Haswell undervolting Software for Windows/Linux?

[zir_blazer]

Due to the reasons stated in this Thread about my obsession with VT-d support, I'm seriously considering purchasing a proper, Workstation-class Supermicro X10SAT Motherboard with the C226 Chipset, to be paired with a Xeon E3-1200 V3 series. Such type of Motherboards usually comes with ZERO options in the BIOS to allow you to run out-of-spec, and that effectively means no chance to change in the BIOS the Multiplier or Voltage, effectively forcing me to run everything at default settings.
Yes, I know that the Xeon doesn't come with a fully Locked Multiplier and bla bla bla. However, at the very least, I would want:

1 - Manual Multiplier control: Mostly for manual underclock. Besides, it should still be possible to raise the Multiplier to the maximum that Turbo uses. On a Xeon E3-1245 V3, that means that I could leave Frequency fixed at 3.8 GHz, instead of 3.4 GHz with Turbo kicking in depending on load, for more consistency.

2 - Manual Voltage control: Because if I can't go above certain Frequency, I would want to be able to go for the lowest possible Voltage than that Frequency can be rock solid at, for the best possible power efficiency. And considering that I shouldn't be able to go more than 400 MHz above nominal Frequency, I should have some interesing headroom to reduce Voltage, and thus power consumption, and also Fan speed. This would also allow me to go for below-default values for Idle. Indeed, for power efficiency, undervolting is a must.

On some previous generations, there were tools that allowed you control of many values from Windows. At least on AMD, I used, and still use, Central Brain Identifier and K10Stat for both things on an A64 and my current AIIX4. So far, I don't know if on Intel side there are tools for that. At least, when I tried to undervolt a Mobile Sandy Bridge from a Notebook I purchased some months ago, I heared that there were no Windows tools to do that, and obviously Notebooks BIOSes doesn't gives you these options, neither many Workstation Motherboards BIOSes does.

In order to use such a Motherboard and still be able to tweak it to my tastes, I need to have control of those two things. Is there any Software based tool for Haswell to do this? Or I NEED to do in from BIOS?

 

[Sequences123]

Modern CPU's already come with sleep states, power down, and underclock/volt when idle. If you're doing this for the sole purpose of saving power when idling, you might be better served with just turning the system off (or if money is an issue, reconsider making such a purchase completely).

 

[zir_blazer]

Full Load power matters more than Idle. You can manually get values much more tight for both Idle and Full Load that whatever automatic power saving features can, because they use conservative values.

 

[fastamdman]

Intel uses extremely conservative voltages for idle on haswell as is. These things barely pull any power at idle. As far as load voltages go, you can set a fixed voltage at the lowest stable point for the over clock or stock clock you are using in the bios. No special software is needed.

 

[zir_blazer]

As far as load voltages go, you can set a fixed voltage at the lowest stable point for the over clock or stock clock you are using in the bios. No special software is needed.

Re-read the first Post. Not every BIOS allows you to do that, specifically the Motherboard I have on sight. That's why I need Software.

 

[Ed1]

Going to be hard for anyone to give you supported app on that MB unless they trying same thing . Have you tried Intel Extreme tuning utility (XTU)

 

[Blastman]

These days a lot of motherboards come with utilities that can overclock and adjust voltages CPU fan speed etc, Check the manufacturer site of your potential m/b to see what they have.

 

[know of fence]

This is how I picture it, you pick the lowest specifiable frequency for undervolting, say 800 MHz then test the required voltages, then repeat for different temperatures in perhaps 5°C steps for a range between ambient and Tmax for that frequency. Then do this for individual cores...

... and this is how you beat conservative pre-sets in electronics, that adjusts different power states a hundred - or was it a thousand - times a second.

The result is a large spreadsheet/matrix of values, that could be avoided/ trimmed down by simply using crude "global" offset values for the Intel pre determined defaults, like use 1 % smaller voltage for all pre-sets, at the cost of increased risk of failure.
I'd like to know how much headroom for offset undervolting there is in those Intel power state tables, that headroom probably only exists to ensure that those settings still work in 5 years or so.
If reliability isn't a big concern, offset undervolting makes sense to me, since it applies to idle/load and all states in between.

