Z170 vs X99

[rosco6912]

Hi all,

I read somewhere that x99 does not take advantage of bandwidth of four channels unless you have four dimms present in the board, is this true?

I am trying to discern pros and cons of z170 vs x99. I've been out of rig building for awhile, and trying to look for less obvious advantages and disadvantages of each platform.

Thank you!

PS: context here is I'm helping my brother select the parts (and build) for an oculus ready machine; $1800-2200 budget. So far I was informed about nvidias Pascal, so we may actually wait for pascal release before we buy... The problem is will the good x99 boards/CPUs be gone or more expensive by then!?

 

[BigDaveX]

I read somewhere that x99 does not take advantage of bandwidth of four channels unless you have four dimms present in the board, is this true?

Well... yes. Each DIMM hosts a single channel, so you need four of them to populate all the memory channels. That being said, it's actually not the end of the world if you only have two DIMMs, which will cause performance to drop a bit, but not to a crippling extent.

Feature-wise, X99 and Z170 are actually very similar - the main difference aside from CPU support being that the former has more SATA ports, while the latter tends to feature more M.2 connectors - so it really depends whether or not you need the 6+ cores of the X99 platform.

 

[rosco6912]

Thanks for the reply Dave.

I am trying to decide if I should build now and just buy GPU later when pascal comes out, or wait and build the whole thing at that time.

Currently split on whether its best to go 5820k overclocked on x99 or 6700k on z170; concerned if I wait for pascal the 5820k might not be easy or cheap to come by, but maybe that won't matter if 6700 has gone down in price by then... Ugh.

 

[Deders]

Skylake can make much better use of memory bandwidth than X99.

Here's an example of my system running with 3200MHz DDR4 Dual channel compared to a X99 system running on 2133 Quad Channel. The memory benches are near the bottom. We were both running our CPU's at 4.6GHz

http://browser.primatelabs.com/geekbench3/compare/4733286?baseline=4738030

 

[Carfax83]

Currently split on whether its best to go 5820k overclocked on x99 or 6700k on z170; concerned if I wait for pascal the 5820k might not be easy or cheap to come by, but maybe that won't matter if 6700 has gone down in price by then... Ugh.

Z170 and X99 are both great platforms. I had this same question myself about a few weeks ago (although I admittedly had a strong lean towards the X99 since I was coming from the X79 platform), and I decided to go with the X99 platform when I saw a sale on the 5930K at Microcenter..

So far I've been highly pleased with my decision. The question you need to ask yourself, is:

1) Do you want or need two to four extra cores?

2) Do you want or need more PCIe lanes?

If the answer is yes, then go with the X99 platform. If the answer is no, then go with Z170..

With DX12 up and coming, it will be easier for developers to use more cores in games. Even now, there are several major 3D engines that can exploit more than four cores, including Frostbite 3, CryEngine, Ubisoft's AnvilNext and Disrupt engines, plus many others..

So even with Skylake's IPC advantage, the X99 platform will be faster due to the utilization of extra cores and the much larger L3 caches...

 

[Carfax83]

Skylake can make much better use of memory bandwidth than X99.

It really depends on the software you're using. The vast majority of consumer software cannot utilize the amount of bandwidth available to the X99 platform, much less exploit it.

That, and the fact that Haswell-E CPUs have a massive L3 cache explain why memory benchmarks can often be unreliable or unpredictable on the X99 platform. When you look at raw bandwidth though, the X99 has much greater bandwidth than Z170. Both the 5960x and the 6700k were using DDR4 2133..

Also Skylake typically scales better with faster memory for consumer software because it has dual channel instead of quad channel, which makes the bandwidth easier to utilize, and the smaller L3 cache which means the CPU has to go to main memory more often than the Haswell-E CPUs which have nearly twice, or more than twice the amount of L3 cache.

 

[Deders]

It does seem to vary with different hardware and configurations. I'm getting almost the same memory copy bandwidth as a 5820K in the same Aida copy test, with nearly half the cache. I think Skylake needs higher speed memory for you to really see the difference, I expect this is down to the way the memory controller works. Couple this with the additional registers (if the software can make use of them) and Skylake is kind of a halfway house between X97 and X99.

I agree with what you said before about extra cores and PCIe lanes being the important factor. Value for money and availability might be another.

 

[Carfax83]

It does seem to vary with different hardware and configurations. I'm getting almost the same memory copy bandwidth as a 5820K in the same Aida copy test, with nearly half the cache.

Here's my test that I just tried with Aida64, with DDR4 2666:

I think Skylake needs higher speed memory for you to really see the difference, I expect this is down to the way the memory controller works.

In the Tech report review, the reviewer speculated that Skylake may have more aggressive prefetching than previous architectures, so it can take better advantage of DDR4 higher bandwidth..

Couple this with the additional registers (if the software can make use of them) and Skylake is kind of a halfway house between X97 and X99.

What additional registers?

 

[Deders]

Not sure exactly of the exact details but I believe they help with running multiple instructions on the same core. Hence why games like GTAV see more of a benefit from Skylake than others.

This might have something to do with it:

The front end in the Skylake microarchitecture provides the following improvements over previous generation microarchitectures:

• Legacy Decode Pipeline delivery of 5 uops per cycle to the IDQ compared to 4 uops in previous generations.

• The DSB delivers 6 uops per cycle to the IDQ compared to 4 uops in previous generations.

• The IDQ can hold 64 uops per logical processor vs. 28 uops per logical processor in previous generations when two sibling logical processors in the same core are active (2x64 vs. 2x28 per core). If only one logical processor is active in the core, the IDQ can hold 64 uops (64 vs. 56 uops in ST operation).

