MSI X99A XPOWER GAMING TITANIUM LGA2011-v3 Review @ [H]

[FrgMstr]

MSI X99A XPOWER GAMING TITANIUM LGA2011-v3 Review - MSI’s X99A XPOWER GAMING TITANIUM is a mouthful, but the XPOWER series has been a favorite of ours here at HardOCP for years now. The latest X99 iteration has much to prove. Is the X99A XPOWER GAMING TITANIUM another pretty face, or is it a fitting addition to the venerable XPOWER line?

I have removed this review for now as I think we did not do some due diligence on our end correctly and have possibly not reported what I would call a "solid" motherboard review. I am pulling all the equipment and parts back into my offices to further investigate this motherboard. Once I figure out what all is going on, I will report back, but at this time I think we are dealing with an equipment failure on our end, and it is not an MSI issue.

 

[MOS8580]

Scary stuff on the Broadwell-E OC degrading quickly. Saw in the Subsystem Testing that it's an Engineering Sample. Are those typically identical to final silicon? Problems relegated to the 10-core version as opposed to CPUs with some cores fused off maybe?

 

[prime2515102]

Do you think this degradation in overclockability is related to the move to 14nm? Have you noticed anything similar with Skylake?

My paranoia wants me to think Intel is sabotaging their overclockability to improve sales (as odd as that sounds).

 

[Quartz-1]

Not stable at stock? Instant fail, TYVM.

And you don't use possessives to denote plurals.

 

[Vudaz]

OMG what a beautiful piece of hardware. But, if you are not an expert in OC is a waste of money Ithink.

 

[gathagan]

I find it incredible that so many boards have difficulty reaching the memory speeds they pretend to support.
This seems to be true of most, if not all, of the X99 boards.

Whatever shortcomings they have, at least the Z170 boards seem to be much more successful in this area.
I noted one review stating that the higher-end Z170 boards seem to be better in this regard than their lower-tier cousins.
Is this a chipset issue, a board manufacturer issue, or just marketing hyperbole?


On an editorial note:
Every motherboard review that Dan has written since July 30, 2013 has included DPC latency testing.

The DPC Latency section of almost every motherboard review from Dan since July 30, 2013 has started with the statement:

"Deferred procedure call latency or DPC testing is something that we’ve been asked about and this is the first article we’ve done which integrates that type of testing."

I'm not trying to be a jerk, but can we all agree that "...this is the first article..." in no longer true?

 

[AK0tA]

Super nice looking board but to many issues for this price point. I am due to build a new system I reckon so I am starting my research phase. I think I want to do a WW I diorama case build, hmmm.

 

[Dan_D]

Scary stuff on the Broadwell-E OC degrading quickly. Saw in the Subsystem Testing that it's an Engineering Sample. Are those typically identical to final silicon? Problems relegated to the 10-core version as opposed to CPUs with some cores fused off maybe?

Yes, it is an engineering sample. ES CPUs manufactured and given to reviewers prior to launch are virtually identical silicon to that of retail CPUs. They are usually cherry picked examples that tend to be on the higher end of the spectrum, but I've never seen an ES CPU that was substantially better or worse than a retail CPU.

Not stable at stock? Instant fail, TYVM.

And you don't use possessives to denote plurals.

That's my feeling on things too. I always caught a lot of flak for my disdain of DFI LanParty motherboards for that. They often required tuning for stability at stock speeds and many enthusiasts excused that. I never did. MSI doesn't get a pass on it either, but at least I know what two or three settings fix the problem. I make sure to mention this in the review so that anyone who buys one can do the same. I've also provided this information in many forum threads concerning MSI motherboards. It's not a problem on their midrange motherboards a lot of times, but their higher end offerings all suffer from this and have for a couple of years or more now.

I find it incredible that so many boards have difficulty reaching the memory speeds they pretend to support.
This seems to be true of most, if not all, of the X99 boards.

Whatever shortcomings they have, at least the Z170 boards seem to be much more successful in this area.
I noted one review stating that the higher-end Z170 boards seem to be better in this regard than their lower-tier cousins.
Is this a chipset issue, a board manufacturer issue, or just marketing hyperbole?

So here is the skinny on memory support for X99 and even Z170. Memory clocks, and what you can achieve are primarily dependent on two separate, but very important variables. The first is the motherboard itself. The PCB has to have enough layers to support the higher frequency signaling. Power phases and trace path design need to be up to the task. The latter is almost a given for MSI, GIGABYTE and ASUS. They use equidistant trace paths that prevent some slots from performing substantially better or worse than others. As long as the power phases are up to the task, higher frequency clocks should be achievable. We've seen this separation more on the Z170 side where some of the less expensive motherboards can't do more than DDR4 3000MHz or whatever.

The second limiting factor is the CPU's integrated memory controller. Often times, the IMC is capable of the higher frequency memory clocks but only at or near stock CPU clocks. As the CPU clocks increase, the memory clocks often have to be brought down in order to achieve stability. Some of my test CPUs will let me go 100 or 200MHz higher if I drop the memory clocks. Having said that, I've seen plenty of Z170 motherboards that can do DDR4 3600MHz even when they are only rated for DDR4 4300MHz or thereabouts, and plenty of X99 motherboards that can't break DDR4 2800MHz. X99 typically won't clock as high as Z170 with regard to memory clocks. Broadwell-E is in my experience at least, no better than Haswell-E in this regard.