 

[VirtualLarry]

Seriously, why bother to save 5W?

 

[cmdrdredd]

Most software relies on the motherboard BIOS options to begin with. It just makes the changes inside the OS. So when you restart the initial boot has defaults from the BIOS and the adjustments are made when the software loads up either with windows or by starting it manually. If the BIOS has no adjustments, then the software you may find won't be able to make any changes to voltages.

 

[FwFred]

This would also allow me to go for below-default values for Idle. Indeed, for power efficiency, undervolting is a must.

I wouldn't worry too much about idle. Haswell Xeon supports C7--just make sure the motherboard does as well. In C7, the clocks and power are both gated, with all level caches powered off as well. Power is removed from the System Agent.

See the Haswell E3 Xeon datasheet , page 49, Table 10.

 

[zir_blazer]

Going to be hard for anyone to give you supported app on that MB unless they trying same thing . Have you tried Intel Extreme tuning utility (XTU)

I checked it. Last version has Haswell support, but officially needs Chipset Z87 and BIOS support. Should be worth to google around if someone got it working on other Chipsets.

These days a lot of motherboards come with utilities that can overclock and adjust voltages CPU fan speed etc, Check the manufacturer site of your potential m/b to see what they have.

Supermicro already stated that they are not going to provide support for those adjustements on the X10SAT because that is not what it was designed for - they offer for the enthusiast market the C7Z87-OCE which is virtually identical, but I will miss the Chipset's VT-d support (Not that critical as you can still use the Processor itself for it, but will possibily not be able to use it on some controllers connected to the Chipset).
I doubt you can find many Motherboards with Q87 or C226 Chipset and a full feature BIOS. They're on totally opposite markets, and I'm trying to get the closest thing possible that pretty much merges both worlds.

The result is a large spreadsheet/matrix of values, that could be avoided/ trimmed down by simply using crude "global" offset values for the Intel pre determined defaults, like use 1 % smaller voltage for all pre-sets, at the cost of increased risk of failure.
I'd like to know how much headroom for offset undervolting there is in those Intel power state tables, that headroom probably only exists to ensure that those settings still work in 5 years or so.
If reliability isn't a big concern, offset undervolting makes sense to me, since it applies to idle/load and all states in between.

What you said is not something I didn't did already, when I builded this 3 years old AIIX4 620 Deneb RB-C2 that I'm using. Default highest PState is 2.6 GHz @ 1.325V, and lowest 800 MHz @ 1.05V, with IMC 2 GHz @ 1.15V. I can do 2.6 GHz @ 1.15V, 1 GHz @ 0.8V, and IMC 2 GHz @ 1.05V. I used to do those with K10Stat. And I spend weeks bashing it with stress test and testing a bunch of settings until I claimed it was rock solid at those values (IMC one was the hardest one to test). I didn't regret having spended that time.

Now, for Haswell itself, I don't even know the actual undervolting margings. This is because if you google around, you will notice that ALMOST NOBODY tried to undervolt it. One of the few reviews it was mentioned was this one, where default Voltage is 1.1V and they undervolted it to 0.9-0.95V. 20% less Voltage is around 35-40% less power consumption on Full Load.

Seriously, why bother to save 5W?

Seriously, why bother purchasing expensive Motherboards, Power Supplies and exotic cooling, to overclock those last 300-400 MHz beyond the Frequency/Voltage curve where you're getting ridiculous increases in power consumption, heat, and component stress, just to get a few mere MHzs that you will possibily not notice due diminishing returns? And I don't see lots of people here saying that overclocking becomes rather stupid after you reach such a point where you're forcing your Hardware beyond rationale limits.
So yes, I'm bothering to save 5W. Actually, more like 15-25W. Don't see anything wrong with that.

 

[Jodell88]

So yes, I'm bothering to save 5W. Actually, more like 15-25W. Don't see anything wrong with that.

