Yet another sneaky Nvidia trick

[AzN]

I wouldn't call this a sneaky trick.

 

[krnmastersgt]

Originally posted by: Ylurien
Isn't anyone else worried that overclocking their PCI-E bus is going to cause system instability? It seems like everyone is overjoyed by this news...?

If the PCI-E bus has to be overclocked for a 9600GT to achieve the same (actually slightly worse) results than an 8800 at stock PCI-E speeds, doesn't that mean the product (9600GT)is inferior? I realize that the number of shaders and ROPS and everything is quite different, but for the price, why would anyone still be considering the 9600GT?

I don't want to have to overclock my PCI-E bus just to get close to what I could get if I were to spend a few more dollars on an 8800GT. This new information certainly explains why the 9600GT runs even hotter than the 8800GT.

It seems to me like Nvidia is doing these kinds of things so that they can skip an entire generation of video card technology - and avoid the costs that would have gone into the research and development for that generation - and still maintain a lead over ATi.

By what means did you come to that conclusion? The 9600 GT still performs well without the pci overclock, HOWEVER with it, theres even more performance coming out of the card.

 

[Cookie Monster]

Im not so sure why is this shady. If they implemented in the first place knowing that increasing PCI-e frequency could potentially kill hardware, then theres no need to worry about this.

Linkboost feature has been a feature that has been dropped. Directly from NVIDIA's website

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

Same goes for 780i. The 650i/750i never supported this in the first place.

With MOST pci-e frequency remaining at default value (i.e no OC has taken place) i have no idea why techpowerup is calling this a shady trick.

 

[Ylurien]

Alright - brain lapse on my part. For some reason I thought the PCI bus was connected to the PCI-E bus, but of course that's not true.

So ok we can overclock the bus to get higher speeds with the 9600GT. But doesn't that effectively mean that we are overclocking how fast the card runs, which will increase temperatures? How can we then have much room to overclock the core or the memory of the card itself?

Isn't there a set amount that any card can be overclocked, whether it's coming from an overclock in the PCI-E bus or from an overclock on the card itself? If we overclock an 8800GT as far as it can go on the card itself, and then overclock a 9600GT using a combination of a PCI-E bus overclock and an on-board overclock, won't they be pretty much the same sort of Mhz boost for each card? Or is the 9600GT just inherently better with heat, etc., and therefore more overclockable?

I probably am too clueless about this, so if anyone could enlighten me...

 

[BFG10K]

Can't this be put to rest by testing it on Intel chipsets to see if the observed (as in benchmarked) performance changes versus an 8800GT on the same board under same conditions?

Yes, but then Riva Tuner isn't in question given it was the one that found the discrepancy in the first place.

 

[krnmastersgt]

Originally posted by: Ylurien
Alright - brain lapse on my part. For some reason I thought the PCI bus was connected to the PCI-E bus, but of course that's not true.

So ok we can overclock the bus to get higher speeds with the 9600GT. But doesn't that effectively mean that we are overclocking how fast the card runs, which will increase temperatures? How can we then have much room to overclock the core or the memory of the card itself?

Isn't there a set amount that any card can be overclocked, whether it's coming from an overclock in the PCI-E bus or from an overclock on the card itself? If we overclock an 8800GT as far as it can go on the card itself, and then overclock a 9600GT using a combination of a PCI-E bus overclock and an on-board overclock, won't they be pretty much the same sort of Mhz boost for each card? Or is the 9600GT just inherently better with heat, etc., and therefore more overclockable?

I probably am too clueless about this, so if anyone could enlighten me...

Overclocking the 9600 through the pci-e bus yields the same results as an overclocked 8800 gt. Yes there is a limit to how much the card can be overclocked, ocing the pci-e bus will have a direct effect on the core clock however it operates faster than just raising the core clock alone, that's why it's so interesting.

 

[nullpointerus]

Yep, I remember reading about this in an article a LONG time ago.

