TomsHardware: Core i7-3970X Extreme Review: Can It Stomp An Eight-Core Xeon?

[Idontcare]

After one year of dominating the enthusiast space, Intel's Core i7-3960X is being replaced. The new Core i7-3970X features higher clock rates, but also imposes a 150 W TDP. Just for kicks, we're putting it up against another 150 W CPU: the Xeon E5-2687W.

article link

Interesting article because they bench the flagship 6-core SBE as well as the even more expensive (and multiplier locked) 8-core xeon chip. Loads of benchmarks too.

Makes me jelly of those LGA2011 owners

 

[Smartazz]

Too bad there still isn't an 8 core SB-E. Do you think it's because of thermal issues?

 

[ShintaiDK]

Too bad there still isn't an 8 core SB-E. Do you think it's because of thermal issues?

Powerdraw. And it gives lower stock clocks(3.1 vs 3.5). Thermals could be an issue too, atleast on stock coolers.

 

[Insert_Nickname]

Too bad there still isn't an 8 core SB-E. Do you think it's because of thermal issues?

Nah, if your cooling setup can handle a 130W TDP, it can properly handle a 150W TDP without too many modifications... :hmm:

It almost begs for water-cooling...

Powerdraw. And it gives lower stock clocks(3.1 vs 3.5). Thermals could be an issue too, atleast on stock coolers.

What stock-cooler?. LGA-2011 doesn't have one. Intel has their own cooler but that's sold separately. Considering that the same-size stock-cooler on my old i7-920 could not keep it under 90C under full load at stock speed, I would not trust it anyway...

 

[WhoBeDaPlaya]

Heh, "a massive 150W TDP". It's like everyone forgot that we had cooling and power delivery solutions to cope with that level of power draw 4 years ago.

 

[Smartazz]

Powerdraw. And it gives lower stock clocks(3.1 vs 3.5). Thermals could be an issue too, atleast on stock coolers.

I'm sure there are enthusiasts who can handle a high clocked 8 core Sandy Bridge.

 

[ShintaiDK]

I'm sure there are enthusiasts who can handle a high clocked 8 core Sandy Bridge.

There is always someone for something. The question is if there is enough.

 

[boxleitnerb]

Personally I find this CPU pretty superfluous. They have the performance crown already. If it was at least with the same TDP as in the mainstream sector like 2600K and 2700K...

 

[Makaveli]

Doesn't provide enough of a boost to be worth it.

Also I think the xeon would spread its legs more in server workloads and not just rending.

In this review the 400mhz gap and ECC memory are making the gains looks much smaller.

 

[Nemesis 1]

Personally I find this CPU pretty superfluous. They have the performance crown already. If it was at least with the same TDP as in the mainstream sector like 2600K and 2700K...

True but its also true it uses the same power as AMDs best chip at 50% better performance

 

[inf64]

50% better performance? How about some facts from the article? Time to completion of whole THG test suite(ST and MT workloads) for 3970x is 91.5 minute. FX8350 took 116 minutes . From this we have: 91.5/116=0.788 or ~21% faster than 8350. 3770K is similarly 12% faster than 8350 when you take its total runtime in the equation.

 

[minitron]

Bad math.

116 - 91.5 = 24.5

24.5 / 91.5 = ~26.7%

 

[Smartazz]

Bad math.

116 - 91.5 = 24.5

24.5 / 91.5 = ~26.7%

Regardless, much less than 50%.

 

[minitron]

With his terrible assumption you can just add up the total times for all the workloads.

In heavily threaded applications the i7-3970X is ~57% faster than the FX-8350.

 

[guskline]

WOW battle of the UBER chips!

 

[Idontcare]

In heavily threaded applications the i7-3970X is ~57% faster than the FX-8350.

Which is great for those situation where an employer is willing to spend top-dollar on workstations for their professionals, absolute performance delivers top ROI when salaries are in the equation.

But I am also quite impressed by the fact that an i7-3970X is nowhere close to being only 57% more expensive than an FX-8350. For those situations out there in the working world, as well as the home, the performance/dollar delivered by the FX-8350 is impressive.

But I suspect it still doesn't matter because even an FX-8350 is probably going to be complete overkill for the majority of desktop PC users. CPUs like the 3570k/3770k and the FX-8350 or 3970X must be rather limited in global annual volumes.

I mean just look at how rare the Intel extreme cpus are around here, and we are already a rather niche portion of the enthusiast spectrum.