If you can get Haswell's idle power consumption down to 9-15W I'll be impressed.

To answer your original question. Look into PHC for undervolting your processor. I would also advise you to use kernel 3.10 or greater to use the intel-Pstate driver and the thermald daemon.

EDIT: I just realised that the advice I gave you conflicts with itself. The intel pstate driver replaces the acpi-freq driver that you need for PHC. Therefore, PHC would not work. It's up to you how you want to proceed in Linux.

 

[rekd0514]

I think we are just going to be SOL on undervolting/clocking on the X10SAT. I agree with some of the above though that the difference at idle will be very small anyways. The load power consumption would be where you would save more, but it doesn't sound like you are going to put any extreme stress on the board 24/7 anyways.

You will save more by getting the correct components for the job of power efficiency. Look into finding a gold or platinum power supply 400W+ depending on how many drives you are going to have, this will likely make the biggest difference out of anything else. Find some power efficient drives and use some Samsung 1.35v memory for a little more savings.

 

[PPB]

Undervolt in offset mode and pray it is stable in all the sleep power saving states? It should undervolt all your Pstates by the same ammount.


If the voltage on those Pstates isnt already pushed to the absolute minimun required for stable operation, you might shave some power consumption without sacrificing stability.

Seriously no K10stat/phenomsrtweaker kind of software for intel core processors? Considering we see the power efficency yadda yadda so often in these forums, i'm atonished no one developed some kind of program that would fit your needs, even more that besides the OP anyone else cares about finding one, apparently.

 

[zir_blazer]

I think we are just going to be SOL on undervolting/clocking on the X10SAT. I agree with some of the above though that the difference at idle will be very small anyways. The load power consumption would be where you would save more, but it doesn't sound like you are going to put any extreme stress on the board 24/7 anyways.

Considering that at worst case scenario that Motherboard will run at stock, and that its Server nature should give it wider tolerance and safety margins by design to warrant the 24/7 operation that you are supposed to be paying for, I doubt it will be possible to actually put dangerous stress on it.
However, the point is that I will be wasting power for no reason. If I can shave 20W out on Full Load, it means that it will generate less heat, and that I could use that headroom to substantially reduce Fan speed and thus noise. And the best part is that I can do so WITHOUT spending a single dollar.

You will save more by getting the correct components for the job of power efficiency. Look into finding a gold or platinum power supply 400W+ depending on how many drives you are going to have, this will likely make the biggest difference out of anything else. Find some power efficient drives and use some Samsung 1.35v memory for a little more savings.

I will be out of cash after purchasing the Xeon and the Supermicro, so a more efficient Power Supply is not an option (Nor will it reduce the useless extra power consumption of the Xeon @ stock). Purchasing a fat ass copper Heatsink with heatpipes with a low noise Fan isn't an option, neither, because I also have to dive into the wallet.
For RAM, I'm using two pairs of these. They come with Hynix MFR ICs and supposedly are currently the ICs capable of reaching the highest Frequencies. However, at 1.35V, there was a Samsung IC that offered better scaling at Low voltages than these. The only modules that I have proof that uses them were the GSkill TridentXs and they were quite a bit more expensive, through at the moment I didn't expected that I would be 2 months happily running 1333 MHz 8-8-8-24 @ 1.35V. May change after jumping into Haswell.

Seriously no K10stat/phenomsrtweaker kind of software for intel core processors? Considering we see the power efficency yadda yadda so often in these forums, i'm atonished no one developed some kind of program that would fit your needs, even more that besides the OP anyone else cares about finding one, apparently.

These days Motherboards makers aren't as draconian as they used to be a decade ago or so, pretty much forcing you to have a high end Motherboard if you wanted to do anything that wasn't running at stock. I think that currently, you have many Motherboards like the ones from AsRock that provides a decent arsenal of BIOS options even on budget. It makes sense that there isn't a current need of these types of tools.
I recall using Clockgen on a Socket 939 ASUS A8N-VM CSM, that was a decent Motherboard, but the near-empty BIOS always made my cry. Two months after getting it, Biostar rocked the budget and high end market (Remember the TForces?). I always feel that I am missing something in the BIOS...

