Tech Power Up's article on the 9600 "Dirty Trick"

[ECH]

It appears that per the article (if I understood this correctly) they found a discrepancy between the core clock speed reported by the driver and the core clock speed read from the clock generator directly. When you increase PCIe frequency instead of getting more bandwidth you over clock the video card. Nvidia's mobile GPUs are suppose to do this as well. Others have tied this in with not just manual PCIe frequency over clocking but the use of Link Boost as well. What some are saying (in other forums) is that:
-PCIe Frequency is not intended as a means to over clock the video card
-high PCIe frequencies can lead to HDD failure and other instability problems on some motherboards
-the average reviewer/ user wouldn't know the actual GPU clock of the video card
-if the PCIe Frequency is higher then 100MHz the video card was benchmarked at a higher OC then other competing video cards
-Shader clock is not effected
-Other comparison video cards do not have this feature when benchmarked.
-We really don't know if reviewers are increasing the PCIe Freq. when conducting benchmark reviews. Assumptions can be made but there is no proof that they don't increase PCIe Freq.

The article can be read here:
Source

There are a few other forums discussing it but I didn't see it mentioned in this forum. There are mix reactions to this news some love the idea while others question it's use. What are your thoughts?

 

[IbaneZ]

Linkboost has been around a while.
http://www.nvidia.com/object/IO_31319.html

NVIDIA LinkBoost™ technologyi—automatically increases PCIe bus speeds on the motherboard when LinkBoost-enabled GPUs, such as the NVIDIA GeForce 7900 GTX, are installed

What's dirty about this? Free performance is nice ain't it? Blame the lazy reviewers for not knowing what they're doing.

 

[trinibwoy]

Also, how many reviewers overclock their PCIe buses?

 

[AlexV]

Linkboost has been around a while.
http://www.nvidia.com/object/IO_31319.html



What's dirty about this? Free performance is nice ain't it? Blame the lazy reviewers for not knowing what they're doing.

It's not quite that...did you read W1zzard's article on techpowerup?I wouldn't call it a dirty trick, it's more of a silent change that might do significant stuff:instead of the PLL getting the base frequency from the crystal oscillator(traditionally 27Mhz for nV cards), it gets it by dividing the PCIE bus frequency by 4. If the PCIE frequency is the standard 100, you get 25 and you get the clocks the nV driver lists for the 9600GT. But with Linkboost enabled, the PCIE frequency jumps to 125Mhz...thus the base frequency becomes 125/4=31.25, thus making the clockrate:31.25*26/1=812.5MHz...without the user knowing it.

OCing the PCIE bus generally provides little real-world performance....OCing the core by a considerable ammount, like in the above example, provides a more than detectable real world improvement. The issue seems to be that nV hasn't openly advertised this and thus it might seem shady. OTOH, one should assume that basic testing is conducted with default clocks(including PCIE), and thus the point would be moot for reviewers.

 

[ShaidarHaran]

Whoa! 125MHz PCIe clock? Seems a bit high to me. I can see why data corruption of HDs with some boards would be an issue there.

 

[ECH]

here are a few that are confirming that it works

http://www.xtremesystems.org/forums/showpost.php?p=2808463&postcount=125
http://www.xtremesystems.org/forums/showpost.php?p=2807416&postcount=97
http://www.xtremesystems.org/forums/showpost.php?p=2806865&postcount=46

 

[AlexV]

Whoa! 125MHz PCIe clock? Seems a bit high to me. I can see why data corruption of HDs with some boards would be an issue there.

nV's documentation on Linkboost suggests an 125% improvement, and the numbers they give corelate to an increase from 100% to 125%. Dunno how it plays out IRL as I won't touch an nV based mobo anytime soon

 

[Davros]

Linkboost has been around a while.
http://www.nvidia.com/object/IO_31319.html



What's dirty about this? Free performance is nice ain't it? Blame the lazy reviewers for not knowing what they're doing.

as long as reviewers make it known
for example my mboard has an auto overclocking facility 5% to 30% and if it was set to 30% and not mentioned during a review that would be dishonest

 

[Blazkowicz]

that makes me think of the insane 486 DX50 (not DX2). that gives you a 50MHz VL bus, in theory 50% faster than PCI . Is the ISA bus clocked at 12.5MHz on such a system?
should I try o/cing a DX2/66 to 100MHz?

 

[ECH]

Wait a minute, aren't most (if not all) NV 780i's motherboards run a PCIe frequency of 125Mhz as the default and it is not adjustable?

 

[dizietsma]

Wait a minute, aren't most (if not all) NV 780i's motherboards run a PCIe frequency of 125Mhz as the default and it is not adjustable?