It's only dishonest if (a) the tech. is being enabled solely for a specific card and (b) nVidia is purposely not bringing this to any of the reviewers' attention.

Anyway...

Older versions of nTune (i.e. back in the NF4 Ultra days) started their automated system tuning routines by overclocking the PCI-E bus. I asked why on these forums at that time -- I was stuck with a 6200TC and was trying to OC the heck out of it -- but no one seemed to know what PCI-E bus overclocking would accomplish.

Naturally, the system would typically freeze solid at these modified PCI-E clock settings shortly after the auto. stability test concluded. The bottom line is that I don't recall seeing any real-world benefit at the time, and support was pretty crappy then, too.

The 9600 GT is a much higher performing part, so PCI-E bandwidth could play a small role.

(This kind of thing *might* have worked out well for nVidia motherboards and nVidia AGP cards since the AGP bus was used for nothing but the graphics card.)

However, other things use the PCI-E bus. Onboard gigabit network adapters, disk controllers, etc. may be based off the PCI-E bus. As time goes by, more and more peripheral cards (e.g. sound cards, TV tuners, etc.) will be moving to the PCI-E bus. Overclocking the bus itself may destabilize all these kinds of components.

To find out exactly what is running off the PCI-E bus on your computer, you can open Device Manager and select the option View --> Devices by connection. Look for devices off the PCI- Express ports, not directly off the PCI bus itself.

Some people here are very confused. There is a big difference between PCI and PCI-E. They run at different speeds (e.g. 33/66 mhz vs. 100 mhz), and they can be overclocked independently. So you must NOT confuse what is running on the PCI-E bus with what is running on the PCI bus.

On my board (Gigabyte P35-DS3R), it looks like this:

PCI bus
|
+--HD Audio ( <-- PCI )
|
+--P35 PCI-Express Root Port
+-----8800 GTS ( <-- PCI-E !! )
|
+--ICH9R SATA RAID Controller ( <-- PCI )
|
+--ICH9 PCI-Express Root Port 4
+----JMB36X RAID Controller ( <-- PCI-E !! )
|
+--ICH9 PCI-Express Root Port 5
+----Realtek 8111B Gigabit Ethernet ( <-- PCI-E !! )
|
+--ICH9 SMBus Controller ( <-- PCI )
...
...

I skipped a bunch of stuff, but you can see where I'm going with this. The bolded items will be running out of spec when the PCI-E bus is overclocked. That's why we have PCI-E locks on the system board when we overclock the NB, right?

It's OK if nVidia designed their boards to handle these wierd bus overclocks.

But ... at these out-of-spec. PCI-E speeds:

1. I (personally) don't trust nVidia to test their onboard PCI-E peripherals (i.e. RAID, LAN).
2. What about PCI-E expansion cards, which CANNOT be certified by nVidia?

Automating this system-wide bus overclock as a design feature for a little GPU tussle between ATI and nVidia ... well ... that was simply reckless and idiotic, at best. nVidia abandoned this misfeature a long time ago.

Hopefully, no one went into their BIOS and tried this without considering the consequences...

 

[Piuc2020]

Originally posted by: krnmastersgt
That makes a lot of sense actually, someone will probably say that isn't entirely fair but I like the way the 9600 is shaping out, and it's big brothers are soon to come out which if they follow the same designs, will be monstrous cards. Maybe we can finally play Crysis at 19x12 with max settings at over 60 fps.

Do all the wishful thinking you can, and even if quantum mechanics says otherwise, it won't change the fact that your claims are incoherent, unfounded and quite simply, unreal.

 

[Zap]

Originally posted by: Demoth
Don't care how they did it, the power draw is still low relative to it's power.
...
Good riddance to 8400GTs and other garbage cards like it.

The power draw is lower probably because of fewer active transistors. This is how Intel's Silverthorne promises to be such a low power CPU, by having very few transistors compared to other modern CPUs.