 

[sm625]

And I don't see lots of people here saying that overclocking becomes rather stupid after you reach such a point where you're forcing your Hardware beyond rationale limits.

That type of attitude has been on the rise steadily since 2008. I think it is stupid to use an aftermarket cooler on a non-delidded cpu when you can apparently get the nearly same performance from a delidded cpu on stock cooling for a lot less money.

So yes, I'm bothering to save 5W. Actually, more like 15-25W. Don't see anything wrong with that.

You're not trying to save power for the sake of saving power. It is usually about saving money. And saving 20 watts is saving roughly $20 a year. It doesnt make any sense to spend any more than $50 to gain that much power savings, assuming at least 3 years of 24/365 uptime.

 

[know of fence]

sm6254, you are being overly reductionist in your arguments. Engineering in general is about delicate compromises, something that off-the-cuff summations cannot resolve.
When people buy electronics they primarily want to get a good value for their money, rather than a return on their investment. Components that save power typically offer both.

 

[kzoax_spain]

ok, this post was interesting for my, ...

and this other too:

http://forums.anandtech.com/showthread.php?t=2326402&page=2...

I was one of the first guys who "invented" underclocking ;-) ... at least anybody (nobody) did that... years ago in 2005... 8 years ¡¡

it is true that in the series of dual processors reduce heat, and noise was relevant ... -25% Was fantastic ... i did this in a lot of MB...


but at this time there are some important issues:


1 - The intel icore 7 processor and voltage reductions apply aggressive in haswell... BIOS do something this in good MB... my other old MB, asus G45 chipset do this ¡¡¡.. sleep mode "call they"...


2 - Although 775 cpu ... the undercloking go Fantastic .. I fear that these processors (1150) have a way "boost" ... 3.1 Ghz to 3.9, or 3.4 ~ 5-4Ghz in haswell may exist some stability problems when the equipment requires the full power in such cases the voltage drops slightly and if we are in UNDERV it's possible shutdown ...

3 - Sincerely ... I just bought a haswell 4770S, I have no motherboard today ... or memory or anything ... that's what I'm looking for ...


This cpu has Vd-D x etc ... But I need a board with those options ... VT d X and some controls manual in bios...

Unfortunately the information is confusing ...

ASUS P9D-WS seems to have the options ... but the information on the website of asus is quite erroneous ... the pdf manual is better but... more info is need ¡

----

Server Boards WILL NOT ALLOW UNDERV because they have to ensure stability to voltage variations in all world... and 100% loads in the system ... server+xeon and experiments are incompatible...

in fact, i think that in xeon cpu the Voltaje are more UP, because need stability on all situations... you can see that some TPD in xeion mode are go up... 95W, when similar TPD in haswell domestic are 84W...

The question is know wich real MB have support VT-d or VT-x ....and don´t cost double...

let me doubt much of those "alleged compatibility" with asrock boards... it´s dificult to believe...

----

Another thing:

It is common if you select the voltage manually in the bios, lost some control options and savings in idle states, bios control go to manual, not control automatic ... at least that happened with my asus boards ...

I suppose that if you apply a UNDERV her, to save energy at 100% full, lose efficiency in energy saving modes on idle ... and system was slowly ... act slowly under high load on the system ... this is so that more errors occur reading bits and is needed more confirmations ... on other problems to signals...

example: a old intel 820D run in a more new board, When sleeps-name-bios mode used, UNDERV ... system was very slow .. (not sleep mode), only it´s a name...

the whole system jams when the cpu power required ... no falling over ... but we are losing effectiveness, the task is made ​​more slowly and thus spend more energy over time ... addition to the added risks of possible shutdowns ...

