How to find your real GDDR frequency

Many people ask me how do I know the manufacturers frequencies for the GDDR in video cards. Actually it's not that difficult to find out so I thought of writing a small guide so I that I do not have to explain myself every time and also so that others can figure it out easily.

I have to give my thanx to GanGstaOne for helping in figuring this out.

How to figure it out

Every manufacturer of GDDR memories prints extensive information about the memory on the memory itself, this is usually in the form of a code made up of numbers and letters.


In the case of Samsung, the code will look something like this
K4XXXXXXXX-[Suffix]
The Suffix is usually just 2 characters, other times 4 or even more. But almost always, the last 2 characters of the code are the access time.


In the case of Hynix, the code is a bit different.
H5XXXXXXXXX-[XX]X
The [XX] is the access time code, so its not necessarily the last 2 characters. It is usually the first two characters after the dash.
On more recent GDDR the access time code is the first two characters of the second line (the line right below the long code on top).

In the case of Qimonda, its similar to the other two manufacturers but its a bit different for GDDR3 and GDDR5 memories (meanwhile, Qimonda went bankrupt).

Qimonda - GDDR2/GDDR3
HYBXXXXXXXXXXXX-[XX]
Again, the access time code is the number after the dash, the [XX] part in the code above. However, please note that this code may be just one character not two as it is usually for the other manufacturers.

Qimonda - GDDR5
IDGXXXXXXXXXX -[XX]
As above, the access time code is the number after the dash, the [XX] part in the code above.

I do not know any other method of finding this code besides having physical access to the memory module itself.

What is the access time/cycle time?

Simply putting is the fastest time at which the memories are designed to be accessed. This is calculated in ns (nanoseconds). This also means that the access time can be translated into the memory frequency and it is very easy to do this, just divide 1000 by the access time. (Example: 1.6ns means 1000/1.6 = 625 Mhz ). I won't go into details why it is calculated this way.

So now that you know where do find the access time code on the memory, here is a list of what each code means:

Code - Access time - Frequency for Samsung GDDR
25 --- 2.5ns ------ 400Mhz
22 --- 2.2ns ------ 450Mhz
20 --- 2.0ns ------ 500Mhz
19 --- 1.875ns ---- 530Mhz
18 --- 1.8ns ------ 550Mhz
16 --- 1.6ns ------ 625Mhz
15 --- 1.5ns ------ 650Mhz
14 --- 1.4ns ------ 700Mhz
12 --- 1.25ns ----- 800Mhz
11 --- 1.1ns ------ 900Mhz
1A --- 1.0ns ------ 1000Mhz
09 --- 0.9ns ------ 1100Mhz
08 --- 0.8ns ------ 1200Mhz
7A --- 0.77ns ----- 1300Mhz
07 --- 0.71ns ----- 1400Mhz
6A --- 0.66ns ----- 1500Mhz
06 --- 0.62ns ----- 1600Mhz
5C --- 0.56ns ----- 1800Mhz
05 --- 0.5ns ------ 2000Mhz
04 --- 0.40ns ----- 2500Mhz
03 --- 0.33ns ----- 3000Mhz
28 --- 0.28ns ----- 3500Mhz




Code - Access time - Frequency for Hynix GDDR
60 --- 6.0ns ------ 166Mhz
55 --- 5.5ns ------ 183Mhz
50 --- 5.0ns ------ 200Mhz
43 --- 4.3ns ------ 233Mhz
40 --- 4.0ns ------ 250Mhz
36 --- 3.6ns ------ 275Mhz
33 --- 3.3ns ------ 300Mhz
28 --- 2.8ns ------ 350Mhz
26 --- 2.6ns ------ 385Mhz
25 --- 2.5ns ------ 400Mhz
22 --- 2.2ns ------ 450Mhz
20 --- 2.0ns ------ 500Mhz
18 --- 1.8ns ------ 550Mhz
16 --- 1.6ns ------ 600Mhz
14 --- 1.4ns ------ 700Mhz
12 --- 1.2ns ------ 800Mhz
11 --- 1.1ns ------ 900Mhz
N0 --- 1.0ns ------ 1.0Ghz
N1 --- 0.9ns ------ 1.1Ghz
N2 --- 0.8ns ------ 1.2Ghz
N3 --- 0.77ns ----- 1.3Ghz
N4 --- 0.71ns ----- 1.4Ghz
N5 --- 0.66ns ----- 1.5Ghz
N6 --- 0.62ns ----- 1.6Ghz
N7 --- 0.57ns ----- 1.75Ghz
N8 --- 0.56ns ----- 1.8Ghz
T0 --- 0.5ns ------ 2.0Ghz
T1 --- 0.444ns ---- 2.25Ghz
T2 --- 0.4ns ------ 2.5Ghz
T3 --- 0.363ns ---- 2.75Ghz
R0 --- 0.333ns ---- 3.0Ghz
R1 --- 0.30ns ---- 3.25Ghz
R2 --- 0.28ns ---- 3.5Ghz

