Why not Quad memory channels in Skylake/Broadwell?

What I really meant to ask is why not quad memory channels in quadcore Skylake/Broadwell CPU's, but I can't edit the title of this post.

If Intel could stuff 4 memory channels in the i7-3820 (Sandybridge-E) why can't they do it with all quad core CPU's? Or would it just be a waste of silicon real-estate?

The CPU really has nothing to do with it... well, the number of cores doesn't anyway (the memory controller of course would).

It would stand to reason that Intel wants to keep their HEDT and mainstream platforms separate.

EDIT: I am sure there are more technical reasons too of course. But it has nothing to do with the number of cores I don't imagine.

That is one of the reasons I had my heart set on a 5820k for my new build. But for the extra cost of a good mobo+Haswell 6 core, I just couldn't do it. And for what I do with a PC I'm pretty sure I'll never notice the "shortcoming". Skylake and DDR4 isn't lacking anything I've found. If I do run in to a scenario where 16 GB isn't enough, another $90 will get me a total of 32 GB of DDR4 3000. Is it possible DDR4 made the quad channel unnecessary?

Alaric said:

That is one of the reasons I had my heart set on a 5820k for my new build. But for the extra cost of a good mobo+Haswell 6 core, I just couldn't do it. And for what I do with a PC I'm pretty sure I'll never notice the "shortcoming". Skylake and DDR4 isn't lacking anything I've found. If I do run in to a scenario where 16 GB isn't enough, another $90 will get me a total of 32 GB of DDR4 3000. Is it possible DDR4 made the quad channel unnecessary?

Click to expand...

Haswell-E does use DDR4. I think I've seen memory benchmarks where Skylake beat Haswell-E. I think latency was one metric and there was at least one other (maybe write bandwidth).

But for desktop apps and gaming maybe memory bandwidth is nearly irrelevant?

That is one of the reasons I had my heart set on a 5820k for my new build. But for the extra cost of a good mobo+Haswell 6 core, I just couldn't do it.

Click to expand...

Last time I did the math, it was around $100 difference. Motherboard was around $30-40 (saying $150-170 for solid z170 and $180-$210 for solid X99 - That said, you could have bought a solid X99 board for the same price as your Z170), CPU difference is $40 at newegg. Ram is not much more either. But, yeah...

Quad channel wasn't really a benefit in the DDR3 days. DDR4 had nothing to do with it.

But for desktop apps and gaming maybe memory bandwidth is nearly irrelevant?

Click to expand...

It is (nearly irrelevant).. correct. There are some situations where bandwidth is needed in gaming... but it depends on the game (what was it, fallout 4? I vaguely recall discussing this with you Mags, LOL..) and very few respond to more bandwidth. Now, when you get to multi-gpus, it tends to help a bit. It also helps with an integrated GPU as the iGPU uses system ram as buffer.

If you SLI or Xfire, bandwidth becomes more important if IRC. Woomack will be the one to dig out the truth on this. I'm hoping he'll be along soon, because now you've got me curious on the subject all over again.

edit:

Last time I did the math, it was around $100 difference. Motherboard was around $30-40 (saying $150-170 for solid z170 and $180-$210 for solid X99 - That said, you could have bought a solid X99 board for the same price as your Z170), CPU difference is $40 at newegg. Ram is not much more either. But, yeah...

Click to expand...

Yeah, but $40 is $40! And my Skylake is newer! And shiny! And I needed a $200 mobo that was the right color scheme for my build! And....ah hell.

I think it depends on the platforms you compare. For instance my LGA775 system with a quad core at 3.6GHz and 8GB of RAM in dual-channel was easily beaten by an LGA1366 system with a quad core with hyper-threading (4C/8T) at 2.66GHz (E5640) with 6GB of RAM in triple-channel (or an LGA1156 system with a quad core without hyper-theading at 2.93GHz and 4GB of memory, for that matter).

Though my current LGA1150 system with a non-hyperthreaded i5 4690K (stock 3.5GHz) and 16GB of memory (perhaps overkill) can beat all three of those setups with relative ease (it stomps the LGA775, even overclocked at any speed practically). And even with the E5640 overclocked to 4.4GHz, it still comes up short of an i5 4690K when last I benchmarked the two.

I've never actually owned an LGA2011 or LGA2011-v3 system, so can't speak to their abilities or strengths.

I know that my LGA1150 system has a memory transfer rate of around 20GB/s the last time I checked (at 800-900MHz/DDR3-1600 to DDR3-1800). LGA775 was around 5.5-7GB/s at my best, and LGA1366 was around 12-16GB/s IIRC (a little fuzzy here on its number, as I've not tested LGA1366 RAM transfer rates lately). I think the transfer speeds are so high on dual channel alone that there's probably little point in having three or four channels from a raw RAM performance perspective at this time. When LGA1366 came out by contrast, there was a lot of performance to gain by increasing channels since the only competing Intel platform at the time was LGA775 in the mainstream and server markets IIRC, and AMD was no real competition in RAM performance at the time, as they were in socket AM2 at the time which was still on DDR2 memory that was limited to 5-6GB/s on their platform.