As for "garbage" cards, I'm sure Nvidia is thinking of how it can cripple an 8500GT with a 64 bit memory interface and rebadge it as a 9400GS or something.

Originally posted by: nullpointerus
To find out exactly what is running off the PCI-E bus on your computer, you can open Device Manager and select the option View --> Devices by connection. Look for devices off the PCI- Express ports, not directly off the PCI bus itself.

Hey, thanks for the tip. I knew it was there but didn't even think about looking in it. I see that the add-on EIDE controller on my abit IP35-E is on PCI-E. Nice.

 

[krnmastersgt]

Originally posted by: Piuc2020

Originally posted by: krnmastersgt
That makes a lot of sense actually, someone will probably say that isn't entirely fair but I like the way the 9600 is shaping out, and it's big brothers are soon to come out which if they follow the same designs, will be monstrous cards. Maybe we can finally play Crysis at 19x12 with max settings at over 60 fps.

Do all the wishful thinking you can, and even if quantum mechanics says otherwise, it won't change the fact that your claims are incoherent, unfounded and quite simply, unreal.

What claims are you disputing? As for this info about the PCI-E bus overclock and it's effect on the 9600, there is an article linked you know, and it's more than just 1 page.

 

[Sylvanas]

I don't think its shady, but I do think Nvidia should have at least explained the situation. I have been running 125mhz PCI-E for like 4 months now, no problems.

 

[Dadofamunky]

Originally posted by: krnmastersgt
That makes a lot of sense actually, someone will probably say that isn't entirely fair but I like the way the 9600 is shaping out, and it's big brothers are soon to come out which if they follow the same designs, will be monstrous cards. Maybe we can finally play Crysis at 19x12 with max settings at over 60 fps.

Don't hold your breath
heh

Not an ATI fanboy, I just know that DX10 makes video cards cry

 

[krnmastersgt]

Originally posted by: Dadofamunky

Originally posted by: krnmastersgt
That makes a lot of sense actually, someone will probably say that isn't entirely fair but I like the way the 9600 is shaping out, and it's big brothers are soon to come out which if they follow the same designs, will be monstrous cards. Maybe we can finally play Crysis at 19x12 with max settings at over 60 fps.

Don't hold your breath
heh

Not an ATI fanboy, I just know that DX10 makes video cards cry

Hopefully tri-sli will get us there I don't even play Crysis, nor do I want to, I just want to know that the most demanding game out there can actually be run at a high res with max settings and still remain beautiful and seamless.

 

[nRollo]

Originally posted by: krnmastersgt

Originally posted by: Dadofamunky

Originally posted by: krnmastersgt
That makes a lot of sense actually, someone will probably say that isn't entirely fair but I like the way the 9600 is shaping out, and it's big brothers are soon to come out which if they follow the same designs, will be monstrous cards. Maybe we can finally play Crysis at 19x12 with max settings at over 60 fps.

Don't hold your breath
heh

Not an ATI fanboy, I just know that DX10 makes video cards cry

Hopefully tri-sli will get us there I don't even play Crysis, nor do I want to, I just want to know that the most demanding game out there can actually be run at a high res with max settings and still remain beautiful and seamless.

So far 3 way SLi hasn't been a big factor with Crysis other than running higher AA at unplayable settings.

My own experience with it is that I'm running Crysis at 19X12 0X16X with mixed medium and high settings on my 25X16 monitor.

If I were in the market for 3 way SLi, Crysis is not what I'd buy it for at this point.

It does help on some other DX10 games though like Lost Planet, and once you get past the cost and supporting hardware requirements is very user friendly.

 

[SniperDaws]

Ok ive just done 2 benchies of 3dmark 06 heres the setup

Q6600 @ 3.2Ghz
2048 DDR2 XMS2
9600GT @ Stock 650/1625/900
P35C DS3R Rev 1.1 Intel P35 chipset.

