Broadwell-E Review

[Arachnotronic]

http://www.guru3d.com/articles-pages/core-i7-6950x-6900k-6850k-and-6800k-processor-review,12.html

Looks like Haswell-E's DDR4 IMC hit's saturation point very quickly, but derives the latency benefits from the higher speed ram.

Definitely looks like DDR4 2400 is the highest Haswell-E can handle in terms of actual throughput.

I'm not following. How can you conclude this from the link that you provided? I see 6950X tested at 2133mhz and 3200mhz, but 5960X is tested only at 2133mhz. Any further explanation would be helpful.

 

[24601]

I'm not following. How can you conclude this from the link that you provided? I see 6950X tested at 2133mhz and 3200mhz, but 5960X is tested only at 2133mhz. Any further explanation would be helpful.

Sry, i meant Broadwell-E, staying up all night must be getting to me o.o

The brain's autocorrect is a frightening thing to behold.

I was looking at the throughput scaling between the 2133 speed bin compared to the 3200 mhz speed bin

 

[Arachnotronic]

Sry, i meant Broadwell-E, staying up all night must be getting to me o.o

I was looking at the throughput scaling between the 2133 speed bin compared to the 3200 mhz speed bin

Makes more sense

Yes, you appear to be correct. I wish they had run some tests with a 2400MHz kit just so we could verify though.

Anyway, the quad channel alone drives an absolutely gargantuan amount of bandwidth. I wouldn't pay for more than a 2400MHz kit for BDW-E.

 

[biostud]

Not much improvement over Haswell, but I guess that was expected. It would have been nice to see better overclocks with lower voltage. So not really much better than Haswell-E when comparing overclocked CPUs.

 

[exar333]

This can't be too surprising due to Broadwell's issues. I will consider going back to HEDT with Skylake-E (or whatever the new iteration will be). I want 8C/16T that will overclock well, at or under ~$600 or so.

I can honestly say I am enjoying my 4C/8T Skylake as much as my 6C/12T SB-E/HW-E setups.

 

[Arachnotronic]

Not much improvement over Haswell, but I guess that was expected. It would have been nice to see better overclocks with lower voltage. So not really much better than Haswell-E when comparing overclocked CPUs.

What I found interesting was the power consumption. It has improved but not a whole lot compared to HSW-E.

I think 14nm has turned out to be something of a bust for Intel in terms of performance/power improvements over 22nm. The focus on density probably dinged them here.

 

[ZGR]

http://www.guru3d.com/articles-pages/core-i7-6950x-6900k-6850k-and-6800k-processor-review,12.html

Looks like Broadwell-E's DDR4 IMC hits saturation point very quickly, but derives the latency benefits from the higher speed ram.

Definitely looks like DDR4 2400 is the highest Broadwell-E can handle in terms of actual throughput.

It looks like 3200 MHz has some solid gains. That is excellent!

 

[Grooveriding]

$2300 in Canada for 6950X. Yeah I'll be spending that on big die Pascal Titan cards when they come out and getting an upgrade I can feel. Looks like it doesn't overclock that well either, so it would be a downgrade...

 

[zentan]

From a price/performance perspective only i7 6800k makes sense for many. The rest of the line up may not interest that many.


From Tweaktown's review:

Higher frequency memory handling seems to have improved from Haswell-E

Both the Corsair Dominator Platinum 3200C16 (4x8GB) and G.Skill Trident Z 3200MHz C14 (4x8GB) kits were overclocked to their XMP settings with ease on both motherboards. The ease of the 3200MHz overclock, especially with the tight C14 CAS latency on the G.Skill shows that Intel has made strides on improving the 3200MHz multiplier and overall mainstream memory XMP compatibility for Broadwell-E. At the Haswell-E launch, overclocking to this frequency was not a simple task, usually requiring a lot of voltage adjustments and even BCLK dividers, but with Broadwell-E, it is easy, breezy, and beautiful.