In a sense, the marketing guys are probably not lying. I've had manufacturers make claims and even show proof of being able to hit insane memory clocks.

On an editorial note:
Every motherboard review that Dan has written since July 30, 2013 has included DPC latency testing.

The DPC Latency section of almost every motherboard review from Dan since July 30, 2013 has started with the statement:

"Deferred procedure call latency or DPC testing is something that we’ve been asked about and this is the first article we’ve done which integrates that type of testing."

I'm not trying to be a jerk, but can we all agree that "...this is the first article..." in no longer true?

I will update my template to correct that.

 

[DrLobotomy]

Dammit MSI!!! This is the wrong board to be screwing around.

Good to see you guys found these issues so we don't have to the [H]ard way.

Seems to me if the manufacturers would just cut out 1 or 2 models and spent more time on the rest we might not have this flakiness that seems to hound the x99 platform.

 

[grtitan]

Any recommendation for a good X99 matx mobo?

 

[darkxmen]

Another great review. Would love for you guys to use the lower end cpu's on your testing, like the 5820k or its counter-part 6800k. i think on a average more people use those cpu's than the out of reach 10 core. Would also love to see some EVGA reviews in the future.

 

[FrgMstr]

Another great review. Would love for you guys to use the lower end cpu's on your testing, like the 5820k or its counter-part 6800k. i think on a average more people use those cpu's than the out of reach 10 core. Would also love to see some EVGA reviews in the future.

What is the point of using those CPUs when we are really only looking for problems with motherboards? Not sure what that is going to show us, please explain.

EVGA refuses to sample us parts of any kind.

 

[gathagan]

Any recommendation for a good X99 matx mobo?

Can't address quality, but your choices for mATX X-99 boards are pretty slim:
ASRock Fatal1ty X99M Killer
ASRock X99M Extreme4
ASRock Fatal1ty X99M Killer/3.1
EVGA X99 Micro2
Gigabyte GA-X99M-GAMING 5
Asus X99-M WS

 

[FrgMstr]

That is odd did evga get but hurt over a bad product that yall called them out on?

Years ago.

I also had a long conversation with them using Folding for its own advertising needs, which I did not like either. They did not like my thoughts on that.

 

[CrazyElf]

Thanks for the review.

Owning an MSI X99A XPower as my main system (the yellow and black version), I feel like apart from the metal reinforced PCIe sockets, this "Titanium" variant is a downgrade.


On the X99A XPower, you get:
- 2 USB 2.0 ports, 8 USB 3.0 ports, and 2 USB 3.1 ports
- Dual LAN (second one is an Intel i210AT)


One thing I actually like about the non-Titanium version is that you get more USB 3.0 ports (I personally prefer more ports as I have a lot of peripherals).

The X99A XPower had:
- 2x Asmedia ASM 1074 for 8x USB 3.0
- 2x Asmedia ASM 1142 for 2x USB 3.1

against that, this new board has
- 2x Asmedia ASM 1142 (only 1 is a type C port)
- VIA VL805 (needs more USB ports in the back IMO)

Net it is a downgrade.

Both chips have the same main LAN i218V (which is built into the X99 PCH), the Wifi (although the Titanium version has the Intel 8260 versus the Intel 7260 on the X99A XPower yellow version), and an OC Socket.


I guess so in summary you get:
- Steel reinforced PCIe slots
- A U.2 slot, but it cannot be used at the same time with the M.2 slot

But you lose:
- Dual LAN
- More USB 3.0 slots on the back


I kind of find this new refresh lacking.

Things I would have wanted:
- Keep the dual LAN and the same number of USB 3.0 slots
- More USB 3.0 slots as USB 3.1
- A backplate to add rigidity to the board (kind of like the ones they added to the Asus Maximus Formula series); MSI has copied it on the Z170 MPower
- Better DPC latency (could probably be fixed in the BIOS)
- Maybe upgrade the VRM entirely - perhaps DirectFET Mosfets ( IR6811/IR6894 )
- Increase the distance from the PCIe slots between the CPU socket, along with the RAM and the GPU
- More PCB layerse would have been nice (10 or ideally 12 layer)



Edit:
I am worried about how fast that 14nm ES degraded. Do you have any more 6950X that have showed similar behavior? It kind of makes me glad I got a 5960X.

 

[FrgMstr]

I am worried about how fast that 14nm ES degraded. Do you have any more 6950X that have showed similar behavior? It kind of makes me glad I got a 5960X.

We have two, both have done the same thing. Dan has one he uses, I have one I use. Both are ES samples, but I would suggest that retail parts would be no different.

I have a MSI X99A XPower in my personal box right now. Been a great motherboard.