• The LSD in the IDQ can detect loops up to 64 uops per logical processor irrespective ST or SMT operation.

• Improved Branch Predictor

http://www.intel.co.uk/content/dam/...4-ia-32-architectures-optimization-manual.pdf (page 32, section 2.1.1)

It would also be interesting to see the difference in SIMD paths between the two architectures.

For instance Sandra tells me this:

So 32 native simultaneous AVX2 registers per core and 16x FMA registers.

It's not comprehensive because Sandra picks the best to use, so any form of SSE is dismissed in favour of AVX or FMA.

Could you run a single threaded multimedia CPU benchmark to compare apples to apples, just for curiosities sake?

 

[rosco6912]

Wow I'm confused now

 

[Deders]

Wow I'm confused now

Just think about what you want your machine to be able to do.

 

[rosco6912]

I want a fresh build PC to do oculus rift as well as possible for as long as possible for $1800-2200

 

[Deders]

So you need to keep framerates as high as possible to avoid spoiling the illusion. It's a tough call. Are there any benchmarks for games that show that X99 beats an i7 6700K?

 

[Carfax83]

So 32 native simultaneous AVX2 registers per core and 16x FMA registers.

That's the same as Haswell. x86-64 architectures have the same amount of registers, which is why I asked you about the supposed additional registers in Skylake.

Both Haswell and Skylake can issue 32 single precision, or 16 double precision flops per cycle.

Could you run a single threaded multimedia CPU benchmark to compare apples to apples, just for curiosities sake?

Like which one? Anything single threaded is going to disadvantage Haswell-E though..

 

[Deders]

That's the same as Haswell. x86-64 architectures have the same amount of registers, which is why I asked you about the supposed additional registers in Skylake.

Both Haswell and Skylake can issue 32 single precision, or 16 double precision flops per cycle.



Like which one? Anything single threaded is going to disadvantage Haswell-E though..

I meant the same Sandra benchmark. It was just to see how many SIMD registers each core has. I'm sure that additional registers were mentioned somewhere in the Skylake thread. I know Sandy and Ivy bridge for instance had more SIMD registers than Nehalem. Notably SSE.

It does still have the optimised frontend which is able to handle 5 uops per cycle instead of four. As well as the other benefits listed above.

 

[Carfax83]

So you need to keep framerates as high as possible to avoid spoiling the illusion. It's a tough call. Are there any benchmarks for games that show that X99 beats an i7 6700K?

The best CPU reviews are done by either Tech Report or Hardware.fr.

In the hardware.fr review, all of the Haswell-E CPUs are beating the 6700K by 22% in Crysis 3..

This proves what I said earlier. Skylake's IPC advantage isn't enough to overcome the advantage of Haswell-E's extra cores in games that can make use of them....and the amount of games that can make use of them will be growing this year due to DX12.

 

[Deders]

The best CPU reviews are done by either Tech Report or Hardware.fr.

In the hardware.fr review, all of the Haswell-E CPUs are beating the 6700K by 22% in Crysis 3..

This proves what I said earlier. Skylake's IPC advantage isn't enough to overcome the advantage of Haswell-E's extra cores in games that can make use of them....and the amount of games that can make use of them will be growing this year due to DX12.

That particular level is renowned for CPU usage, I often bring it up myself. On the other hand the Arma3 benchmarks below massively favour Skylake. (Edit: as do all the other games in that article.)

Also one of the advantages of DX12 is being able to use much less CPU to draw much much more.

 

[Carfax83]

That particular level is renowned for CPU usage, I often bring it up myself. On the other hand the Arma3 benchmarks below massively favour Skylake. (Edit: as do all the other games in that article.)

Yes, because CryEngine scales to 6 cores, just like Frostbite 3. Anything that uses more than four cores is going to perform better on Haswell-E, whilst anything that uses only four cores is going to be better on Skylake. Watch Dogs is the only other game in that review that can scale to 6 cores, and the 5930K is almost as fast as the 6700K despite the clock speed disadvantage.

Skylake has a significant clock speed advantage against the Haswell-E CPUs in that review as everything is running at stock clocks.

BTW here's my single thread Sisoft Sandra scores. That's at 4.4ghz, so we're not at parity clock speed wise..

Also one of the advantages of DX12 is being able to use much less CPU to draw much much more.

Yep. The truth is, both platforms are awesome performance wise. That's why the OP's choice should come down more to features and price than anything else..

 

[Deders]

Didn't have the time yesterday but I thought it might be a good idea to put up the multithreaded results just to give the full picture.

 

[Carfax83]

Damn, and I already uninstalled it LOL! It has a service running in the background, and that made me suspicious.. What in the hell does benchmarking software need a service running in the background for?

 

[Deders]

Not sure but there a lot of different software technologies combined. It likes it if you restart after you install so that GPGPU benchmarks work properly. Could it be that the restart is needed to get the service running?

 

[maxkillz]

Many Skylake users, including myself, are experiencing random freezeups during gaming. If I'd have to do it again, I'd go with X99. Here's a thread with people having the problems: http://www.tomshardware.com/forum/id-2830772/skylake-build-randomly-freezing-crashing.html

 

[Magic Carpet]

Just sexy! UT fans should pay close attention. If you don't need iGPU and/or tiny form-factor, X99 is the way to go, imo.

 

[Deders]

Many Skylake users, including myself, are experiencing random freezeups during gaming. If I'd have to do it again, I'd go with X99. Here's a thread with people having the problems: http://www.tomshardware.com/forum/id-2830772/skylake-build-randomly-freezing-crashing.html

Have you tried upping the ram and/or vccio?

 

[Shmee]

X99 also has an upgrade path by the looks of it, to BW-E