Wednesday, July 5th 2017

Update on the Intel X299 Platform "VRM Disaster"

We have some updated information on the X299 Platforms VRM issues from the same overclocker who initially discovered the issue, renowned overclocker der8auer. In an updated YouTube video, der8auer first updated his viewers with new information on his testing techniques, and basically concluded that all issues initially detected (throttling included) are still is an issue even after extensive testing, only in some instances it is difficult to detect not only if you are throttling, but even specifics such as what precisely is throttling. He goes into extensive detail, but a brief summary of the videos main points can be found below for your consumption.
His first major point is to watch power consumption. If it drops drastically, something is probably throttling. It does not mean (as some critics assumed) that the X299 system is not as energy hungry as reported: It is, it just has the nasty habit of not always reporting a throttle in clock rate. Furthermore, throttles can happen "back and forth" very quickly, and CPU-Z's refresh rate is not fast enough to always catch these. He recommends using HwInfo for frequency and VRM temp monitoring purposes.

Second, once you know there is a throttle occurring, there are two major types of throttling. The first is CPU throttling, which will occur via the multiplier backing off on the CPU cores slowly. The second is VRM throttling, which will manifest itself as a sharp drop in multiplier instantly (in his example, it went from 4.5 GHz to 1.2 GHz during a VRM throttle). These dramatic drops are related to VRMs and the poor build issues on various boards. Sadly, as of right now these issues still exist as a very complex minefield for potential buyers to navigate.

Third, he backed off his criticism of the 8-pin connectors saying they were probably okay for "moderate overclocking" (being that he is an extreme overclocker, that's probably enough for most of us "24/7ers"). The initial issue of overheating cables appeared to be more the cable design of the Superflower PSU than an actual limit of the 8-pin CPU connector. He still says it is better to get an 8+4 or 8+8 pin board if pursuing "serious overclocks", however.

Oh, and just to make it more confusing, in several UEFIs you can raise the throttle-temp point for the CPU (though most 24/7 overclockers probably shouldn't) and on some boards, even more throttling types are present (GIGABYTE seems to have some kind of "VCCIN throttle" that kicks in randomly if left on auto).

The end result of all this? I would say it's a very confusing and very strange product landscape, the likes of which is not likely to be friendly to overclockers, neither casual nor extreme. Pick your boards carefully dear user, perhaps this time even more so than most.Source: YouTube user "der8auer"
Add your own comment

43 Comments on Update on the Intel X299 Platform "VRM Disaster"

#1
FR@NK
Glad to see he updated this "Disaster" would only happen under very worst case scenario.
Posted on Reply
#2
R-T-B
FR@NK said:
Glad to see he updated this "Disaster" would only happen under very worst case scenario.
There are so many ways to be throttled if you watch the video, I'm not quite certain we can say that. Worse, you don't even know you are being throttled half the time. It's as confusing as it could be if they were trying to go for an intentionally confusing product.
Posted on Reply
#3
Aenra
Not often that numbers alone can be so amusing. Many thanks to all the "techs" and "engineers" at Intel for this.
Posted on Reply
#4
RejZoR
And people were going batshit crazy over poor initial RAM support with Ryzen (which can, is and will be improved via BIOSes). How do you improve a broken VRM segment on a motherboard? I bet AMD is chuckling like an evil mastermind in a volcano and Intel is still in full panic mode...
Posted on Reply
#5
Maxx_Power
A lot of these VRM heatsinks today appear to be poorly thermally engineered (might as well include some M.2 heatsinks). They appear to be nothing more than a slap of extruded aluminum carelessly slapped on the board. Then they add more decals over that, which in many cases, are not meant to be removed. I doubt that any kind of thermal modelling was ever done to optimize anything.
Posted on Reply
#6
R-T-B
Maxx_Power said:
A lot of these VRM heatsinks today appear to be poorly thermally engineered (might as well include some M.2 heatsinks). They appear to be nothing more than a slap of extruded aluminum carelessly slapped on the board. Then they add more decals over that, which in many cases, are not meant to be removed. I doubt that any kind of thermal modelling was ever done to optimize anything.
This is my assesment as well. They seem more concerned with the mobos looking like mechs or something than making actually fuctionally good heatsinks.
Posted on Reply
#7
evernessince
R-T-B said:
This is my assesment as well. They seem more concerned with the mobos looking like mechs or something than making actually fuctionally good heatsinks.
That's because they haven't needed to do serious cooling on the VRM in a long time. Why should they bother when Intel CPUs have sipped power for so long. Then bam, x299 hits and all of a sudden Intel CPUs are power hogs again.