 

[grtitan]

Can't address quality, but your choices for mATX X-99 boards are pretty slim:
ASRock Fatal1ty X99M Killer
ASRock X99M Extreme4
ASRock Fatal1ty X99M Killer/3.1
EVGA X99 Micro2
Gigabyte GA-X99M-GAMING 5
Asus X99-M WS

Thanks for the suggestions, will look into them.

 

[ianken]

OMG what a beautiful piece of hardware. But, if you are not an expert in OC is a waste of money Ithink.

Considering it wasn't stable at stock I'd say it's a waste of money no matter what.

 

[FrgMstr]

I have removed this review for now as I think we did not do some due diligence on our end correctly and have possibly not reported what I would call a "solid" motherboard review. I am pulling all the equipment and parts back into my offices to further investigate this motherboard. Once I figure out what all is going on, I will report back, but at this time I think we are dealing with an equipment failure on our end, and it is not an MSI issue.

 

[Jim Kim]

I'm more worried about the degradation of the Intel Core i7-6950X. Is it because of testing methodogy, the motherboard or the gods. Or worse yet is it a sign of imminent failure.
This is about the only review site that I trust to get honest, in-depth and sometimes even fiery reviews during power supply testing. Keep up the excellent work.

 

[Dan_D]

We overclock the piss out of these CPUs. I will be the first to admit I will push past the safe ranges of the voltages if I think it will give me a solid overclock. That said, the thermals on these CPUs usually make that a non-issue as there isn't any room to handle more heat. 1.35v is the most voltage I've had to use in quite some time. We have spoken to one of the motherboard vendors about the CPU degradation and they have seen it too with several CPUs. I think the issue is design related. To the person who asked if we think this is a 14nm issue, I will say that I do not. Skylake doesn't have this issue thus far.

 

[Murzilka]

Oh man, this cpu just can't stop sucking.

 

[striker444]

I posted your findings in the overclock forums as folks buy CPU's there that are guaranteed to overclock. The owner of Silicon Lottery responded with this

"I have not seen a glimpse of degradation personally, nor have any of my customers mentioned anything. I've even put near 1.5V in these for benching, with no ill effects. They might have a faulty chip there, or a buggy bios."

 

[FrgMstr]

I posted your findings in the overclock forums as folks buy CPU's there that are guaranteed to overclock. The owner of Silicon Lottery responded with this

"I have not seen a glimpse of degradation personally, nor have any of my customers mentioned anything. I've even put near 1.5V in these for benching, with no ill effects. They might have a faulty chip there, or a buggy bios."

There is no doubt in my mind about what I have said about CPU degradation with the Broadwell-E. I would also suggest that we do a lot of long term stress testing with our CPUs as well that many do not. That said, I have talked to one major motherboard builder that confirmed my thoughts when I reached out to them about, they have seen the same exact thing in their testing.

 

[Dan_D]

I posted your findings in the overclock forums as folks buy CPU's there that are guaranteed to overclock. The owner of Silicon Lottery responded with this

"I have not seen a glimpse of degradation personally, nor have any of my customers mentioned anything. I've even put near 1.5V in these for benching, with no ill effects. They might have a faulty chip there, or a buggy bios."

I think the difference boils down to the fact that these hand picked CPUs are stability tested at certain speeds for a couple hours and then sent out the door. The testers probably know exactly how to achieve a given clock on those chips. I doubt many of the customers who buy them actually reach the same speeds and sustain them 24/7. If they can't hit the guaranteed speeds it's probably chalked up to cooling or the motherboard itself. Our CPUs didn't show signs of degradation for several weeks after we started beating on them. Our CPUs get installed in lots of motherboards and get benchmarked to death without adequate cooling on the motherboards a lot of the time. We also might have higher ambient temperatures to deal with. Each motherboard is different and some of the motherboards we use do not necessarily have the best power circuitry. A flaky BIOS, an early BIOS or early PCB design can add extra stress to these CPUs. While we usually try and work with up to date and retail motherboards we don't always have that luxury. Sometimes we are shown early samples of things which again get beaten on.

It's important to note that we have engaged in this behavior for years with dozens of CPUs from AMD and Intel. In all that time we've ever seen degradation occur this quickly. With at least one motherboard manufacturer confirming these issues with a larger CPU sampling size than we have, I feel confident in our assertions concerning Broadwell-E. Let me put it to you this way, if I bought a Core i7 6950X for use in my own personal system I sure as hell wouldn't clock it the way we clock our test CPUs. I wouldn't clock a free CPU this high without some guarantee Intel would replace it if and when I killed it. That's how certain I am that Broadwell-E in it's current form (6950X only) can't handle these overclocks over the long haul.

 

[Ultima99]

Broadwell in all desktop forms is a letdown IMO. Lets hope Skylake-X is a different story.

 

[Dan_D]

Broadwell in all desktop forms is a letdown IMO. Lets hope Skylake-X is a different story.