No, default is 100. Mine doesn't have linkboost either but it does have something about nvidia GPU which is set to disabled by default.

 

[Pete]

So the issue is not that (some?) NV MB owners who buy a 9600GT get free performance, just that everyone else doesn't, and if reviewers use (specific?) NV MBs that would present a false picture of precisely how the 9600GT performs relative to everything else? (Ignoring any potential system stability--surely NV isn't going around OCing ppl's systems willy-nilly.) I'm trying to figure out the "dirty trick" here. It's surely not free performance, but it seems to be free performance boost for some that's potentially misrepresented as default performance for all.

Either way, it's a nice piece of investigative journalism. Would it be too cynical to ask if W1zzard's initiative alone seeded this pearl? >:^D

 

[FrameBuffer]

No, default is 100. Mine doesn't have linkboost either but it does have something about nvidia GPU which is set to disabled by default.

is there a definitive link one can find for this info ? According to ViperJohn at techPower : "Simply plugging the card into a motherboard that runs a higher default PCIe bus frequency, such as a reference 780i (125Mhz default PCIe), will cause the 9600GT to run higher core clock speeds "

Checking several 780i motherboard makers sites for product manuals (unfortunately XFX requires a product serial number/code to DL a manual) the most info I have found is a bios setting that allows the user to adjust the PCIe 16x frequency from 100-200mhz. I would assume that naturally the default would be the lowest value, however of course that isnt always true (see memory).

Not sure about the "ethics" involved (OK who are we kidding .. the LACK of ethics in business shouldn't surprise anyone), it does bring into question the validity of product comparisons with products that don't enjoy the benefits that LinkBoost bring. For example if a possible consumer is on the fence about a 9600GT and say a 3870 given their near equal pricing. Reviews using a 590-790 nforce mobo when reviewing the 9600GT would show the 9600 scoring typically within a few % of the 3870. Said consumer buys the 9600 expecting near equal performance, brings it home and plugs it in their Intel/ATI chipset motherboard to find that the 9600GT is now about 15-25% LESS than what they were expecting.

 

[ShaidarHaran]

nV's documentation on Linkboost suggests an 125% improvement, and the numbers they give corelate to an increase from 100% to 125%. Dunno how it plays out IRL as I won't touch an nV based mobo anytime soon

I hear that. Only Intel and ATi (AMD) chipsets for me.

This has me tempted to o/c my PCI-e clock now though

 

[satein]

as long as reviewers make it known
for example my mboard has an auto overclocking facility 5% to 30% and if it was set to 30% and not mentioned during a review that would be dishonest

Errrrr.... nope.... That's not the reviewer's job!! How come people know without reading the review if it is a product specification

It's the NV's job

 

[Twinkie]

***Note: NVIDIA LinkBoost™ technology has been removed as a feature from NVIDIA nForce® 680i SLI

I.e 780i doesn't support this dropped feature. Nor does 650i/750i (it was an exclusive feature for the 680i only). Plus not many people OC their PCI-e frequency so i dont think this really is a dirty trick. Everything is remained at stock.

But from what i can tell this "linkboost" feature will be present in the next wave of nVIDIA mobos. Why? because they still have it as a checkbox under their tech specs meaning they arent done with this dropped feature for current nV mobos yet. (they dropped in the first place because it never had any tangible performance benefits just by increasing the PCI-e fequency). This time around its a little different, seeing as increasing the PCI-e frequency is proportional to the core clock frequency (not shader) according to TPU.

Maybe thats why nVIDIAs a little inclined to explain about this according to TPU because this feature could potentially be tied to a "yet to be announced future product".

 

[ECH]

I.e 780i doesn't support this dropped feature. Nor does 650i/750i (it was an exclusive feature for the 680i only). Plus not many people OC their PCI-e frequency so i dont think this really is a dirty trick. Everything is remained at stock.

But from what i can tell this "linkboost" feature will be present in the next wave of nVIDIA mobos. Why? because they still have it as a checkbox under their tech specs meaning they arent done with this dropped feature for current nV mobos yet. (they dropped in the first place because it never had any tangible performance benefits just by increasing the PCI-e fequency). This time around its a little different, seeing as increasing the PCI-e frequency is proportional to the core clock frequency (not shader) according to TPU.

Maybe thats why nVIDIAs a little inclined to explain about this according to TPU because this feature could potentially be tied to a "yet to be announced future product".

This has nothing to do with Linkboost in particular. Linkboost was mentioned because it does the exact same thing. The 9600 GT can be over clocked with a PCIe frequency higher then 100MHz. Regardless of how the PCIe Frequency is OC'd via manually or through linkboost, it does increase the GPU of the 9600 GT. The real question here is who is using a PCIe Frequency above 100MHz? A lot of people do regardless if they use a Nvidia, Intel, etc chipset. Some do it because the theortical bandwidth does help them gain performance. While others believe that they must increase PCIe frequency in order to properly OC their CPU.