---

I let doubt that you get to 80% to icore7 with 4 cores. (8 threads total) ... it´s dificult get out with this load in domestic users...

in these modes, it is more important to save in "automatic" to speculate on hand ...

about noise, must remember here that this Cpu:

4770S intel (cheaper than a xeon, and with the same options)


TPD has a max of 65 W. .. similar that the best xeion v3...


there is only one xeon with this TPD and costs twice ... the 1285L .... 700 $ ¡¡¡


I have not found why I will have to go for a xeon ... I think it's the same cpu ...


in resume, that with a max of 65W TPD will not cover ever since I guess you use external VGA card ... (Then the energy consumption of the CPU is even lower, this cpu have VGA in... if you don´t used it beter TPD ;-))

also seen as some xeon v3 HAVE NO VIDEO CARD and instead the TPD is equal to or higher than its equivalent, ... 84w ¡

this leads me to think that wings to improve stability and performance under difficult conditions the voltage at the XEON IS HIGHER than in haswell 7icore series 4 ..


The whole subject of best cost with UNDERV may be small .... in that situations in my cpu, 4770S, for example...

It is much more interesting to find a plate with OVERCLOCKING options, , because that this permit do underclocking in BUS, memory ... etc..., (when time passes and the cpu is getting old, you have to speed up the system, (overclocking), but now ? why ¡¡¡ only is 5-10 % better,

I'm writing from an intel D 935, running at 2.8 GHZ ... in UNDERV and When load on overclocking go up ... 3.6 Ghz ... 3.7 or 4.2 with control bios asus... in a very very old asus p5wd2 premium... in stocket 775... with windows 8 and sdd...) with a very old chipset 955x

for me, was very interesting when i used 820D, (version A0) 1.3 v, but after this... intel and BIOS are better and my 935D, have revision D0, and it is a lot of better in his regulation...

in that mode i get better results when i select Mhz bus to 150 Mhz, (200 it´s defaul)... from 3.2 to 2.8 ghz to office, website, and video this is enough... when game or other thing you can used 240 Mhz... to 4 Ghz... but ?... only more noise and wats for some seconds....

ALL this you can not do with a server board...


The question is: find a motherboard with REAL SUPPORT VT-d and tensions and BUS modification options in the bios ... not for do overclocking now, I think that it is not necesary if you buy a icore 7 haswell, there are a lot of powefull, are very very fast...

but in future you will need other cpu or go up...

in TODAY... i only find one motherboard...

I just found this: P9D WS and I'm not sure VT-d-x options... this support icore 7 4 and xeon,...

but you do not have much hope of reducing the consumption of a haswell ... by UNDERV ...

would be more interesting to get off the system bus and run at 2.5 Ghz ... instead of 3.1 ... for example...

other question: this Mb have all that i need, a lot of ...

ASRock Z87 Extreme9/ac

but... have VT-dX ?

ASUS P9D WS have vt d x and control bios underclocking....?


I hope that you understand and excuse me my english...

bye from Spain.

 

[Idontcare]

Welcome to the forums kzoax_spain! :thumbsup:

I was one of the first guys who "invented" underclocking ;-) ... at least anybody (nobody) did that... years ago in 2005... 8 years ¡¡

Great ideas get independently discovered all the time, so don't take it personal when you learn you weren't one of "the first" to invent something, it is still an achievement to have come up with the notion independently even if you are among the "first million" to have done it rather than among the "first dozen".

You will find that underclocking goes WAAAAAAAAAAAAAAAY back in time, to the dawn of electronic circuits themselves even when vacuum tubes were used (before transistors took over).

Underclocking as a means to reduce heat production during summer months versus winter months was very common on the early vacuum tube computers

And in more recent modern times, the 80's and 90's, underclocking your laptop CPU was done by anyone wanting to get another 30minutes out of their laptop battery because the processors back then did not have idle-clockspeeds to save battery (they ran at one clockspeed regardless the CPU load).

