Interesting info on early NV40 memory

[radar1200gs]

You may recall there was some controversy over the speeds of memory on early NV40 card.

It was discussed in this thread. http://www.beyond3d.com/forum/viewtopic.php?t=11444
Two interesting quotes:

DaveBaumann
Senior Member



Joined: 29 Jan 2002
Posts: 6919
Location: Bedforshire, UK
Posted: Mon Apr 12, 2004 1:57 am Post subject:

--------------------------------------------------------------------------------

Depends on what you are doing with the memory.
_________________
'Wavey' Dave
Beyond3D
"The drinks are on the roof"

PaulS
Senior Member



Joined: 11 May 2003
Posts: 465
Location: UK
Posted: Mon Apr 12, 2004 2:30 am Post subject:

--------------------------------------------------------------------------------

Dave is refering to the memory being overclocked, hence the cooling. Nothing to do with ATi.

Bouncing Zabaglione Bros. wrote:
No vent to the outside?


I was thinking the same, since I always thought that was a good idea. Is there a reason they're not doing that this time?
_________________
- Paul

Last edited by PaulS on Mon Apr 12, 2004 2:33 am; edited 1 time in total

Now take at look at a post from this thread over at nvnews:
http://www.nvnews.net/vbulletin/showthread.php?t=32362

lowdog
Registered User


Join Date: May 2004
Posts: 100 Re: Question about which brand has a better 6800 Ultra.

--------------------------------------------------------------------------------

Quote:
Originally Posted by saturnotaku
I thought BFG was the sole maker of NVIDIA's reference cards?


http://www.auspcmarket.com.au/ ...link to the card, S-Media (perfect Nvidia reference card)


Don't know about that. I got my card about 6 weeks ago now and it is a perfect Nvidia reference card. Nvidia sticker on the fan and Nalu picture on the heatsink cover. Came in a white box with nothing else but a DVI/VGA adapter and 60.85 drivers burnt onto a plain gold cd.

It is the exact same card that were sent to reviewers when the card first became available for review.

Oh and regarding the GC20 modules. I e-mailed Nvidia about them being overclocked and they replied that the first few runs of GC16 modules were marked incorrectly by Samsung as GC20 yet were infact GC16. The Nvidia Tech guy who responded to my e-mail said that Nvidia had Samsung retest the mislabled GC20 modules to indeed confirm that they would run at GC16 specifications.....he assured me that the GC20 modules on the first batches of 6800U's were infact GC16 speced and that I had nothing to worry about.

Still seems ppl are getting 6800U with the GC20 modules on current cards atm that run well over 1200MHz.

http://www.nvnews.net/vbulletin/showpost.php?p=374059&postcount=12

Do those who claimed nVidia was cheating and overclocking the ram still stand by their claims in the light of this information?

 

[Evildeus]

What's the point of this thread? Many Board Makers are still OC memory on their cards. FYI
http://www.anandtech.com/video/showdoc.aspx?i=2126&p=2

First up, we have the Albatron Trinity GeForce 6800 Ultra. The bright yellow box advertises 2ns DDR3 Memory; though, at 2ns, 1.1GHz would be a bit of an overclock.

Both cards sport GDDR3 RAM rated at 2.0ns, configured with a bit of a factory overclock. 2.0ns RAM should only be able to hit a 1GHz data rate, but both vendors are running at 1.1GHz data rates.

 

[Mariner]

Do those who claimed nVidia was cheating and overclocking the ram still stand by their claims in the light of this information?

As a caveat to my response, I must first say that I don't give a flying f**k whether or not NV deliberately overclocked memory.

What I will say is that quoting as certain fact unsubstantiated information you read on another messageboard that somebody reported they'd been told by an unnamed person at NV in an e-mail which is not quoted doesn't go anywhere close to convincing me!

Now, if you had quoted similar information which came from an official source at Samsung, I'd be quite certain it was the truth.

Ultimately, as I don't own an early NV40 card it is unimportant to me and if whatever type of memory was used works to specs that's fine, too.

 

[IgnorancePersonified]

LOL!

 

[dizietsma]

Considering the 2ns 1000Mhz memory is clocking up to the 1100-1150 level I would hazard I guess that all the 1.6ns memory modules are clocking up to 1250-1300 levels on 6800 cards.

Actually, I see very little people getting this sort of clock. So what's happening, is it all 2ns still or is it 1.6ns that cannot get to 1200 or have I just missed them ?

nvidia cards normally have overclocked very well for DDR and even hot DDR2, but something seems a bit "sick" with the latest memory overclocks.

