FRONTPAGE Intel i7-6950X Broadwell-E CPU Review

Today we are going to get a chance to review Intel"s next processor in their High-End Desktop (HEDT) platform. The king of the consumer hill used to be the mighty i7-5960X our friend Dino reviewed back in August of 2014. That "extreme" CPU shook up the landscape then by providing the consumer with a high clocked, octo-core CPU with a total of 16 threads. Since then, Intel brought to market their 14 nm Broadwell CPU architecture, on their mainstream platform, through their "tick-tock" cycle. Naturally, the next HEDT processor is based off the same 14 nm lithography, Tri-gate (FinFET) technology is Broadwell-E.

Click to expand...

Click here to view the article.

Lookin good Joe, I was curious to see what if any single thread performance gains there might be with Broadwell E. Great job.

Nice review, Joe!

Nice one Joe!

Would love to see you freeze that baby and let us know what's what!

Good effort on the review very thorough and a good read

Nice review!

Now, to find a spare $1,700!

/sells kidney for $1700


OK, I'm ready to buy one of these

Nice job on the review Joe.

Thanks all! It really is an absolute beast assuming you can use all of its cores.

As soon as I get settled and have more time with the chip, I will get an overclocking guide out as well as take it cold. I had some issues overclocking much past that 4.2 GHz you saw in the article. I reached 4.5 GHz for a screenshot, but was sitting around 1.4V which made me a bit nervous. It was brute force. I have another board I just swapped out very late last night (this morning before I published) and hoping maybe it will get me over the hump.

I have read from another review (anand) they were at 4.1 GHz and 1.31V so it seems like I have a slightly better than average CPU (or at least better than their sample) to drive... not sure what the issue was... (looks in the mirror).

This chip has its small niche, multi-tasking I mean mega-tasking, and dominates in that niche. I agree with the assessment that the CPU is only worth the money to those who use heavily multi-threaded applications for their livelihood.

mdcomp said:

/sells kidney for $1700

Click to expand...

Don't get ripped off, do some research first

Great review, dont exactly like it has a thin substrate like skylake though
I await for someone to bend their $1700 cpu

Hi everyone ,



Hello ,

we all know that games dont use more than 8 threads today ...

so to take advantage of an 8 cores or 10 cores CPU in Gaming you should Disable HT (Hyperthreading) and run the gaming test again to compare it against the 4 cores i7 6700K .

and test it with SLI as well to reach the i7 6700k bottleneck !

let me put it more simple ,

The i7 6700K has 4 cores and can oc to 4.4 ghz easy . this CPU will give us 8 Virtual cores comparable to 2.2 GHZ clock for each virtual core .

However the 8 Coresi7 6900K , With the HT Turned OFF , will give us 8 cores @ 4.4 ghz EACH !

Thats double the speed of the 4 cores i7 ! if the game uses 8 threads .

EVEN if we dont OC the 8 cores , it would be 3.2GHZ VS 2.2 GHZ !!!

if you ask why Disable HT ? simple because the game will never use 16 Virtual cores !!! and the advantage is LOST .

Please run the test again for games with HT turned off in the 8 cores and 10 cores cpu .

and to stress the CPU more , TEST SLI as well , we want the i7 6700K to bottleneck !

THANKS

oh and Intel Should release i5 Broadwel-E CPU , 8 cores without HT , CHEAPER and BETTER for GAMERS

samer1970 said:

Hi everyone ,



Hello ,

we all know that games dont use more than 8 threads today ...

so to take advantage of an 8 cores or 10 cores CPU in Gaming you should Disable HT (Hyperthreading) and run the gaming test again to compare it against the 4 cores i7 6700K .

and test it with SLI as well to reach the i7 6700k bottleneck !

Click to expand...

Games don't even make use 8 threads very well since there's almost no difference in the i5 6500K (4c/4t) and i7 6700K (4c/8t). Are you wanting to see 4 real cores vs 8 real cores vs 10 real cores?

Testing 2-way, 3-way, and 4-way SLI with the i7 6700K and i7 6950X at the same clocks with or without HT enabled on both CPUs would show if, when, and how much 4c/8t or 4c/4t is a bottleneck in different SLI configurations. This would be interesting just to see, and might be useful to a very few people out there.

samer1970 said:

let me put it more simple ,

The i7 6700K has 4 cores and can oc to 4.4 ghz easy . this CPU will give us 8 Virtual cores comparable to 2.2 GHZ clock for each virtual core .

However the 8 Coresi7 6900K , With the HT Turned OFF , will give us 8 cores @ 4.4 ghz EACH !

Thats double the speed of the 4 cores i7 ! if the game uses 8 threads .

EVEN if we dont OC the 8 cores , it would be 3.2GHZ VS 2.2 GHZ !!!

Click to expand...

Assuming your 4.4GHz OC, that means the 6700K has four cores each running at 4.4 GHz and each one of those 4.4 GHz cores can process two threads at once with HT enabled. When HT is turned off, each one of those 4.4 GHz cores will only process one thread at a time. No dividing speed by threads to get a "speed per thread", it doesn't work like that. Also, we can't assume the 6900K will OC like an 6700K, although, it may very well make 4.4 GHz.

samer1970 said:

if you ask why Disable HT ? simple because the game will never use 16 Virtual cores !!! and the advantage is LOST .

Please run the test again for games with HT turned off in the 8 cores and 10 cores cpu .

and to stress the CPU more , TEST SLI as well , we want the i7 6700K to bottleneck !

THANKS

oh and Intel Should release i5 Broadwel-E CPU , 8 cores without HT , CHEAPER and BETTER for GAMERS

Click to expand...