I think it is. The memory controller isn't really improved from Haswell-E which hurts it. The IPC improvements are very small and we lost too much clock speed to achieve it. You have to run at 4.3GHz for most tasks and often times less than that for AVX based instructions. I found I had to run at -2 most times which gave me 4.1GHz with AVX. Were it not for the extra two cores I think the Core i7 6950X would be worthless. I still argue that it is outside of some very rare situations it's not a great buy. I don't know how the 6930K is so I can't speak to that one. I think those lose clock speed compared to the 5960X as well when overclocked. Depending on price, the older 5960X might be a better buy if you can find one.

 

[Murzilka]

You mean the 6900k, not 6930k?

 

[Quix]

Two 6950Xs bite the dust so soon? That's disappointing.

 

[M76]

I was waiting for Broadwell-e to finally replace my aging sandy bridge, but now it seems I'll continue to wait for skylake-e

 

[Silent Scone]

I think the difference boils down to the fact that these hand picked CPUs are stability tested at certain speeds for a couple hours and then sent out the door. The testers probably know exactly how to achieve a given clock on those chips. I doubt many of the customers who buy them actually reach the same speeds and sustain them 24/7. If they can't hit the guaranteed speeds it's probably chalked up to cooling or the motherboard itself. Our CPUs didn't show signs of degradation for several weeks after we started beating on them. Our CPUs get installed in lots of motherboards and get benchmarked to death without adequate cooling on the motherboards a lot of the time. We also might have higher ambient temperatures to deal with. Each motherboard is different and some of the motherboards we use do not necessarily have the best power circuitry. A flaky BIOS, an early BIOS or early PCB design can add extra stress to these CPUs. While we usually try and work with up to date and retail motherboards we don't always have that luxury. Sometimes we are shown early samples of things which again get beaten on.

It's important to note that we have engaged in this behavior for years with dozens of CPUs from AMD and Intel. In all that time we've ever seen degradation occur this quickly. With at least one motherboard manufacturer confirming these issues with a larger CPU sampling size than we have, I feel confident in our assertions concerning Broadwell-E. Let me put it to you this way, if I bought a Core i7 6950X for use in my own personal system I sure as hell wouldn't clock it the way we clock our test CPUs. I wouldn't clock a free CPU this high without some guarantee Intel would replace it if and when I killed it. That's how certain I am that Broadwell-E in it's current form (6950X only) can't handle these overclocks over the long haul.

Dan, can you tell me whether you are using adaptive or manual when running the stress tests for this review?

 

[illustrator]

You mean the 6900k, not 6930k?

Or a 5930K, or a 4930K, or a 3930K. Or something completely different. So many Ks.

 

[CrazyElf]

I think it is. The memory controller isn't really improved from Haswell-E which hurts it. The IPC improvements are very small and we lost too much clock speed to achieve it. You have to run at 4.3GHz for most tasks and often times less than that for AVX based instructions. I found I had to run at -2 most times which gave me 4.1GHz with AVX. Were it not for the extra two cores I think the Core i7 6950X would be worthless. I still argue that it is outside of some very rare situations it's not a great buy. I don't know how the 6930K is so I can't speak to that one. I think those lose clock speed compared to the 5960X as well when overclocked. Depending on price, the older 5960X might be a better buy if you can find one.

The problem is that Skylake E (or X) will be on the same 14nm process.

I don't see it being much faster if you think about it.

Ivy Bridge E vs Sandy Bridge E would be the most direct comparison, of Broadwell E vs Haswell E. Arguably, from a CPU standpoint, Sandy Bridge E was faster due to the higher possible overclocks. In the case of SBE, the golden chips could reach over 5 GHz at times. Ivy Bridge E was slightly faster in IPC, but net was slower due to the slower clock speeds after overclocked. Ivy Bridge E was looking at 4.4 - 4.6 GHz, with golden chips reaching 4.7 GHz (at 24-7 voltages).

Broadwell E is looking similar compared to Ivy Bridge E. The problem is that factoring in the degradation and the lack of IMC improvements, it is a worse deal. Ivy Bridge, although it did not bring much in the way of single threaded performance, at least brought improvements in the way of memory controllers.

Now we compare Haswell E vs Ivy Bridge E. Haswell E got very similar clocks - 4.4 to 4.6 GHz, with golden chips reaching 4.7 GHz. The addition of FIVR did not seem to slow clocks down, both on the HEDT and mainstream platform (a 3770k and 4790k have similar maximum clocks). It would suggest the process is the bottleneck.

This would suggest that we can already extrapolate Skylake E performance. Clocks should be comparable to Broadwell E, 4.2 to 4.4 Ghz, with a few golden chips hitting 4.5 GHz. Considering Skylake's IPC improvements, that is only a slight upgrade over Haswell E. Memory overclocking should see better improvements though, if desktop Skylake is anything to go by.

One big question is, will Skylake E degrade as rapidly as Broadwell E? Are CPUs getting more "voltage sensitive" with smaller nodes? In other words, is say, 1.45V a safe 24-7 overclock at 14nm with both Z170 and X190 (or whatever they call it)?


Edit:
I guess the sad part is that the platform is more exciting than the CPU itself. Skylake E will no doubt bring DMI 3 to the HEDT platform, where it is really needed. Arguably with the proliferation of M.2 drives, even DMI 3.0 might not be enough and a wider bus should be considered (ex: with double the GT/s).