 

[radar1200gs]

The implication at the time was that nVidia was trying to cheat and pull the wool over reviewers eyes by overclocking the memory on reference boards, claiming they were one thing qwhen in fact they were another (in other words people were trying to say that nVidia hadn't changed and were going to continue to lie their ass off with nV40).

Retail boards are another matter altogether. Both ASUS and Albatron make the clock rates of thier parts public knowledge and the overclocks are manufacturer sanctioned overclocks.

 

[gokickrocks]

...and if samsung did some binning, then it can still be considered overclocked

just because something is spec'd at a particular setting does not imply that it is not overclocked (refer to ram for the motherboard...ie: kingston hyperx/corsair/mushkin lvl2 using the same bh5 ram modules for pc3200 and pc3500 specs)

although i am not implying that nvidia was doing the overclocking

 

[Dave Baumann]

What's the point of this thread? Many Board Makers are still OC memory on their cards. FYI
http://www.anandtech.com/video/showdoc.aspx?i=2126&p=2

First up, we have the Albatron Trinity GeForce 6800 Ultra. The bright yellow box advertises 2ns DDR3 Memory; though, at 2ns, 1.1GHz would be a bit of an overclock.

Both cards sport GDDR3 RAM rated at 2.0ns, configured with a bit of a factory overclock. 2.0ns RAM should only be able to hit a 1GHz data rate, but both vendors are running at 1.1GHz data rates.

Given that there are partner boards available at the same time with fully labelled 1.6ns RAM it would appear that Samsung have sorted out their "labelling issue", in which case would it really take two and a half months for these mislabelled modules to pass through, given the sparcity of the RAM availablility?

And if NVIDIA retested the RAM why did they then proceed to label them with the apparently "incorrect" speeds?

And why must everything be a "cheat" to Greg?

 

[radar1200gs]

Dave, first, read what I quoted you as saying in the old thread.

Second, nVidia does not label memory, samsung does!

 

[Dave Baumann]

The implication at the time was that nVidia was trying to cheat and pull the wool over reviewers eyes by overclocking the memory on reference boards, claiming they were one thing qwhen in fact they were another (in other words people were trying to say that nVidia hadn't changed and were going to continue to lie their ass off with nV40).

There was no such "implication", other than how you percieve these things. The boards were labelled with 500MHz RAM and they shipped with modules that correspond to those speeds - anyone can see that and operational speed of 550MHz is greater than they are labelled. The fact that partner boards are still going out with overclocked RAM is a good indication that this is why there is RAM cooling as it was the design is catering for the overclock due to lack of 1.6ns RAM supplies - we know that 1.6ns RAM does not need cooling to run at its standard speed from the X800 boards.

 

[radar1200gs]

The ram cooling is there because nVidia is cautious when it comes to design, not because they intended to overclock (though you could argue they expect their consumers to overclock and provide a margin of safety for that purpose). ATi evidently feels differently about ram cooling.

 

[Dave Baumann]

Dave, first, read what I quoted you as saying in the old thread.

Only in your mind was there an implication of cheating there - prior to the release of 6800U there were all kinds of claims of ludicrusly fast RAM (indeed, there you are saying 800MHz), however I had been warning for a long time that RAM supplies are likely to be very limited and you shouldn't expect hugely fast RAM to start with. Look at it now, 2ns GDDR3 is still supply constrianted, let alone 1.6ns - the "depends what you do with the RAM" was merely an indication that you can get slightly better performance if you ship overclocked, which to all intents and purposes they did (according to their, and Samsungs labelling), and still are doing with partner boards now.

Second, nVidia does not label memory, samsung does!

Look at the link - there is an image there of a reference board with an NVIDIA label saying "500MHz Samsung"; if NVIDIA "retested" the memory, why would they then proceed to incorrectly label them again?

 

[gokickrocks]

there is a possibility that nvidia had overvolted (compared to spec), and not necessarily overclocked, the ram that allowed them to have a more stable clock on the memory which would then give off more heat and in turn would explain the extra ram cooling

 

[Evildeus]

I don't see your point. could make it clearer, because i don't understand what you are trying to say :?

What's the point of this thread? Many Board Makers are still OC memory on their cards. FYI
http://www.anandtech.com/video/showdoc.aspx?i=2126&p=2

First up, we have the Albatron Trinity GeForce 6800 Ultra. The bright yellow box advertises 2ns DDR3 Memory; though, at 2ns, 1.1GHz would be a bit of an overclock.