Code - Access time - Frequency for Qimonda GDDR (Qimonda went bankrupt at the beginning of 2009)
GDDR2/GDDR3
28 --- 2.8ns------ 350Mhz
25 --- 2.5ns ------ 400Mhz
20 --- 2.0ns ------ 500Mhz
16 --- 1.6ns ------ 600Mhz
14 --- 1.4ns ------ 700Mhz
12 --- 1.25ns ----- 800Mhz
11 --- 1.1ns ------ 900Mhz
10 --- 1.0ns ------ 1000Mhz
8 --- 0.8ns ------ 1200Mhz
GDDR5
32 --- 1.2ns------ 800Mhz
36 --- 1.1ns------ 900Mhz
40 --- 1.0ns------ 1000Mhz
45 --- 0.9ns------ 1125Mhz
50 --- 0.8ns------ 1250Mhz


Examples :

The memory of a Radeon 6970M:



The memory of 9800M GTX from a Clevo D900C/M570TU :




The memory of a 8800M GTX from a Clevo D900C:



The memory of a GT 425M MXM card (ASRock) - this is where Hynix writes its timings on more recent GDDR:



The memory of 8600M GS/GT GDDR2 (MXM types):



The memory of a 7800 Go:



Other cards:
ATI M4570: Laptop Forums and Notebook Computer Discussion (thanx to kiwitheiwik)

Why in the end the video card manufacture decides to use other frequencies than the ones specified by the memory manufacturer I have no idea... But I guess that this can give a helping hand in having a successful overclock of the memories.

Known GDDR frequencies:

If you have other pictures of GDDR memories would be great.

Card	Brand	Timing	Frequency	Supplier
Nvidia 680M	MXM Type 3.0b/ ?	04	1250 Mhz (x2)	Samsung
Radeon 7970M	MXM Type 3.0b/ ?	04	1250 Mhz (x2)	Hynix
Radeon 6970M	MXM Type 3.0b/ ?	04	1250 Mhz (x2)	Hynix
GTX 470M	MXM Type 3.0b/Clevo W860/870/x7200	04	1250 Mhz (x2)	Samsung
GTX 460M	MXM Type 3.0b/Alienware	04	1250 Mhz (x2)	Samsung
ATI M5870	MXM Type 3.0b/Clevo W860/870	04	1250 Mhz (x2)	Samsung
ATI M5870	MXM Type 3.0b/MSI GX740	05	1000 Mhz (x2)	Samsung
GTX 285M	MXM Type III /Clevo	1A	1000 Mhz	Samsung
GTX 280M	MXM Type III /Clevo	1A	1000 Mhz	Samsung
GTX 260M	MXM Type III /Clevo	1A	1000 Mhz	Samsung
Quadro FX 3700M	MXM Type III /Clevo/early Lenovo/early HP	11	900 Mhz	Qimonda
Quadro FX 3700M	MXM Type III /later Lenovo/later HP	12	800 Mhz	Samsung
9800M GTX	MXM Type III /Clevo	1A	1000 Mhz	Samsung
9800M GTX	MXM Type III /Clevo - D900C	12	800 Mhz	Samsung
8800M GTX	MXM Type III/Clevo	11	900 Mhz	Hynix
9800M GTS	MXM Type III /Clevo	12	800 Mhz	Samsung
9600M GT	MXM Type II/Clevo/Acer	12	800 Mhz	Samsung
ATI M 4670	MXM Type II/Acer	20	500 Mhz	Hynix
8600M GS/GT DDR2	MXM Type II	25	400 Mhz	Hynix
7800 Go	Clevo M59K	16	625 Mhz	Samsung

 

Nice guide, thanks for the response in the other thread and thanks for this guide as well, +rep.

 

Excellent guide. This is the stuff I like to see. Repped.

 

I updated the post with the 8800M GTX memory frequency.

 

thank you blacky

 

Now here's something worthy of a bookmark.

 

Last weekend I've check the memories of my video card and they are also 1A. I assume that all Clevo 9800M GTX ran with 1A timing as in 1 Ghz.

I've also got the chance to check out an 9600M GT from a Clevo M860 and it has 12 timings that 800 Mhz. The stock frequency for the 9600M GT so there isn't so much room for overclocking.
So I've updated the first post accordingly.

I've also found out something else but I will put that in a new thread.

 

Updated first post with new info on Hynix memories and some older 8600M cards.

 

Very nice blacky

 

Dang - now I wish my memory chips weren't covered by thermal pads - I'd love to find out exactly how fast my DDR2 VRAM is spec'd to go.
EDIT: Managed to find a pic of ONE 9500M GS MXM-II card over at LV2Go - definite Hynix memory, but the code's obscured by the camera angle. Looks like either 22 or 21 ns. No confirmation, though.