I had a lot of trouble with overclocking my memory to 3200MHz with the previous generation 5960X, but I have no such issue with the 6950X. Even with the newer motherboards, both 3200MHz kits from both vendors easily hit 3200MHz using the 32x memory multiplier

http://www.tweaktown.com/reviews/77...950x-10-extreme-edition-cpu-review/index3.htm

Overclocking the i7 6950X

Overclocking all cores like I would on the 5960X yielded a maximum overclock of 4.4GHz. At 4.4GHz, the CPU is a beast, but I fear that many of you will have trouble getting that 4.5GHz mark without overclocking only one or two cores to 4.5GHz while the rest stay below. Voltages are much the same as Haswell-E, but watch the input voltage as it does help stability but also greatly increases temperatures. VCore around 1.3-1.4v can be handed by an AIO cooler like the Corsair H115i I used. I didn't have time to tune much for a maximum overclock irrespective of stability (CPU-Z validation), but I was able to boot into Windows at 4.5GHz, but I quickly crashed. When vendors release their Windows light overclocking programs, I will be able to test maximum overclocks

.

SKYMTL from Hardwarecanuck is getting 4.4GHz as the wall for i7 6950X and 6900K.

Considering it's a 10C/20T against a 8C/16T there's a decent improvement in power consumption

 

[Arachnotronic]

Nice, so the memory controller is much improved. I remember that the one on my 5960X crapped out when I overclocked it much. I have a DDR4-2400 quad channel kit coming for my 6800K so as long as it runs flawlessly, I'm a happy camper.

 

[IndyColtsFan]

I'm wavering between the 6700k/7700k and the 6800k. My main concern with the 6800k is that it is on an older platform and OC ability. I would hope I could reliably hit 4.4 Ghz with it. My goal for this next PC would be to last 5 years, so I feel the 6800K is probably better in that regards. I can't wait to hear some of your own reviews to help me with my decision.

 

[AtenRa]

One 1080 per system for now. 6700K rig gets one and now my soon-to-be-Broadwell-E rig gets one

Also, 12 extra PCIe lanes isn't worth a couple hundred bucks to me, especially since I've run SLI on a 6700K with great scaling.

ok, thought you were going for 1080 sli.

 

[Arachnotronic]

ok, thought you were going for 1080 sli.

Naw, at least not right now. The performance of a single 1080 is so good, especially overclocked (~2ghz GPU, 5.3GHz GDDR5X) that I don't see a need to bother with the hassles of SLI...at least until I run into a game that doesn't give me the FPS I want.

 

[Magic Carpet]

What I found interesting was the power consumption. It has improved but not a whole lot compared to HSW-E.

Speaking of performance per watt, it is amazing to see the 6-core BDW equaling in power consumption the 4-core Skylake in this test, though.

And just matching the DT BDW in this test:

 

[Headfoot]

I don't know if I just got a good chip or what but my 5820k had absolutely zero issues running DDR4 3200 right off the bat. I just set the strap to 125 per the XMP profile and its worked without a single issue or any further configuration since at 4.4ghz on 1.25v. Seeing BDW-E reviews, I'm glad I went with the cheaper 5820k a while back instead of waiting. I'd get no performance as it looks like Haswell E can overclock enough better to outstrip the small increase in IPC if you're getting anything less than the $1700 10 core.

Can't help but feel massively underwhelmed by this release

 

[ElFenix]

What GPU are you running? Are you on Ivy E or Ivy mainstream?

3570k/290.

I expect a processor will last me 5 years these days so future proofing is important. My 3570 is already 4 years old now and it seems like we may just now be getting to a point where i7s are offering tangible gains to their i5 counterparts. So, moar cores?

But 1. Dat price increase; 2. Feels kinda weird 4 years later to be buying a processor only 1 generational change on.

 

[Elixer]

I don't see why AT's review didn't include the one game that is supposed to max out all of the resources on the system, both CPU & GPU?
Ashes of the Singularity should have been shown, since it is pretty much a wash for the other games they picked.

 

[2blzd]

Sry, i meant Broadwell-E, staying up all night must be getting to me o.o

The brain's autocorrect is a frightening thing to behold.

I was looking at the throughput scaling between the 2133 speed bin compared to the 3200 mhz speed bin

Makes more sense

Yes, you appear to be correct. I wish they had run some tests with a 2400MHz kit just so we could verify though.