You know what they say, when you get older you forget the lessons you learned earlier in life and have to learn them all over again.
Posted on Reply
#8
Caelestis
R-T-B said:
This is my assesment as well. They seem more concerned with the mobos looking like mechs or something than making actually fuctionally good heatsinks.
In my opinion this is a result of the side window trend. Everyone wants to show of his/her hardware and mobo manufacturers are following this trend while forgetting about "Form follows function".
Posted on Reply
#9
dwade
RejZoR said:
And people were going batshit crazy over poor initial RAM support with Ryzen (which can, is and will be improved via BIOSes). How do you improve a broken VRM segment on a motherboard? I bet AMD is chuckling like an evil mastermind in a volcano and Intel is still in full panic mode...
Quality of VRM is not Intel's responsibility. RAM support is AMD's responsibility.
Posted on Reply
#10
R0H1T
RejZoR said:
And people were going batshit crazy over poor initial RAM support with Ryzen (which can, is and will be improved via BIOSes). How do you improve a broken VRM segment on a motherboard? I bet AMD is chuckling like an evil mastermind in a volcano and Intel is still in full panic mode...

Despicable :pimp:
dwade said:
Quality of VRM is not Intel's responsibility. RAM support is AMD's responsibility.
Every Ryzen supports RAM for its rated speeds with 2667 MHz being the max in dual channel, unless we're talking about (Intel) XMP which even Intel doesn't guarantee for all its processors.
Posted on Reply
#11
RejZoR
dwade said:
Quality of VRM is not Intel's responsibility. RAM support is AMD's responsibility.
Yes it is Intel's responsibility. They hand over the specifications. EXACT specifications. Motherboard makers can beef it up, but they can't dumb it down below what Intel specifies. Problem with X299 is that they are ranging from quad cores up to what's max, 18 cores? Things tend to go wrong when you do this oposed to having only beefy CPU's on the platform like X99 had...
Posted on Reply
#12
Hood
RejZoR said:
Yes it is Intel's responsibility. They hand over the specifications. EXACT specifications. Motherboard makers can beef it up, but they can't dumb it down below what Intel specifies. Problem with X299 is that they are ranging from quad cores up to what's max, 18 cores? Things tend to go wrong when you do this oposed to having only beefy CPU's on the platform like X99 had...
Do you really think they would make that kind of mistake? It might be because Intel doesn't recommend overclocking, and never has. They allow overclocking on some CPUs, due to demand from enthusiasts, but never recommend it. Board partners are free to do whatever they want, like selling special boards with beefed-up components designed to handle high OCs, and also sell cheap boards with components barely adequate for a mild OC. As we have all experienced with low budget builds. Yes, it's disappointing that the first round of X299 boards are not great overclockers. I'm sure that better boards are coming, and probably some that are even worse (and cheaper), to match up with the lower end X299 CPUs. In the end, it's up to the board partners to sell what customers want.
This so-called "VRM disaster" thing is just a sign of the times, whining about anything that will generate clicks.
Posted on Reply
#13
RejZoR
Just because overclocker is commenting this, I dont think it's OC related. As far as I know the issue was with stock, but high core count CPU's...
Posted on Reply
#14
erixx
I removed all "art decó" from my mobo's headsinks, as they are all very hot. Glued on bugs... Keep them in the cardbox for future selling....
Posted on Reply
#15
Aretak
dwade said:
Quality of VRM is not Intel's responsibility. RAM support is AMD's responsibility.
That's nonsense, frankly. Even putting aside the fact that both AMD and Intel give very conservative figures for RAM speed support with their CPUs, with anything above that essentially being an overclock that's entirely out of their hands, board and BIOS quality have a huge impact on supported RAM speeds (again for both for AMD and Intel). That's the entire reason why board manufacturers maintain their own very specific QVL lists for RAM.
Posted on Reply
#16
cdawall
where the hell are my stars
RejZoR said:
Just because overclocker is commenting this, I dont think it's OC related. As far as I know the issue was with stock, but high core count CPU's...
There is no issue with vrm temps at stock. Actually when using a real case with airflow I saw no issues at 4.7 either on the Asus x299a
Posted on Reply
#17
EarthDog
cdawall said:
There is no issue with vrm temps at stock. Actually when using a real case with airflow I saw no issues at 4.7 either on the Asus x299a
I was good to 4.5 GHz 10c/20t on ASUS PRime X299 Deluxe as well. VRM sink was hot there, but I didn't catch any throttling. I will double check with HWinfo though. :)

As far as what can be done... vendors can make a better VRM sink and make a v2 or offer it up for a trade in/make us pay for it.