For example I had a 386-based laptop that ran at 25MHz stock, I underclocked it to 16MHz to gain extra battery life. (which was nothing special even back then, people were doing this for decades prior)

 

[zir_blazer]

I was one of the first guys who "invented" underclocking ;-) ... at least anybody (nobody) did that... years ago in 2005... 8 years ¡¡

You were 6 years too late. My very first underclocking was with a Pentium MMX 233 MHz around 1999. After two years or so of having it, it started to reset often for no reason when playing Pokemon Red/Blue in a Game Boy emulator, and Super Mario RPG on ZSNES.
I discovered that after several of those restarts, the BIOS would misdetect the Processor as 120 MHz, but it would be rock solid during that session, and didn't seemed much slower. Not long afterwards I discovered that I could get into the BIOS and underclock it to 166 MHz, and it was fine at that Frequency, too, but I didn't needed to reset it tons of time in a row until I get the BIOS to get it to work at 120 MHz.
Sure, chances are that it was dirty and I would get it fully functional after cleaning it. The problem was that clone machines of that era usually had a warranty seal that voided the warranty if you opened the case. So you couldn't do so by yourself, and usually, vendors would always tell you that something else was broken and charge you for a repair.


My first time intentionally undervolting was with a K6-II 550 MHz on a crappy PC Chips Motherboard. The machine had a very crappy Heatsink (That connected to a Molex instead of the Fan header), and at that time I didn't even knew about silicon grease. It worked fine for a year or so, then it became totally unstable.
After getting it in for repairs because it stopped to boot, we got told that the Motherboard was burned and we needed a new one. My mom decided to raise them the middle finger, and bring back home the machine while she thinked whom to ask to repairs. Not long afterwards we figured out that the Motherboard didn't booted because the DIMM didn't perfectly fitted on the slot, and it would boot only if seated in some random manner, through at the slight movement it would fail again.
Still, it didn't fixed stability problems, just boot ones. That machine ended up dissambled over a shoes box (Yes, I already had a open testbed when I was 12 years old) just because it was easier to make the DIMM seat properly that way that inside the case. I discovered that everytime that the machine resetted, if I go into the BIOS Hardware Monitoring part, I would see the CPU Temperature Sensor at near 100°C. I touched the Heatsink because it seemed wrong, and I regretted it. So I figured out that temperature was related to my restarting issues.
The first time I removed the Heatsink, we discovered that instead of the K6-II 550 MHz @ 2.3V that BIOS said (That is remembered as a factory overclock model, anyways), it was just a mere 500 MHz model. Oh, the irony! A model not even binned to get to 550 MHz, trying to go hand-to-hand with a factory overclock, it was its fate to lose and be worse than it. We also discovered later that my cousin machine, purchased at the same time but in a different computer shop, also shared this trait of having been vendor-overclocked. That was an era when end user shops abused Unlocked Multipliers to charge you more.
I remembered about my old Pentium MMX (That I didn't had anymore, because it got traded back in for this K6-II), and tested it at 400 MHz. Not only it worked flawlessly, but it was actually FASTER than at 500 or 550 MHz - believe it or not, there was some sort of Thermal Throttling back then. I recall Ultima Online become unresponsive often at 500 MHz, while it was much more fluid at 400.
Not much time after, that machine got a much more powerful Heatsink with proper silicon grease (Something I didn't know back then), that I had used with an AXP 1800+ Palomino that I builded with my own hands (I didn't know at that moment how to recognized Thoroughbreds). It STILL wasn't stable at 550 MHz @ 2.3V. Then I discovered that if I lowered the Voltage to 2.0 when at 500 instead of the default 2.2V of that model, I would still be perfectly stable, but ran cool as ice. Last time I hear of it after my mom gave it away, it was still working, through with the bothersome DIMM booting issues. Took me several years of abuse until I learned its quircks.

The question is know wich real MB have support VT-d or VT-x ....and don´t cost double...

let me doubt much of those "alleged compatibility" with asrock boards... it´s dificult to believe...

There were already two guys that reported VT-d working as intended on cheap AsRock Motherboards. So yes, it works.