Anyway, the quad channel alone drives an absolutely gargantuan amount of bandwidth. I wouldn't pay for more than a 2400MHz kit for BDW-E.

Wait I thought there was actual proven perormance benefits once you started going over DDR4-3000? I was planning on get 32 gigs of 3200. Its dirt cheap right now...what about 2400 being the max throughput? Im confused

 

[Arachnotronic]

Wait I thought there was actual proven perormance benefits once you started going over DDR4-3000? I was planning on get 32 gigs of 3200. Its dirt cheap right now

I am not sure, I thought that was for SKL only. I hope some website does a detailed memory test with BDW-E.

 

[24601]

Wait I thought there was actual proven perormance benefits once you started going over DDR4-3000? I was planning on get 32 gigs of 3200. Its dirt cheap right now...what about 2400 being the max throughput? Im confused

Regardless of the throughput scaling (Which historically has been horrible on all Sandy-E/Ivy-E/HW-E and now BW-E), the latency definitely shows linear gains based on total actual latency of the DDR4.

Hell, HW-E doesn't even scale past DDR4-2133 in terms of throughput, with the write performance capping at below DDR4-1800.

If SKL throughput scaling remains linear up to DDR4 4800 (It's been shown to definitely remain linear up to at least 3600, and defnitely scales up to at least 4000), SKL with 2 channel will have more throughput than HW-E with 4 channels.

 

[2blzd]

So when people, like the ones in this thread, saw they're running 3200 DDR4 on Haswell-E, you're saying its a waste and has no improvement over 2400?

Im so confused. that doesnt make any sense

 

[24601]

So when people, like the ones in this thread, saw they're running 3200 DDR4 on Haswell-E, you're saying its a waste and has no improvement over 2400?

Im so confused. that doesnt make any sense

Would depend entirely upon whether the application and usage environment you plan on implementing for yourself benefits from lower total latency main memory access in the absence of throughput increase.

But if you were doing it purely for the imagined extra throughput, don't.
It doesn't give extra throughput.

Skylake's IMC is the first in a long, long time that has actually given good scaling of throughput upon nominal and advertised main memory speed increase.

 

[Arachnotronic]

Would depend entirely upon whether the application and usage environment you plan on implementing for yourself benefits from lower total latency main memory access in the absence of throughput increase.

But if you were doing it purely for the imagined extra throughput, don't.
It doesn't give extra throughput.

Skylake's IMC is the first in a long, long time that has actually given good scaling of throughput upon nominal and advertised main memory speed increase.

Skylake is limited to dual channel. BDW-E has a beefy quad channel controller

 

[24601]

Skylake is limited to dual channel. BDW-E has a beefy quad channel controller

I was answering a question about DRAM throughput scaling on HEDT systems, but if you insist i will talk about skylake's DDR4 IMC/Cache architecture scaling.

Simply running SKL at DDR4 3600 (which gives linear throughput increase on Skylake) already gives you comparitively a "beefy 3 channel controller"

A bump to DDR4 4800, assuming linear throughput increase, would give SKL comparatively a "beefy 4 channel controller"

Given that many of the DRAM manufacturers are currently at the beginning of releasing mass market products based on smaller lithography, DDR4 3600-4000+ should be commonplace without large markups fairly soon.

http://www.eetimes.com/document.asp?doc_id=1329380

http://www.extremetech.com/extreme/226195-samsung-announces-new-10nm-class-ddr4

http://www.kitguru.net/components/m...-of-dram-using-10nm-class-process-technology/

 

[2blzd]

I'd be using this for Premiere Pro and other multi-threaded work where I need all the bandwidth I can get. I've seen and read benchmark guides that show the higher end rendering machines using faster ram. The bottleneck for most GPU acceleration applications has been CPU and ram and I've been told (and seen) faster ram makes a significant difference. Premiere Pro is GPU accelerated and I want a system that can fully push, take advantage of and utilize all of the gpu horsepower it can.

16gb of ddr4-3200 is going for around $70...so x2...$140 for 32gb ddr4 3200...I thought that was really good...but your comments about throughput are confusing me. are you talking about raw bandwidth?