 

Can you give a link to that pic ?

 

WOW, excellent info, when I am gonna home I am going to check it out. BTW, I can't overclock my Hynix GDDR2 more than 450MHz without stability problems (my stock clocks 400MHz lower than the one mentioned in this thread but I might have different type).
Question: Is it possible my memory is designed to be clocked at 475MHz but shows stability problems at lower speed?
Between 400-450MHz is fine once I go more than that, even 5MHz, then the graphic card downclocks to 'safety' speeds and I have to restart the laptop in order to be able to overclock the card or even use it at default settings.

 

No, that's not how it should behave. When the memory frequency goes above what it can take, you will get artifacts in games or game crashes but not to safety speed. The downclock of your card is depended on the GPU core and its temperatures not of the memories (there is no way to measure memory temperatures).

Second of all, you should have powermizer working right? And that means that the overclock you apply to the memories will only show in games.

To tell you a strange thing that happened to me. I couldn't get the memories to go above 850Mhz on my card, the computer will crash even in Vista aero with higher clocks. I do not know exactly what I did. But after taking the card out, cleaning the connectors and testing it on a different laptop now the problem has completely disappeared and I can run my memories at 950 Mhz with no problem. If your card acts strange while not gaming, check the current frequencies with GPU-Z. If the current frequencies are not the overclocked ones (let's say just 100 Mhz) then the problem is somewhere else, not with the memories.

As a last thing, you might be having 2.2 nanoseconds access time that is 450 Mhz and then everything should make perfect sense. However, it is very strange that Hynix memories have 2.1 access time when officially they say they don't have memories with such access time.

EDIT: what are you overclocking the card with ? what program ?

 

Great information, reputation jellybean for you.

This might come handy for the unmarked modules i have lying around.

 

Nice, thank you.

I would love to add more cards to the guide. It will save people from having to fry their video memory and/or opening up their notebooks.

 

My 9800M GTS has Samsung memory with the suffix "HC 12", so I'm guessing the GDDR frequency is 800Mhz; which is the stock memory frequency of the 9800M GTS.

 

That is correct. You can still overclock them though, but the chances of damaging them are higher.

The HC in front means it is halogen free and that the maximum supported temperature is 85C.

By the way, I will also add your card to the guide. Thnx.

Does anyone know from what notebook this is?
http://www.ixbt.com/video2/images/g71/pic8.jpg

I've been trying to figure out but I can't. It's where I got the info on the 7800 go. Any help is welcomed.

 

Could it be from an HP Pavillion? I remember the dv8000t used 7600 go and 7800 go

 

I don't think so, because the cards in the picture look like MXM cards.

Given it has an nforce4 chipset, I think it's a D900T but I can't be sure.

 

Excellent guide! Very simple and informative. Great job, Blacky!

 

It is definitely not a D900T, the motherboard is completely the wrong shape.

 

Then I have no idea what that laptop is.

 

I've traced down the b@stard in the :
http://www.ixbt.com/video2/images/g71/pic8.jpg

It's a Clevo M59K
Here's the pic:
http://ztronics.com/components/com_virtuemart/shop_image/product/PROSTAR_5989__59_497a6d506c336.jpg

 

Yep, looks about right

 

I now have undeniable PROOF that Hynix chips are sold in 2.1 ns varieties. Take a gander at this image of an Acer 8600M GS.
That memory chip message doesn't get any clearer.

 

That is actually an access time of 2.5 ns . Sorry. Remember, it's the first two characters after the dash .

EDIt: I think I might have made a small mistake in my initial post. I will correct it later. I must apologise for this.

The MXM II DDR2 memories are 400 Mhz, not 475. Damn, why was I so stupid not to see that.
I've updated the initial post.

 

Ah. That would be my bad.

 

Sir Blacky. I have recorded the numbers for my memory on my GTX 280M. I am very surprised to find that the memory on the GTX 280M is the same as the 9800M GTX; the suffix ends in 1A.

9800M GTX


GTX 280M

 

Thank you so much. Yes, it looks like the 280M uses almost the same GDDR as the 9800M GTX. Although they are different, you can notice the model number is different 858 vs 304. but I am not sure what is the difference unless I can have the full code of the 280M GTX because I can't read that from your image.

Anyway, thanx again.

I will update the post with the 280M GTX and it's max frequency, 1000 Mhz.

 

Yes, I noticed that also. Sorry, but the text is very faint on the chips. However, the full code reads as K4J1032400 HJ1A on both cards.

 

If it's the exact same code then it means they are most likely identical. So... means we can safely push the memories in the 280M up to 1Ghz. Nice.