 

[Silent Scone]

The problem is that Skylake E (or X) will be on the same 14nm process.

I don't see it being much faster if you think about it.

Ivy Bridge E vs Sandy Bridge E would be the most direct comparison, of Broadwell E vs Haswell E. Arguably, from a CPU standpoint, Sandy Bridge E was faster due to the higher possible overclocks. In the case of SBE, the golden chips could reach over 5 GHz at times. Ivy Bridge E was slightly faster in IPC, but net was slower due to the slower clock speeds after overclocked. Ivy Bridge E was looking at 4.4 - 4.6 GHz, with golden chips reaching 4.7 GHz (at 24-7 voltages).

Broadwell E is looking similar compared to Ivy Bridge E. The problem is that factoring in the degradation and the lack of IMC improvements, it is a worse deal. Ivy Bridge, although it did not bring much in the way of single threaded performance, at least brought improvements in the way of memory controllers.

Now we compare Haswell E vs Ivy Bridge E. Haswell E got very similar clocks - 4.4 to 4.6 GHz, with golden chips reaching 4.7 GHz. The addition of FIVR did not seem to slow clocks down, both on the HEDT and mainstream platform (a 3770k and 4790k have similar maximum clocks). It would suggest the process is the bottleneck.

This would suggest that we can already extrapolate Skylake E performance. Clocks should be comparable to Broadwell E, 4.2 to 4.4 Ghz, with a few golden chips hitting 4.5 GHz. Considering Skylake's IPC improvements, that is only a slight upgrade over Haswell E. Memory overclocking should see better improvements though, if desktop Skylake is anything to go by.

One big question is, will Skylake E degrade as rapidly as Broadwell E? Are CPUs getting more "voltage sensitive" with smaller nodes? In other words, is say, 1.45V a safe 24-7 overclock at 14nm with both Z170 and X190 (or whatever they call it)?


Edit:
I guess the sad part is that the platform is more exciting than the CPU itself. Skylake E will no doubt bring DMI 3 to the HEDT platform, where it is really needed. Arguably with the proliferation of M.2 drives, even DMI 3.0 might not be enough and a wider bus should be considered (ex: with double the GT/s).

What makes you think these degrade, or even that the memory controller is not improved?

 

[Dan_D]

You mean the 6900k, not 6930k?

Correct. All the K's get confusing and the last generation had a x930K nomenclature to denote it was right behind the top end CPU.

Dan, can you tell me whether you are using adaptive or manual when running the stress tests for this review?

Manual normally for the sake of speed. Unfortunately I don't get that much time to dial these motherboards in. Once I dial in what a given CPU needs, it is a lot faster to simply punch in the settings I know work. That said, I do use adaptive and offset voltage adjustments to see if the behavior of a given motherboard is better that way. In some cases a manufacturer recommends that. Adaptive voltages can also be tricky with using Prime95 and AIDA64 for stress testing. Too much voltage sometimes gets used with adaptive mode. Its often best to use offset voltages for that which limit the amount of voltage that the VR will output.

Or a 5930K, or a 4930K, or a 3930K. Or something completely different. So many Ks.

The problem is that Skylake E (or X) will be on the same 14nm process.

I don't see it being much faster if you think about it.

Ivy Bridge E vs Sandy Bridge E would be the most direct comparison, of Broadwell E vs Haswell E. Arguably, from a CPU standpoint, Sandy Bridge E was faster due to the higher possible overclocks. In the case of SBE, the golden chips could reach over 5 GHz at times. Ivy Bridge E was slightly faster in IPC, but net was slower due to the slower clock speeds after overclocked. Ivy Bridge E was looking at 4.4 - 4.6 GHz, with golden chips reaching 4.7 GHz (at 24-7 voltages).

Broadwell E is looking similar compared to Ivy Bridge E. The problem is that factoring in the degradation and the lack of IMC improvements, it is a worse deal. Ivy Bridge, although it did not bring much in the way of single threaded performance, at least brought improvements in the way of memory controllers.

Now we compare Haswell E vs Ivy Bridge E. Haswell E got very similar clocks - 4.4 to 4.6 GHz, with golden chips reaching 4.7 GHz. The addition of FIVR did not seem to slow clocks down, both on the HEDT and mainstream platform (a 3770k and 4790k have similar maximum clocks). It would suggest the process is the bottleneck.

This would suggest that we can already extrapolate Skylake E performance. Clocks should be comparable to Broadwell E, 4.2 to 4.4 Ghz, with a few golden chips hitting 4.5 GHz. Considering Skylake's IPC improvements, that is only a slight upgrade over Haswell E. Memory overclocking should see better improvements though, if desktop Skylake is anything to go by.

One big question is, will Skylake E degrade as rapidly as Broadwell E? Are CPUs getting more "voltage sensitive" with smaller nodes? In other words, is say, 1.45V a safe 24-7 overclock at 14nm with both Z170 and X190 (or whatever they call it)?