I can agree that stressing the CPU more with multi-GPU setups would be an interesting read, but I don't see what you're going for with the CPU and HT testing unless you want 4 vs 8 vs 10 real cores (see if games take advantage of more real cores better than cores+threads). I mean, we know that there's slight to no IPC improvement between 6700K and 6950X, so a game that only takes advantage of 4c/8t will only take advantage of 4c/8t in the 6950X as well. Therefore, since there is little to no IPC improvement between 6700K and 6950X, there would be little to no performance increase between 6700K and 6950X in that game (results in the review confirm this). However, with the 6950X you would have 6c/12t free and unused by that game, which means more things could be done while gaming simultaneously without slowing down the game performance. The 6950X is about heavy multi-tasking and heavily multi-threaded applications. Games are only going to use ~40% of the 6950X's resources, the game (software) is the limiting factor.

MattNo5ss said:

...

Testing 2-way, 3-way, and 4-way SLI with the i7 6700K and i7 6950X at the same clocks with or without HT enabled on both CPUs would show if, when, and how much 4c/8t or 4c/4t is a bottleneck in different SLI configurations. This would be interesting just to see, and might be useful to a very few people out there.
...

Click to expand...

There should not will never be a case of bottle neck unless multiple M.2/PCI-E SSDs and multi-GPUs are in the config. There are more than enough lanes for these CPUs, and the mobo makers should know how to architect with this availability.

The PCIe lanes are not really the question though. Its the fact that with most high end cards in SLI/CFx that the faster your CPU speed is, the better your FPS are. In those situations, I do not believe that threading has anything to do with it. We saw similar scaling for a quad as you did with quad + HT, as you did with a hex + HT (not sure about a dual core).

Alright so I don't have time to go through a complete tear down (I'm typing these up in meetings).

Here is a quick glimpse at what can cause a common bottleneck in desktop mobos: An Intel CPU has a bunch of available lanes (a lot more than what is utilized). Each of these lanes are grouped into channels (since PCI-E is point to point). Each channel has a window to receive and transmit data. So if the channels are distributed correctly, and the signals on the channels are laid out correctly (think physical link), than the n-number of devices connected to these channels will have enough time to move their data to and from the CPU. What happens when you don't have these basics designed correctly (or gimped due to costs), than you will see bottlenecks in different tests.

A great example is multi-gpu with m.2. The GPUs typically will receive the best designed channels so that they never have issues. Yet an m.2 that is now just entering the market will not. Why gimp these channels? Simple answer: cost. More lanes = more board layers. PCI-2.x, 3.0 and future will be harder and harder to implement. PCI-E has a great recovery system, but that creates high overhead (who cares with PC). Faster cores helps mask this problem. Higher PCI-E bus somewhat helps, but ruins the channel signal integrity.

To add: Think channels as highways. With an X amount of cars present, the more lanes you have, the smaller your window. The smaller number of lanes increase the window. PCI-E uses windows for data transfer. Only 1 window at a time for communication.

I'm hoping to buy/bin 2-3 of these, keep the best, then sell (or return) the reject(s). No 'real world' advantage but dang I want to bench 2D and newer 3D with this beast.

Gotta wait for some cold results and more reviews and guides before even thinking about pulling the trigger on one of these, though.

Dolk said:

There should not will never be a case of bottle neck unless multiple M.2/PCI-E SSDs and multi-GPUs are in the config. There are more than enough lanes for these CPUs, and the mobo makers should know how to architect with this availability.

Click to expand...

That statement was about processing power of the CPU as number of GPUs increase, and when are the 4 cores of the 6700K not able to crunch enough data for all the GPUs, not really number of PCIe lanes available. It would help answer the question: How many GPUs can be added before I need a CPU upgrade? You know, a Pentium G3258 has 16 PCIe lanes just like a 4700K has 16 PCIe lanes, but a 2c/2t Pentium will surely bottleneck a double/triple GPU setup.

The available PCIe lanes are 16 vs 40 (6700K vs 6950X), so the 6950X could do a little better in configs with more than 2 GPUs since motherboards won't need a PLX chip.

Dolk said:

Alright so I don't have time to go through a complete tear down (I'm typing these up in meetings).

Here is a quick glimpse at what can cause a common bottleneck in desktop mobos: An Intel CPU has a bunch of available lanes (a lot more than what is utilized). Each of these lanes are grouped into channels (since PCI-E is point to point). Each channel has a window to receive and transmit data. So if the channels are distributed correctly, and the signals on the channels are laid out correctly (think physical link), than the n-number of devices connected to these channels will have enough time to move their data to and from the CPU. What happens when you don't have these basics designed correctly (or gimped due to costs), than you will see bottlenecks in different tests.

A great example is multi-gpu with m.2. The GPUs typically will receive the best designed channels so that they never have issues. Yet an m.2 that is now just entering the market will not. Why gimp these channels? Simple answer: cost. More lanes = more board layers. PCI-2.x, 3.0 and future will be harder and harder to implement. PCI-E has a great recovery system, but that creates high overhead (who cares with PC). Faster cores helps mask this problem. Higher PCI-E bus somewhat helps, but ruins the channel signal integrity.

To add: Think channels as highways. With an X amount of cars present, the more lanes you have, the smaller your window. The smaller number of lanes increase the window. PCI-E uses windows for data transfer. Only 1 window at a time for communication.

Click to expand...

Still not sure why we are talking about PCIe lanes...

You can thank samer for that because it's relative. 6700k has a lot less lanes than 6950x so at some point they will become saturated

Very nice review Joe.

i didn't expect it to overclock worse than a haswell-e. bummer. no oc gains and no ipc gains. this can't be right.