 

Updated with memories of the 260M. Also Samsung 1Ghz max.
http://forum.notebookreview.com/attachment.php?attachmentid=34340&d=1243260218

 

Quadro FX 3700M, Qimonda chip 1.1ns = 900MHz.

 

Ah, nice. I will have to add Qimonda codes now .

Maybe you have a picture of the memory chip.

 

Just as a note for GDDR manufacturers worldwide.
Qimonda went bankrupt in February 2009.

http://www.computingsa.co.za/article.aspx?id=1001876

Now only Hynix and Samsung are left.

 

Maybe I am asking a stupid question. If I have a 8600m GT with ddr2 it has 8 slots of memory qimonda 32mb each one at ddr2 (25) 400mhz, and If I get some VGAram ddr3, and a qualified person change them, It would work? Even if I change the bios info from one that has natively gddr3?

I asked this because I have two 8600MGT, one new working in my laptop, and another that its broken( Memories burnt)

 

Yes, if you get a qualified person to change the GDDR and you flash the video BIOS afterwards to reflect the higher GDDR3 clocks ... there is no reason why it should not work. Just be careful cause GDDR3 will get a bit hotter, so make sure you have good memory heat pads.

 

Thanks. Just another thing, its about the broken gpu. The symptoms that shows are this: Wrong colors, green vertical lines when screen is black or dark. And sometimes system doesnt starts with it.
Just to know if the core is damaged too or not with this symptoms.

 

Looks like a typical case of bad video memory. Impossible to tell if the core is affected or not.

 

Hi again does you know online stores to find them?

I found this in ebay:

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=150378896372&ssPageName=ADME:X:RTQ:US:1123

 

The codes only indicate their normal speed, not their max speed, correct?

I have the samsung k4j1032400-hc12 which normally is 800mhz, but I have it OC to 900mhz without any problems.

 

Sort of. They indicate the specified manufacturer frequency. As long as you are within their specifications your GDDR should operate normally and no damage should come to it. Of course they could go higher, but this would be a bit more risky given that the GDDR is generally the first to fail in a GPU, that's all.

 

http://forum.notebookreview.com/showpost.php?p=5264325&postcount=158

 

From my experience I could disagree with you on that. I had GPU failures without any heat problems... just because I've pushed the GDDR frequencies to high. Heat does shorten the life span of your GPU indeed, but is just one of the factors.

 

Pushing the memory above manufacturer spec is indeed a key factor in why GPU memory fails, but as to what extent also plays a role. If the memory is only clocked slightly above rated spec, say 1025MHz on 1000MHz rated memory, then it's not too bad. But if one pushes the memory out of proportion, then yes, the GPU memory will be the main cause of failure, and even moreso when thermally mistreated.

It seems from your instances of GPU failure tends to refer to the quality control aspect as well. There are a bunch of users on this forum alone that have pushed their memory past max rating on both the minor and major level, and under high thermal stress, and their GPU's are still alive and kicking despite months of long term use. I don't know what GPU's you have or how far you pushed them beyond their maximum rated frequency, but all three of the said causes for failure are present and should be handled responsibly.

 

I suppose then that we do have an agreement.
As regarding to failure, we both know that not one GPU is identical to another. As a result one may take very high clocks another might not, it's always a gamble.

If I recall correctly, I pushed the GDDR of a 7900 GTX Go to about 720 MHz (I was doing it incrementally) and monitoring temps trying to find how high it will go. When I hit 720 system froze and it took about 6 months for the GPU to completely die (heroically I might say). It happens, what can I say, but this is why now when I OC my GDDR I am extra careful.

 

The 7900 mobile cards had their problems (i had a 7900 gs go). But newer cards are a lot better.

I don't think a 100mhz OC is damaging.

 

I used to overclock and overvolt my 6600 Go to 400/400 from 300/300 straight since July 2005 to February 2009 on my M38AW. The memory is rated for 300MHz, so I was running ~33% above rated spec. To this day, the memory is still error free and the GPU still hauls *ss in CS:S. Hell, I'm typing on it right now.

 

You going to sell it soviet?
I myself would keep it, it seems like it served you well.

 

There are some things that I have sentimental value for. In regards to technology, my D400V and M38AW are my sword and shield. They have indeed both served me well and I would never sell them. A lot of people were spellbound as to why I sold my M570TU. My M570TU was the big exception because I was offered such an amazing deal for it. Also, I never did anything momentus on it; all the hardware and software achievements that I have made on that notebook were merely evolutionary rather than revolutionary. Yes, I do miss it as it was my first notebook bearing a top shelf mobile GPU, but that's really nothing compared to the vast amount of knowledge gained from research and experimentation on my D400V and M38AW.

Plus, I need a backup notebook right now as the second wave of midterms are approaching very soon before Thanksgiving break.