Edit:
I guess the sad part is that the platform is more exciting than the CPU itself. Skylake E will no doubt bring DMI 3 to the HEDT platform, where it is really needed. Arguably with the proliferation of M.2 drives, even DMI 3.0 might not be enough and a wider bus should be considered (ex: with double the GT/s).

It isn't that simple. Process nodes may have inherent problems or advantages, but that's only part of the story. Other things in the architecture can maximize or inhibit CPU clocks. Deeper pipeline stages for example have been used to increase the clocks of a CPU. AMD and Intel have both done this from time to time. We really can't extrapolate too much about Skylake-E from Haswell-E. We can extrapolate more from Skylake than anything else. This suggests that we will see slightly higher clocks than we are getting with Broadwell-E. I hate trying to predict the future but if I had to guess, I'd bet we'll get back up to 4.5GHz or thereabouts with it depending on the final core count of that processor line. I think Broadwell-E only clocks so badly because of the extra cores. Skylake is on the same 14nm process as Broadwell and it clocks as well as Devil's Canyon does. I don't think the process node by itself is an indicator of the CPUs being more voltage sensitive. We aren't seeing the same degradation on Broadwell or Skylake.

I do agree that the platform may be where the excitement is for Skylake-E. X99 was a good step in the right direction, but I think the next platform will be the one we've truly been waiting for in the HEDT segment. That is *IF* everything that goes into the chipset makes the final cut for it. As we know, that doesn't always happen. SAS was dropped from X79 for example despite it being a native element of that chipset.

What makes you think these degrade, or even that the memory controller is not improved?

Broadwell-E definitely has issues with degradation. Not only have we seen this with our chips, but one manufacturer confirmed they've seen it DOZENS of times. As for the memory controller, we can tell it wasn't improved. Broadwell-E doesn't benefit from the higher memory frequencies that Skylake does in games and it doesn't clock any higher with RAM speeds than Haswell-E does. That to me suggests a lack of improvement. Having said that, the limitations we see with regard to memory clocks may also lie with X99 itself in whole or in part. It's difficult to tell. I don't think its a motherboard design issue or the X99 refresh motherboards would be vastly superior to the X99 launch motherboards and that just isn't the case.

 

[Silent Scone]

Correct. All the K's get confusing and the last generation had a x930K nomenclature to denote it was right behind the top end CPU.



Manual normally for the sake of speed. Unfortunately I don't get that much time to dial these motherboards in. Once I dial in what a given CPU needs, it is a lot faster to simply punch in the settings I know work. That said, I do use adaptive and offset voltage adjustments to see if the behavior of a given motherboard is better that way. In some cases a manufacturer recommends that. Adaptive voltages can also be tricky with using Prime95 and AIDA64 for stress testing. Too much voltage sometimes gets used with adaptive mode. Its often best to use offset voltages for that which limit the amount of voltage that the VR will output.





It isn't that simple. Process nodes may have inherent problems or advantages, but that's only part of the story. Other things in the architecture can maximize or inhibit CPU clocks. Deeper pipeline stages for example have been used to increase the clocks of a CPU. AMD and Intel have both done this from time to time. We really can't extrapolate too much about Skylake-E from Haswell-E. We can extrapolate more from Skylake than anything else. This suggests that we will see slightly higher clocks than we are getting with Broadwell-E. I hate trying to predict the future but if I had to guess, I'd bet we'll get back up to 4.5GHz or thereabouts with it depending on the final core count of that processor line. I think Broadwell-E only clocks so badly because of the extra cores. Skylake is on the same 14nm process as Broadwell and it clocks as well as Devil's Canyon does. I don't think the process node by itself is an indicator of the CPUs being more voltage sensitive. We aren't seeing the same degradation on Broadwell or Skylake.

I do agree that the platform may be where the excitement is for Skylake-E. X99 was a good step in the right direction, but I think the next platform will be the one we've truly been waiting for in the HEDT segment. That is *IF* everything that goes into the chipset makes the final cut for it. As we know, that doesn't always happen. SAS was dropped from X79 for example despite it being a native element of that chipset.

Broadwell-E definitely has issues with degradation. Not only have we seen this with our chips, but one manufacturer confirmed they've seen it DOZENS of times. As for the memory controller, we can tell it wasn't improved. Broadwell-E doesn't benefit from the higher memory frequencies that Skylake does in games and it doesn't clock any higher with RAM speeds than Haswell-E does. That to me suggests a lack of improvement. Having said that, the limitations we see with regard to memory clocks may also lie with X99 itself in whole or in part. It's difficult to tell. I don't think its a motherboard design issue or the X99 refresh motherboards would be vastly superior to the X99 launch motherboards and that just isn't the case.

Broadwell-E's IMC has improved. This was not only noted in ASUS' overclocking overview, but several users over at Overclock.net including myself have established that a lot of these processor are capable of running exceptionally tight sub timings, close to or at the minimal spacing required at speeds of 3200Mhz, as well as 3400Mhz being far more easily obtained at an operation level. There were very few Haswell-E CPU capable of such a feat. I guess you could argue in terms of raw frequency things have not improved, but that's not the same as saying things haven't changed at all.