From the real-world tests in the past dual channel and Quad channel memory had no impact. Does anyone have some links for new reviews of real-world benchmarks?

I could only find memory scaling in your link.

wingman99 said:

From the real-world tests in the past dual channel and Quad channel memory had no impact. Does anyone have some links for new reviews of real-world benchmarks?

Click to expand...

Fallout 4 benchmarks do respond to faster RAM speed:

http://wccftech.com/fallout-4-performance-heavily-influenced-by-ram-speed-according-to-report/

wingman99 said:

From the real-world tests in the past dual channel and Quad channel memory had no impact. Does anyone have some links for new reviews of real-world benchmarks?

Click to expand...

Fallout 4 benchmarks do respond to faster RAM speed:

http://wccftech.com/fallout-4-performance-heavily-influenced-by-ram-speed-according-to-report/

I think I was commenting that Anand article some months ago as they were comparing all memory kits not looking at ranks etc. Results can be different and only bandwidth can be up to 20% different in some cases. I thought that reviewers on Anand know that but clearly not.

You can't direcly compare dual and quad channel performance because they're using different CPUs, cache etc. Everything is affecting performance, especially cache speed. Skylake has smaller but faster cache. Also in quad channels are longer traces, longer delays etc. You can see access time in benchmarks like AIDA64. Dual channel will always have better access time.

Home/office CPU series don't have quad channel controllers because it would raise die size, wattage and price. For some reason Intel raised price of their chipsets anyway but 100 series chipsets/motherboards are still much cheaper than X99.
Where can you see better quad channel performance ? In heavy multitasking, rendering of large files on optimized applications etc. My colleague sees it while rendering scenes for animations that they make at work ( mainly TV commercials ).

At home you won't see the difference between quad and dual channel ( out of benchmarks ) but you can see difference between 4 and 6 cores so it can be good reason to move to 6 cores ( if you can use it ). At least it was calculating if you were comparing 6700K+higher series board with 5820K+any lower but still high end X99 board what was giving not much higher price. Broadwell-E cost more so you can't really compare it like that now and I guess that Intel wanted to make larger price difference between 4 and 6+ cores.

I think I had posted here in the past, I do run compute software where ram performance is as important as CPU. As a rough balance, for a Skylake quad core, ideally I'd need the ram to be numerically the same speed as the CPU clock. e.g. quad core Skylake running at 4 GHz, need 4000 MT/s dual channel ram to be "practically unlimited". This also assumes dual rank ram, and there is a stiff performance penalty for single rank, which is particularly annoying as it is rarely specified on ram kits and even 8 GB modules are moving towards single rank these days.

The side effect of the above is, I find the i3 is a great performer. You get two fast cores, and two ram channels to feed it. Nothing can be overclocked nor needs to be, so it can be done very cheaply compared to an OC i5 or i7 setup where you have to consider quality components. Note: this is for a compute scenario where you don't care the cores are in separate boxes or not, you just need a lot of them.

Edit: noticing which forum I wrote this on, I should clarify, I don't consider that the i3 can be usefully overclocked in any meaningful way for this application. The method that breaks AVX would cost you far more in performance than any clock gain, and maybe the Asrock non-Z ram OC could help a little but it would be minor enough that should only be a consideration if it doesn't impact cost to implement.

wingman99 said:

I could only find memory scaling in your link.

Click to expand...

Sorry, what were you looking for?

magellan said:

Fallout 4 benchmarks do respond to faster RAM speed:

http://wccftech.com/fallout-4-performance-heavily-influenced-by-ram-speed-according-to-report/

Click to expand...

That would be about the only title tbat sees such impeovements (that are not sli or igpu).

Number of cores do not have anything directly to do with number of channels, but at higher core counts given the same memory bandwidth, memory will more often become the bottleneck. So it makes sense to scale number of channels with core count.

Adding channels add cost on many layers. Larger and more complicated memory controller (silicon area), more pins and I/O drivers (extra cost on the CPU and socket), and more traces to route (extra cost on motherboard by potentially requiring additional PCB layers).

This is why -E sockets always have more pins than the non-E counterparts. Most of the extra pins are probably used for the extra channels.

Tech Tweaker said:

I know that my LGA1150 system has a memory transfer rate of around 20GB/s the last time I checked (at 800-900MHz/DDR3-1600 to DDR3-1800). LGA775 was around 5.5-7GB/s at my best, and LGA1366 was around 12-16GB/s IIRC

Click to expand...

AIDA64 on my 5930k says I have ~50 GB/s (on DDR4-2600).

AIDA64 results scale with cores, FYI.

EarthDog said:

AIDA64 results scale with cores, FYI.

Click to expand...

Isn't that what we're talking about? More cores being the need for more memory channels with a side effect of more bandwidth?

It's not quite a simple count of cores as clock counts towards it also. You need enough combined CPU demand to saturate the memory interface. The problems I see only really kick in with quad cores and dual channel interface, particularly with higher clock CPU and/or lower clock ram (e.g. DDR3-1600).