VT-d specifications allows for several DMA Remapping engines, that are the parts where the actual IOMMU job gets done. Haswell itself has two, according to Page 42 of this PDF:

Two Intel VT-d DMA remap engines.
— iGFX DMA remap engine
— Default DMA remap engine (covers all devices except iGFX)

Now, it seems that the Q87 and C226 Chipsets ALSO have their own DMA remapping engines, reason why they support VT-d themselves. Supermicro said that Intel even suggest a workaround to make devices connected to the Chipset to use the Processor DMA remapping engine instead (That is having a PCIe device with VT-d enabled on one of the Processors PCIe Slots). However, even if it works, if you want the full, proper, VT-d support, you need these Chipsets. Reason why I'm aiming for a Server Motherboard. I can either be happy with VT-d on a Z87, or get a C226 for proper VT-d then figure out if I can undervolt by Software.


BTW, there are some niche, high end Motherboards that are intended for Dual Processor Xeons and support overclocking. EVGA used to, with these. However, "niche" means ridiculous expensive, considering that the enthusiast/gaming/overclocker crowd is currently looking as cash cow by manufacturers. You can, but not for cheap.

 

[kzoax_spain]

Welcome to the forums kzoax_spain! :thumbsup:



Great ideas get independently discovered all the time, so don't take it personal when you learn you weren't one of "the first" to invent something, it is still an achievement to have come up with the notion independently even if you are among the "first million" to have done it rather than among the "first dozen".

You will find that underclocking goes WAAAAAAAAAAAAAAAY back in time, to the dawn of electronic circuits themselves even when vacuum tubes were used (before transistors took over).

Underclocking as a means to reduce heat production during summer months versus winter months was very common on the early vacuum tube computers

And in more recent modern times, the 80's and 90's, underclocking your laptop CPU was done by anyone wanting to get another 30minutes out of their laptop battery because the processors back then did not have idle-clockspeeds to save battery (they ran at one clockspeed regardless the CPU load).

For example I had a 386-based laptop that ran at 25MHz stock, I underclocked it to 16MHz to gain extra battery life. (which was nothing special even back then, people were doing this for decades prior)

uhmm... yes. you are correct.

Maybe... i was thinking when i was reading you in that example:

It is difficult to attribute the invention of radio to a single person. In different countries recognize on a local parenting Aleksandr Stepanovich Popov made ​​his first demonstrations in St. Petersburg, Russia; Nikola Tesla in St. Louis (Missouri), Guillermo Marconi in the UK or the commander Julio Cervera in Spain.

of course it is very interesting for me, because i see that a lot of people think the same thing...

I've always thought that improvements in overclocking are rather small, 5-10% of better in test ... 15 or 20 is to limit ... with an obvious risk of losing information during the process ... who knows when something fails, a brief change in the electrical supply and the system crashes as Newton's apple ... maybe it was in one weeks or some years, but ... if you are in That Moment write something interesting in word or openoffice ... you lost all info or if you are win in starcraft and cpu shutdonw ... ohhh;-)

At that time very well designing INTEL excess voltage on the cpu ...

But I think soon after and has been tuned much underclocking improvements (with UNDERV) are rather small ...

thanks for the info. it was interesting.

;-)

 

[kzoax_spain]

You were 6 years too late...


There were already two guys that reported VT-d working as intended on cheap AsRock Motherboards. So yes, it works.

VT-d specifications allows for several DMA Remapping engines, that are the parts where the actual IOMMU job gets done. Haswell itself has two, according to Page 42 of this PDF:


Now, it seems that the Q87 and C226 Chipsets ALSO have their own DMA remapping engines, reason why they support VT-d themselves. Supermicro said that Intel even suggest a workaround to make devices connected to the Chipset to use the Processor DMA remapping engine instead (That is having a PCIe device with VT-d enabled on one of the Processors PCIe Slots). However, even if it works, if you want the full, proper, VT-d support, you need these Chipsets. Reason why I'm aiming for a Server Motherboard. I can either be happy with VT-d on a Z87, or get a C226 for proper VT-d then figure out if I can undervolt by Software.


BTW, there are some niche, high end Motherboards that are intended for Dual Processor Xeons and support overclocking. EVGA used to, with these. However, "niche" means ridiculous expensive, considering that the enthusiast/gaming/overclocker crowd is currently looking as cash cow by manufacturers. You can, but not for cheap.

fond memories.