As for degradation, it would depend on under what conditions the manufacturer is seeing this. You noted in the review that AVX offset was used, and we've established you were using a manual voltage. If running tests such as Prime in conjunction with the AVX offset, it is mandatory to leave the voltage function in Auto. If it is in Manual like you are using, then the drop in frequency is all you will be getting, and not the concurrent voltage drop that you would whilst in Auto.

Needless to say, it's fairly pointless doing things this way. If you are applying upward of 1.3 to 1.35v and running Prime under this condition, then the current will still be too great. This is not really any different to Haswell-E, the same conditions were just as damning in tow with similar voltages.

I guess ultimately it depends on what the user is doing for the degradation to occur. Most will not be hammering the CPU in this way. Or if they are, they should probably be aware of the amount of current passing through the CPU.

 

[Dan_D]

Broadwell-E's IMC has improved. This was not only noted in ASUS' overclocking overview, but several users over at Overclock.net including myself have established that a lot of these processor are capable of running exceptionally tight sub timings, close to or at the minimal spacing required at speeds of 3200Mhz, as well as 3400Mhz being far more easily obtained at an operation level. There were very few Haswell-E CPU capable of such a feat. I guess you could argue in terms of raw frequency things have not improved, but that's not the same as saying things haven't changed at all.


As for degradation, it would depend on under what conditions the manufacturer is seeing this. You noted in the review that AVX offset was used, and we've established you were using a manual voltage. If running tests such as Prime in conjunction with the AVX offset, it is mandatory to leave the voltage function in Auto. If it is in Manual like you are using, then the drop in frequency is all you will be getting, and not the concurrent voltage drop that you would whilst in Auto.

Needless to say, it's fairly pointless doing things this way. If you are applying upward of 1.3 to 1.35v and running Prime under this condition, then the current will still be too great. This is not really any different to Haswell-E, the same conditions were just as damning in tow with similar voltages.

I guess ultimately it depends on what the user is doing for the degradation to occur. Most will not be hammering the CPU in this way. Or if they are, they should probably be aware of the amount of current passing through the CPU.

Fair enough. I actually didn't put two and two together on the manual voltage and AVX offset but that makes sense. There is always a learning curve when we start messing with a new CPU. I know I have pushed that CPU far harder than it probably should have been. You'd be surprised how little information we have to go on when we first get these CPUs and start beating on them. I burned a GIGABYTE motherboard when the Phenom chips first came out because AMD told us next to nothing about the chips and what voltages were safe. That said, my CPU isn't an isolated example. We never know where the safe zones are besides what Intel or a motherboard manufacturer tells us. I've used well beyond 1.35v with that CPU many times. This may very well have and probably did contribute to the degradation. Manual voltage settings have often been the quick and dirty way to setup a system for overclocking. Not only does this save us time but it also allows us to more or less use the same values on all motherboards to see if there is a difference in overclockability between them. It provides a good apples to apples baseline. We rarely see too much differentiation in motherboard overclockability these days. Its interesting to see when some motherboards won't use settings that another will run perfectly well with. We mostly see differences in how much vCore has to be set to run at 4.3GHz or 4.5GHz or how much CPU input voltage is required to do the same things.

I think sometimes that people are confused by the overclocking goals. We are here to push the motherboard primarily, and see if there is a difference in what one can achieve with a given CPU across multiple motherboards. I am not trying to max out the CPU specifically. It's nice when I can, but it isn't our primary goal. Our goal is to isolate motherboard overclocking capability and motherboard quality while stressing its components. I do try offset and adaptive modes with varying degrees of success on these motherboards. Generally, I can achieve the same results by setting the CPU vCore to manual and being done with it.

 

[Silent Scone]

Fair enough. I actually didn't put two and two together on the manual voltage and AVX offset but that makes sense. There is always a learning curve when we start messing with a new CPU. I know I have pushed that CPU far harder than it probably should have been. You'd be surprised how little information we have to go on when we first get these CPUs and start beating on them. I burned a GIGABYTE motherboard when the Phenom chips first came out because AMD told us next to nothing about the chips and what voltages were safe. That said, my CPU isn't an isolated example. We never know where the safe zones are besides what Intel or a motherboard manufacturer tells us. I've used well beyond 1.35v with that CPU many times. This may very well have and probably did contribute to the degradation. Manual voltage settings have often been the quick and dirty way to setup a system for overclocking. Not only does this save us time but it also allows us to more or less use the same values on all motherboards to see if there is a difference in overclockability between them. It provides a good apples to apples baseline. We rarely see too much differentiation in motherboard overclockability these days. Its interesting to see when some motherboards won't use settings that another will run perfectly well with. We mostly see differences in how much vCore has to be set to run at 4.3GHz or 4.5GHz or how much CPU input voltage is required to do the same things.

I think sometimes that people are confused by the overclocking goals. We are here to push the motherboard primarily, and see if there is a difference in what one can achieve with a given CPU across multiple motherboards. I am not trying to max out the CPU specifically. It's nice when I can, but it isn't our primary goal. Our goal is to isolate motherboard overclocking capability and motherboard quality while stressing its components. I do try offset and adaptive modes with varying degrees of success on these motherboards. Generally, I can achieve the same results by setting the CPU vCore to manual and being done with it.

