized chip, it would also be 12 threads, but it would outperform Intel by nearly 20%.Or were these talking points only meant to be applied from Sandy Bridge to Bulldozer, but not going the other way? Aren't double standards fun?[/citation]
Dude... stop. You're argument is a ridiculous mix of oversimplifications and misunderstandings. There is no double standard. The way the two architectures handle threading are not directly comparable, your inferences are irrelevant and are again based on oversimplifications and misunderstandings.

You also fail to make the conceptual leap between this portion of your argument and your previous criticism of SB-E. The i7-3960X does have 25% of its cores and L3 cache fused off, and yet it still outperforms the FX8150 by a sizable margin in heavily threaded workloads (a supposed strength of the Bulldozer architecture), all while having a lower stock clock, providing more overclocking headroom, and consuming less power.

And like a true troll create a new account once again, if you decide to respond.

 

[hyrule571]

Pentium 4 Extreme Edition, Socket 478 FTW! Still floating on my 1 Gallatin core =]...

Thinking about building a system off of the x79...

 

[userngalan]

expensive

 

[DjEaZy]

... another socket... minimal gains... expensive as S#!T!!! Do intel know the state of global economy? how they gonna make money out of this?

 

[Guest]

It would be interesting to see how well this cpu performs compared to the AMD Interlagos Opteron I bet the Intel wins with 6 threads vs 16 threads from AMD even in multithreaded benchmarks.

 

[Reynod]

So its a crippled Xeon then ... hmmm ... and over a grand a pop too?

I'll check in the forums to see how many rich suckers buy one of these and we can all have a giggle.

Those of you with 2600k's must have huge smiles on your faces ... still!!

After the BD release those smiles must be leaving lines by now ...

There is nothing her to warrant checking my bank account ... moving on.

Nice article Chris!

 

[occupant]

I am willing to bet the 8 core Sandy-Bridge-E will be the 3990X and it'll be out in a quarter or two, AFTER the Xeon introduction. Stick around, things will get interesting. The 3990X will be the $1000 gottaupgradethegibson chip and the 3960 will drop to $550ish and the 3930 will drop to $350ish and THEN we will see a nice round of "answers" from AMD. In the meantime, I'm holding onto my II X4 630, i7-920, and i7-950 desktop systems. More than adequate even after all this time. Sad to see the i7-920 at the BOTTOM of any chart, but I know it's light-years ahead of all the i3s, i5s, Phenom this, A-Series that, and darn near everything else out there under $200.

 

[shin0bi272]

You guys should really look into using dBpoweramp for your audio conversion software. It can rip in almost any format AND utilizes all cores of the processor. So you'd actually see a difference in ripping times with a 6core cpu vs a 4 core cpu and if either of those had HT you'd see a difference there too.

 

[blazorthon]

Everyone is bashing Bulldozer but no one is putting it in a situation where it even works right. The thread scheduling problem is more or less fixed in windows server 2008r2 so if you want to test bulldozer that's the way to go. By the way, the 16 core Interlagos CPUs doubled the performance of the old 12 core Magny-Cours so they aren't worthless, quite the opposite. Admittedly the Magny-Cours were under-performers compared to current Xeons, but this is a right step forward in the enterprise market for AMD.

I admit AMD made a crap move releasing Bulldozer before a desktop OS can utilize it properly and should have done a Phenom II die shrink, then released Bulldozer on a 22nm process after windows 8 (if it works right, I've read conflicting reports). That would have helped the ridiculous heating issues (slightly) and fixed compatibility problems Bulldozer faces now.

Moving on, SB-E so far seems like a waste of money compared to the current SB quad core CPUs. similar single threaded and slightly threaded performance for triple the price means you need to be using heavily threaded programs or a lot of single/lightly threaded programs at once for it to be worth your time. That doesn't make it worth your money at $1000 when there are cheaper alternatives without sacrificing performance with dual CPU boards. If you need that much performance you could grab two 8 core Bulldozer Opterons (cheap ones are well under $500). I'm not saying I wouldn't love to have a 3960x, just that I don't want to pay for it when it isn't the only option or best in it's price range.

Honestly, it isn't extremely overpriced for what it is, just somewhat. If it was an 8 core or Intel had improved single/slightly threaded performance too then it would be a decent, reasonable price. I want to see if the 3930k is worth it's $600 and I bet it either is worth it or much closer to being worth it than the 3960x is to being worth $1000. Unless money really is no object this CPU is too overpriced for it's performance.

And for the comment by luc vr above me, if you disable hyperthreading on the 3960x to get 6 threads (doesn't make sense, you are doing a highly threaded comparison) and then compare it against a 16 core Interlagos you've just wasted your time. It doesn't matter which one would be faster at that point and they probably wouldn't be too far off from each other (my money is on the Opteron because even if it fails it probably won't be by much and you can get two such Opterons for a little more than $1000, the 3960x with HTT won't come close to two). The extreme editions from Intel have often trumped some/most Xeons and Opterons in performance.

Per core/thread speed is NOT the point of most 8-16 core CPUs even with Xeons.

As for 3rd_timer, you have failed because the extra 6 threads on an Intel HTT CPU aren't equal to a core. last I heard they get you about a 30% increase in performance.

 

[zulutech]

[citation][nom]srgess[/nom]I heard when windows 8 come out this processor will get more benefit lawl[/citation]
Man, theres no l at the end of law.