 

[inf64]

@minitroll

My math is not bad. When you say something is faster by factor of x than something else,then if you have total runtime of benchmarks of both products you divide the faster runtime(product X) with the one being slower. Substract from 1(or 100%) what you got and that is by how much product X is faster than product Y.
3970x is 22% faster than 8350.

Now if you would to say that FX is slower by xx amount than 3970x than your would take 116m and divide by 91.5. FX8350 is therefore 26.7% slower than 3970x.

As for validity of their benchmark ,of course it's valid. They took all the runtimes of all the chips in their test suit and compared them. The completion time in C11.5 is factored in, don't worry. But still with this outlier 3970x is just 22% faster than 8350. Also ST tasks are figured in too so they push the average a bit towards the intel side(since core is better in ST performance of course). 5x price difference for average of 22% better performance in their mixed(ST+MT) test suit.

But you cannot argue facts,you can try but you will fail .

 

[minitron]

No, you're wrong.

The FX (116s) takes 24.5 seconds longer to complete the task than the i7 (91.5s) which means the i7 completes the task 26.7% faster; you want to divide the additional time the FX take by the time the i7 takes. By your awful math the i7 completes the task in 79% of the time it would take the FX which is not the same as the i7 being 22% faster.

I understand this is high level stuff for you but maybe one day you'll get it. But probably not...

 

[bunnyfubbles]

Too bad there still isn't an 8 core SB-E. Do you think it's because of thermal issues?

there are 8 core Xeons on 32nm, the problem is that they're completely locked down outside of the typical maybe couple extra MHz on BCLK overclocking like we see with Sandy and Ivy 1155 non K CPUs, and thus a heavily overclocked 6 core is actually going to be faster than the fastest 8 core Xeon and completely not worth it for an enthusiast who is willing to blow past stock TDP with proper cooling.

but yeah, its too bad the 3970X wasn't an unlocked 8-core, I really don't know why Intel doesn't release one, I suppose there might be enough idiots with too much money for their own good who might try to pair a $1000+ enthusiast chip with a shoddy cooling solution. Another potential scenario might be yields just aren't good enough to justify selling the chips as mere $1000 EE CPUs that are capable of being high clocked Xeons and priced near $2000, and maybe Intel knows there just aren't enough extreme enthusiasts who will match or beat that price

 

[AtenRa]

No, you're wrong.

The FX (116s) takes 24.5 seconds longer to complete the task than the i7 (91.5s) which means the i7 completes the task 26.7% faster; you want to divide the additional time the FX take by the time the i7 takes. By your awful math the i7 completes the task in 79% of the time it would take the FX which is not the same as the i7 being 22% faster.

I understand this is high level stuff for you but maybe one day you'll get it. But probably not...

Intel XEON E5 takes 91.5 seconds to complete the run
FX8350 needs 116 seconds to complete the run

FX8350 needs 24,5 seconds more to complete then run

116 seconds are ~26,7% more than 91,5 seconds

Edit : That makes the FX8350 ~26,7% slower than Intel XEON E5 because it needs 24,5 seconds more to finish the run.

Intel XEON E5 needs ~21,1% less time to complete the run than FX8350

That makes Intel XEON E5 ~21,1% faster than FX8350.

 

[inf64]

Thanks Aten but I think it's like integral/differential calculus to this guy. He probably needs a math course to figure it out...
It's funny how he fails to realize that X being faster than Y by ZZ amount is not the same as Y being slower than X by WW amount. It's all so very complicated .

 

[Sheep221]

Intel has gone that far with performance that it can now compete with itself.

 

[octavian33]

We all know that Intel makes the faster processor. We also all know that intel is way more expensive. So my simple question is this, why doesnt amd just make a board that will take 2 FX-8350's. I mean is there a reason that 2 processors running in unison wouldnt at least or exceed intels 3970x in multithreaded applications, and at a cheaper price? I am a total newb at multi processor set ups but it seems logical to me.

 

[-Slacker-]

Work station motherboards do that already, with up to 4 sockets, whereas insane number of cores are not needed in the desktop market. Yet.

 

[Insert_Nickname]

We all know that Intel makes the faster processor. We also all know that intel is way more expensive. So my simple question is this, why doesnt amd just make a board that will take 2 FX-8350's.

Oh, they do... its just called Opteron instead...

Your choice of socket C32 or G34. Boards are a little more expensive then your average desktop board though...