Both cards sport GDDR3 RAM rated at 2.0ns, configured with a bit of a factory overclock. 2.0ns RAM should only be able to hit a 1GHz data rate, but both vendors are running at 1.1GHz data rates.

Given that there are partner boards available at the same time with fully labelled 1.6ns RAM it would appear that Samsung have sorted out their "labelling issue", in which case would it really take two and a half months for these mislabelled modules to pass through, given the sparcity of the RAM availablility?

And if NVIDIA retested the RAM why did they then proceed to label them with the apparently "incorrect" speeds?

And why must everything be a "cheat" to Greg?

 

[dizietsma]

Well someone will have to get their skates on or else this is going to look more and more optimistic

http://times.hankooki.com/lpage/tech/200404/kt2004041917251911780.htm

 

[radar1200gs]

Depends on when the memory was retested. I assume that for the early reference boards nVidia really did think they only had 500 mhz ram to work with.

I'll ask the author of the nvnews post what his sticker says (if present).

In any case remarking a sticker does not equal remarking ram!

Another reason nVidia may run memory hotter than ATi is due to DDR auto-precharge which effectively keeps the memory in an active state for longer (don't know if ATi make use of this or not). It was a large part of the memory incompatability problems seen with nForce1 since nForce also uses DDR auto-precharge and not all DDR memory modules supported it at the time (most should nowadays).

 

[Dave Baumann]

Depends on when the memory was restested. I assume that for the early reference boards nVidia really did think they only had 500 mhz ram to work with.

<smacks head> So to all intents and purposes they sent out boards that were overclocked. If the RAM was actually 2ns RAM then it was overclocked, if it was mislabelled RAM that they thought was 500MHz, then they still send out overclocked boards! I.e. they designed it to cater for RAM that they thought was 500MHz and sent it out at a rate faster than they thought the RAM was specced by Samsung for.

Regardless, this is all academic as partner boards are still being sent out with 2ns RAM – the fact that some boards have 1.6ns is a good indication that if there was actually a “labellingâ€￾ issue, this is now resolved, hence the design of the board clearly had this in mind.

In any case remarking a sticker does not equal remarking ram!

So, what do we have – Samsung apparently getting things wrong and labelling chips incorrectly, then NVIDIA testing the memory and deciding to label them as 500MHz even though they test them to 500MHz, and then board partners proceeding to continue label their boxes with 2ns RAM, and indeed shipping them with 2ns RAM.

If it looks like a duck, quacks like a duck, perhaps it is a duck?

Another reason nVidia may run memory hotter than ATi is due to DDR auto-precharge which effectively keeps the memory in an active state for longer (don't know if ATi make use of this or not).

Given GDDR3 is primarily ATI’s specification, I would say they are using it to the full extent.

 

[Unit01]

oh boy what a thread.
So because NV was providing faulty information ( if that is correct ) on their memory cheaps and people drew conclusions according to those facts at the time.

Then we have new information from two anonymous persons on some internet board. That brings new light to the issue.

The people back then were lying?

Cheating in RAM? wtf are you talking about? you mean overclocking only? but decided to put in the word cheating to make it more spicy and hopefully to those you accuse to feel more ashamed?

If you see RAM chips that say 5ns don't you assume that it's 200Mhz ram?
And if the manufacturer and RAM assembler forgot to re-label them, and you don't know it, are you supposed to to assume like n months ahead of time that "yeah, this ram is probably mislabeled" instead of taking the information that is available at the time?

 

[radar1200gs]

Auto-precharge was introduced with DDR-1.

ATi tells us they design for lower power requirements, which is why I assumed they may not use auto-precharge (would also partly explain why nVidia almost always gets better memory efficiency than ATi too).

EDIT: about the ram and boards shipping with GC20 memory - it's actually GC16 wrongly labeled (according to the poster and the email).

Why don't you try confirming the story with nVidia Dave? or have you burned all your bridges there?

 

[Unit01]

Auto-precharge was introduced with DDR-1.

ATi tells us they design for lower power requirements, which is why I assumed they may not use auto-precharge (would also partly explain why nVidia almost always gets better memory efficiency than ATi too).

EDIT: about the ram and boards shipping with GC20 memory - it's actually GC16 wrongly labeled (according to the poster and the email).

Why don't you try confirming the story with nVidia Dave? or have you burned all your bridges there?

Why don't you provide solid facts instead with solid evidence?
Since it's clear you're the one whose accusing and wanting to make a point?