Scores with PCI-e @ 100Hz

3dMark score 11527
SM2.0 4683
SM3.0 4387
CPU 5091


Score With PCI-e @ 110Hz

3dMark Score 12176
SM2.0 5003
SM3.0 4667
CPU 5081


So a little jump in performance there

 

[taltamir]

This sounds to me like nvidia FINALLY made a platform useful...

Driver based OC when combining an nvidia motherboard and an nividia card... which they KNOW is going to be stable because they made both parts!
See this is pure genius!

I have a G92 8800GTS 512 and P35 mobo with the 174 drivers (thanks to the modded inf file from http://www.laptopvideo2go.com/ ... how would I check the the PCIe speed though? I don't think the drivers could just change my motherboard's settings like that. This has to be something done while the drivers are active (in windows), and even then I doubt it could be done for a P35 board. But I would check..
According to what people are saying here, the bus increase increases the core clock since it works as a multiplier off of the bus speed. Well my core clock is at the default according to GPUz, 650mhz.
So I am not getting the OC on an 8800GTS 512 on vista64 with 174 modded drivers with a P35 mobo.

 

[Janooo]

NV had a good laugh.
Everybody thought it's a some kind of magic. People thought NV is tweaking the GPU and all they did is a speed increase.

 

[Tempered81]

Originally posted by: SniperDaws
Ok ive just done 2 benchies of 3dmark 06 heres the setup

Q6600 @ 3.2Ghz
2048 DDR2 XMS2
9600GT @ Stock 650/1625/900
P35C DS3R Rev 1.1 Intel P35 chipset.

Scores with PCI-e @ 100Hz

3dMark score 11527
SM2.0 4683
SM3.0 4387
CPU 5091


Score With PCI-e @ 110Hz

3dMark Score 12176
SM2.0 5003
SM3.0 4667
CPU 5081


So a little jump in performance there

Nice it works on p35 boards, as well as nvidia, and therefore is not a 780/650/680i exclusive feature, and has nothing to do with linkboost.

And the 9600gt is the only nvidia card that does this.
Interestingly it reports the same core speed at 100, or 110 yet notice the obvious performance increase in 3dmark. I'd say that's a tad shady.

 

[taltamir]

wait a minute. those scores from sniper... did he manually overclock it, or did he just upgrade the driver? if he manually overclocked it, did he check that the new driver causes it to automatically overclock?

 

[Tempered81]

nah, he went into his bios with the card clocked at stock speeds. He went to PCI-E frequency and changed it from 100 to 110. booted windows and ran 3dmark06.

Then he went back to bios and set the PCI-E back to 100mhz (which is default) ran 3dmark again, and got a lower score.

 

[SniperDaws]

Originally posted by: jaredpace
nah, he went into his bios with the card clocked at stock speeds. He went to PCI-E frequency and changed it from 100 to 110. booted windows and ran 3dmark06.

Then he went back to bios and set the PCI-E back to 100mhz (which is default) ran 3dmark again, and got a lower score.

Correct


but i do have an auto setting for my PCIe would this make any diffrence.

i tried 110Hz with my card at 740/1850/975 and it locked up halfway through return to proxycon or whatever its called.

 

[Tempered81]

its locking because the card inst actually running at 740/1850/975, it would be running at 814/1850/975 if the shaders aren't linked. If you had them linked, it would be attempting 814/2070/975

 

[SniperDaws]

oh yeah.......DOH!

 

[Janooo]

Originally posted by: jaredpace
...


Nice it works on p35 boards, as well as nvidia, and therefore is not a 780/650/680i exclusive feature, and has nothing to do with linkboost.

And the 9600gt is the only nvidia card that does this.
Interestingly it reports the same core speed at 100, or 110 yet notice the obvious performance increase in 3dmark. I'd say that's a tad shady.

Now, this is not nice. That's called lying.
What they were thinking. That somebody wouldn't find out?

 

[the unknown]

What about people with locked BIOS who bought their comp from a retailer like Dell or Gateway? I'm guessing they won't be able to take advantage of this?