No worries! The Asus Edge guide for the Thermal Control Tool covers the conditions that AVX offset needs to be used

How to get the best performance from Broadwell-E processors - The ASUS Thermal Control Tool - Edge Up

 

[FrgMstr]

Going back up with a redraft with a BIG FYI.

This is the second version of this review. The original review we published had some issues pointed out that were not MSI's problems. We identified two parts on our testing hardware that were having issues. I brought Dan's entire test bench back to my office where we could start isolating issues, and it came down to RAM and power. I had no issues getting the MSI X99A XPOWER GAMING TITANIUM motherboard stable at stock clocks with an Intel Core-i7 6950X processor as was previously stated. Our sincere apologies to MSI for this error in our review process.

 

[CrazyElf]

Thanks for the re-upload. The AVX negative offset seems like the only "cool" feature of this new generation of CPUs. It may be useful for certain applications.

At least now, the X99A XPower Titanium is probably not to blame for not being stable at stock. Feature wise though, yeah, in some ways this board is still kind of underwhelming for being a flagship (granted the very top now for MSI is the X99A Godlike Carbon).

As discussed, I am hoping to see the yellow/black come back and with a crazy VRM (it would have to be XL ATX like on the X79 XPower in order for there to be enough room).

It isn't that simple. Process nodes may have inherent problems or advantages, but that's only part of the story. Other things in the architecture can maximize or inhibit CPU clocks. Deeper pipeline stages for example have been used to increase the clocks of a CPU. AMD and Intel have both done this from time to time. We really can't extrapolate too much about Skylake-E from Haswell-E. We can extrapolate more from Skylake than anything else. This suggests that we will see slightly higher clocks than we are getting with Broadwell-E. I hate trying to predict the future but if I had to guess, I'd bet we'll get back up to 4.5GHz or thereabouts with it depending on the final core count of that processor line. I think Broadwell-E only clocks so badly because of the extra cores. Skylake is on the same 14nm process as Broadwell and it clocks as well as Devil's Canyon does. I don't think the process node by itself is an indicator of the CPUs being more voltage sensitive. We aren't seeing the same degradation on Broadwell or Skylake.

The 6900K from what we can tell, despite have 8 cores, isn't doing any better than the 10 core 6950k. The 6800k and 6850k, with 6 cores, seem to average about 100 MHz higher on average.

So it seems like:
- Big penalty from 4 to 6 core
- 6 core to 8/10 core seem to have a small penalty
- 8 to 10 core penalty is probably very small

Were it possible, it'd be very interesting to see how the giant HCC dies would compare in overclocking. Sadly the 2699v4 is locked down tight. Perhaps with Skylake E if there is a way to port over the Sky OC, that might happen.

Yet even looking in comparison to Skylake, I still expect Skylake E to be in the clocks I've specified though. 4.3 to 4.4 GHz for most. Lucky chips at 4.5 GHz and the very best at 4.6 Ghz. That's about 200 MHz slower than Skylake, and if you consider Haswell E vs Devil's Canyon, you are looking at perhaps 200 MHz on average slower, maybe a bit more (say 200 - 300 MHz slower on average).

I just hope that you are right and that I am wrong about Skylake E. I'd hate to see Skylake E chips being prone to degradation.

I do agree that the platform may be where the excitement is for Skylake-E. X99 was a good step in the right direction, but I think the next platform will be the one we've truly been waiting for in the HEDT segment. That is *IF* everything that goes into the chipset makes the final cut for it. As we know, that doesn't always happen. SAS was dropped from X79 for example despite it being a native element of that chipset.

I was disappointed when I learned that desktop Skylake E would not be getting the 6 channel RAM.

Sadly we get (and take this of course with a grain of salt):
- 48 PCI-E 3.0 lanes (up 8 from 40 for Haswell E)
- 10 USB 3.0 ports (X99 has 6, and one big disappointment is that there is no native USB 3.1 support)
- 2 integrated ports and I think to 8 SATA 3.0 (unchanged from X99)
- Quad channel RAM, although no doubt the RAM speeds will be faster

I hope that we can get comparable speeds and timings to Skylake, only in quad channel.

Broadwell-E's IMC has improved. This was not only noted in ASUS' overclocking overview, but several users over at Overclock.net including myself have established that a lot of these processor are capable of running exceptionally tight sub timings, close to or at the minimal spacing required at speeds of 3200Mhz, as well as 3400Mhz being far more easily obtained at an operation level. There were very few Haswell-E CPU capable of such a feat. I guess you could argue in terms of raw frequency things have not improved, but that's not the same as saying things haven't changed at all.

The extent of the improvements though, are disappointing compared to prior generations. Ivy Bridge for example, despite being a node shrink of Sandy Bridge saw much better speeds.

As Dan has noted, it may be less significant seeing that Broadwell E doesn't benefit and perhaps a limitation with the X99 platform. In that case, we'd best hope that X190 brings improvements in this area.