But what about those that reviewed the 9600 GT? Now they have to speak up in order to quell any doubts on their review (be it they did increase the PCIe Frequency because they didn't know or otherwise). As it stands there is no proof that reviewers (in general) don't using higher PCIe frequencies because it was never an issue if they did until now.


Here is the problem in a nutshell, according to AlexUnwinder (I assume you know who he is) programs like nTune, GPU-Z and ExpertTool, etc only show the target clock or what clock rate you are requesting. The only programs that monitor real PLL clocks: RivaTuner and Everest posted here. What this means is that any review that didn't use Rivatuner or Everest and, the reviewer used a PCIe frequency higher then 100MHz (regardless if he knew it or not) showed an incorrect GPU reading. More importantly, showed a lower GPU reading providing OC'd performance results.

ALSO

According to ViperJohn, those who use 780i's motherboards along with a video cards that is PCIe 2.0 compliant runs a 125Mhz PCIe bus frequency and it's not adjustable. That is found in this post (regardless of what you were told or what you read). What this means (if true) is that any reviewer using a 780i MB wouldn't have known for the most part unless they used Rivatune or Everest. Even if they did wouldn't have automatically thought something was wrong.

So I have to ask, can any show that a 780i doesn't use 125MHz PCIe bus using a 9600GT?

 

[Twinkie]

What you are saying is correct. But i was trying to explain why the 9600GT could exhibit such behavior, and the only anwser is tied with the dropped feature of the 680i, "linkboost".

I dont see the problem with this because of the fact that most reviewers and users never touch the PCI-e fequency settings in the BIOS (why should they in the first place? what tangible performance benefits does one get when you increase the PCI-e frequency?). Same goes for mobos based on nVIDIA chipsets. There is no "linkboost" taking effect so the PCI-e frequency remains at stock. Since this is the case, i dont see the big fuss about this because there is no OCing taking place for the 9600GT because the PCI-e frequency is remained at their default value which is 100MHz.

Unless some motherboards run their PCI-e frequency higher than normal which i doubt since they will run at their rated specs, i dont see how this can be classified as a dirty/shady trick.

Who here actually touches their PCI-e frequency?

edit - although im making assumptions, im not to familiar on users OCing their PCI-e frequnecy but rather setting it to their default value so that there is no instability issues when OCing their systems.
edit 2 - The real question is, what does this mean to the consumer? Considering that nVIDIA has kept quiet about this is still puzzling seeing as the 9600GT performs faster on their own chipsets (780i, not sure about 750i or the 6x0i) due to the OC taking place.
For the most part, im trying to figure out what part of this can be considered a "dirty trick". Since viperjohn stated that the 780i run their PCi-e frequency higher than normal (which i didnt know but thanks for the info ) then test systems based on the 780i should be notified about this.

 

[ECH]

What you are saying is correct. But i was trying to explain why the 9600GT could exhibit such behavior, and the only anwser is tied with the dropped feature of the 680i, "linkboost".

I dont see the problem with this because of the fact that most reviewers and users never touch the PCI-e fequency settings in the BIOS (why should they in the first place? what tangible performance benefits does one get when you increase the PCI-e frequency?). Same goes for mobos based on nVIDIA chipsets. There is no "linkboost" taking effect so the PCI-e frequency remains at stock. Since this is the case, i dont see the big fuss about this because there is no OCing taking place for the 9600GT because the PCI-e frequency is remained at their default value which is 100MHz.

Unless some motherboards run their PCI-e frequency higher than normal which i doubt since they will run at their rated specs, i dont see how this can be classified as a dirty/shady trick.

Who here actually touches their PCI-e frequency?

edit - although im making assumptions, im not to familiar on users OCing their PCI-e frequnecy but rather setting it to their default value so that there is no instability issues when OCing their systems.

We will agree to disagree here.
-There is no proof that reviewers (in general) don't use higher PCIe Freq. No explanation is needed nor is one necessary to explain why they do or don't
-According to ViperJohn, a 780i using a video card that is PCIe 2.0 does increase the PCIe Frequency. I ask again, can you or anyone else prove otherwise other then what you read?

 

[cho]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3192&p=3

the bus speed of pcie 1.x , 2.x and 3.x all are 100mhz not 125mhz.

there is no more "link boost" in 6xx/7xx series chipset mainboard AFAIK, but the GPU Ex is exist that was set to disabled by default. the GPU Ex allow some driver-level optimizations as it requires a 90-series driver.