But at that time did not have the same protection systems by excess heat, today is very difficult to burn excess processor ... the UNDERV was justified at the time, who knows if not apply some "planned obsolescence" in the form of over-voltage ... you save energy, and you save cpu live... too...

In any case, the current object is save energy, save on noise above ... with underV.

in these cases is little justification ... as modern processors, but I also would try "our invention" ...

Personally I think a cpu xeon has some minimum requirements even more high energy than a domestic-users series, precisely because they have to ensure that the system will work, even if a bit hot.

Anyway, you do not forget that "to indicate OFF voltages auto", and select "to manual options" on bios maybe is bad now..,

some very good in power saving on 100 % load of system and energy ... in ASUS bios occurs at least so and I fear that may be the majority, it makes sense, "if you regulate to manual" is that you do not want the automatic regulation of "bios + chipset + cpu", therefore excess saving in the few moments that the cpu to 100% of the load, will lose at times of low consumption, where the bios would use much lower voltages than you can regular manual; I am sure of this, because i see in some boards...

On the other hand, can raise the voltage baar losses data as bit errors and repeated requests, in short, the tests I did in "Chess" fritz, indicated that using the same frequency, voltages very low, the processor estimated less ... very little diferents, ~ 1%, but keep in mind that these are errors that sometime can accumulate leading to a catastrophic failure and data loss ram that point in the best ... or fail in HDD data...

---
About VT-d-x in asrock, i need see in captures, runs OK, PDF info is interesting but... i need confirmation this...

Do you think that in asus P9d WS with c226 have VT-d-x by default, I don´t sure about this... all are new.....

---



Now i am finding info about use ECC RAM, (not errors), with a haswell and have some options to up... i think that ECC ram is better stabilty, and not are a lot of diferent of money...

i see you

bye from Spain.

 

[zir_blazer]

*BUMP*


So, I have been sitting with a Haswell-based Xeon E3-1245V3 and the Supermicro X10SAT for almost a month. As expected, I can't do anything from within the BIOS to undervolt. Shame that I actually need it...


Under some circunstance (Happened once during Windows XP installation in a VM) I already saw the Processor hitting 100°C, and the Motherboard overheating alarm starts to make the speaker beep. This was with Turbo disabled, so the heat buildup should have been slower than with it enabled. Haswell runs hot, everyone knows that, and when you're during a summer heat wave on a poorly-ventilated room with no air conditioner, its even worse. Stock heatsink simply isn't enough, if not purporsefully subdimensioned.

For temperature issues, you either add more cooling, or reduced the heat generated. Spending money on more cooling is not currently an option, and even with cash on my wallet, it would taste bitter to do so.
Due to the fact that coming from an Athlon II X4 620 @ 2.6 GHz which was beyond adequate, this Haswell is currently overkill, and I would gladly reduce generated heat by either: Undervolting @ stock Frequency, or a deep manual undervolt/underclock. Haswell being a mobile-oriented architecture is extremely power efficient, and even with a undervolt @ stock is possible to reduce power consumption by 1/3. However, not being able to undervolt or underclock from within from BIOS, and without any Software option to do the same (Which on previous AMD platforms was possible thanks to Central Brain Identifier and K10stat allowing me easily to do so), makes all of Haswell tweaking capabilities go totally to waste. And reducing temperature during prolonged full load is a must, as I need a safety margin to not reach throttling.
Basically, my only two free options to reduce temperature at the moment, are: Disabling Hyper Threading, because it adds to power consumption, and delidding, something which I don't have the balls to do with a 300 U$D brand new toy under warranty. One would expect that on a world that is so obsessed with "going green" you would have more tools to do so, yet at this moment I have less than in older platforms...

 

[PPB]

Its a shame that one of the bullet points of going intel is totally negated by such a lazy modding community. Thankfully on the amd side there are geniuses like the stilt. Crossing fingers for him to release his apu tool out of beta so he can focus on the fx stand alone tool