As it stands, these are good at stock in a case with some airflow. Overclocking seems to change things dramatically depending on the overclock.
Posted on Reply
#18
GhostRyder
Sounds similar to how the FX 9590 and 9370 were back on AMD. Only a few boards were rated to actually support it because of high power consumption. Moral of the story is, if your going to buy the very high end chips (Lets say 8 core +) then you should spend on a higher board or wait especially if overclocking. Even so, I bet better boards will come with the issues resolved. Though to me, this is exactly why I think having a huge range of core options on a platform like that is a foolish idea depending on your power consumption range. I still don't get why there are 4 core models again on that platform...
Posted on Reply
#19
EarthDog
It was all of AMD's FX octo and even hex cores on some boards...
Posted on Reply
#21
cdawall
where the hell are my stars
EarthDog said:
I was good to 4.5 GHz 10c/20t on ASUS PRime X299 Deluxe as well. VRM sink was hot there, but I didn't catch any throttling. I will double check with HWinfo though. :)

As far as what can be done... vendors can make a better VRM sink and make a v2 or offer it up for a trade in/make us pay for it.

As it stands, these are good at stock in a case with some airflow. Overclocking seems to change things dramatically depending on the overclock.
This was a particularly good chip 4.7ghz blender stable at 1.25v on just a noctua U14S and a second fan. VRM was a bit warmer than I would prefer, but considering I still run an FX9370 in something I am no stranger to this.
Posted on Reply
#22
Bayonet
There are specific situations under which this overheating occurs, I recommend anyone interested also watch Tom from OC3D's YouTube video as he was in collaboration with der8auer in investigating the issue.

But to summarise, to reach a CPU power draw of ~400w requires (at least on ASUS boards) a power limit increase and the disablement of (somewhat hidden) throttling mechanisms in BIOS, combined with running Prime95 with small (12k) FFTs. This results in VRM temperatures greater than 100°C unless assisted by direct active cooling. Basically the heatsinks are not good enough on X299 motherboards.
Posted on Reply
#23
cadaveca
My name is Dave
Bayonet said:
But to summarise, to reach a CPU power draw of ~400w requires (at least on ASUS boards) a power limit increase and the disablement of (somewhat hidden) throttling mechanisms in BIOS, combined with running Prime95 with small (12k) FFTs. This results in VRM temperatures greater than 100°C unless assisted by direct active cooling. Basically the heatsinks are not good enough on X299 motherboards.
The platform is only rated to 300W. So it is only those that are ocing to the limit with delidding and/or doing extreme benching that this will really affect. My 7900X CPUs draw less than 280W @ 4.7 GHz, and hitting 400W isn't possible with most traditional cooling methods so far; the CPU will throttle and overheat before the VRM does. :p That's also exceeding the CPU's power consumption, for a 140W CPU, by almost 200%.

Of course, I'm looking at CPU power draw over the EPS connectors only, not full system, so anything like this being a serious issue would have been obvious right away, since I see real-time power draw over those wires. AFAIK, I'm the only person that does this... most reviewers report full system power for power consumption.

Isn't this the same dude that said we should push to 5 GHz and delid? (I know that it is, but... perspective).
Posted on Reply
#24
cdawall
where the hell are my stars
cadaveca said:
The platform is only rated to 300W. So it is only those that are ocing to the limit with delidding and/or doing extreme benching that this will really affect. My 7900X CPUs draw less than 280W @ 4.7 GHz, and hitting 400W isn't possible with most traditional cooling methods so far; the CPU will throttle and overheat before the VRM does. :p That's also exceeding the CPU's power consumption, for a 140W CPU, by almost 200%.

Of course, I'm looking at CPU power draw over the EPS connectors only, not full system, so anything like this being a serious issue would have been obvious right away, since I see real-time power draw over those wires. AFAIK, I'm the only person that does this... most reviewers report full system power for power consumption.

Isn't this the same dude that said we should push to 5 GHz and delid? (I know that it is, but... perspective).
My test chip can do 5ghz with the top on it on air, unstable easily :roll: However the minute you pump AVX through it the temps shoot up and it restarts. Power consumption was pretty high at that point. I think high end water will easily do 5ghz without a delid...but power consumption will be through the roof.
Posted on Reply
#25
nem..
:peace: :lovetpu:
Posted on Reply
Add your own comment