Are RV770 Cores Cherry Picked for Radeon HD4870's ?

[Big.Wayne]

Hello Xtremers,

I know its early days but a few of us across the pond were wondering if ATI have done any speed-binning on the RV770 GPU cores and cherry picked the best ones for use on the HD4870 cards?

I've not had enough time to read through the gazillions of posts here to view the overclocking results from both HD4850 and HD4870 and see how high these new cards can go?

Thanks in advance

[zerazax]

I'm sure that while they are probably higher binned, they already have higher voltages set than the RV770PRO and their BIOS is twice the file size to handle GDDR5...

[metro.cl]

Sure they bin the cores but saying Cherry Picked is too much.

[cadaveca]

Binning based on memory control, not max clocks?

[Big.Wayne]

Ok thanks guys, soz if I got the wrong jargon but there is a fine line between the meaning of speed binned, Cherry Picked and Golden Sample!

I'm gonna make another coffee and search for the eye-drops before trying to see how well the cards are clocking on the GPU, hopefully a database will magically appear soon! *cough*

[metro.cl]

Ok thanks guys, soz if I got the wrong jargon but there is a fine line between the meaning of speed binned, Cherry Picked and Golden Sample!

I'm gonna make another coffee and search for the eye-drops before trying to see how well the cards are clocking on the GPU, hopefully a database will magically appear soon! *cough*

I wouldnt say so, Bining is when you go for lots of sales and high volume, nothing really complex here, you get normally 2 or 3 bins.

Cherry picking i whn you go to the best bin test a lot of units and come up with the best one, that is or can go far beyond the average of the high bin.

[xMrBunglex]

i'm not sure if there is any consensus on what the 4850 cores are capable of attaining yet because none of the third party OC apps we use are compatible with the new cards yet. don't quote me on this, but i think most or all 4850's are able to hit the CCC Overdrive limit. all those people need is an ATiTools update, some pencil mods, and aftermarket cooling to really see what they can do.

[Big.Wayne]

metro.cl: Understood, many thanks!

xMrBunglex: you are of course quite correct . . . what was I thinking!

[alexio]

I'd say most RV770 meet the RV770XT specs but ATI puts them on the slower HD4850 cards to meet demand. I'm not sure if all chips are tested at XT specs, probably not, but I'm pretty sure that ATI could qualify a lot more of them if they wanted to.

[Crankybugga]

I'd say most RV770 meet the RV770XT specs but ATI puts them on the slower HD4850 cards to meet demand. I'm not sure if all chips are tested at XT specs, probably not, but I'm pretty sure that ATI could qualify a lot more of them if they wanted to.

My guess is that they are all same binned gpu's, only the PCB changes and thus voltage supplied, the 4850 still have to be 100% fit even though slower.

[AliG]

I'd say most RV770 meet the RV770XT specs but ATI puts them on the slower HD4850 cards to meet demand. I'm not sure if all chips are tested at XT specs, probably not, but I'm pretty sure that ATI could qualify a lot more of them if they wanted to.

that's my bet, and they just set the voltage lower to prevent people from reaching xt speeds on the pro without some modding.

And also Golden chip is a chip that was produce from regular binning but just happens to be amazing, it is completely based off luck and thus is a golden chip instead of a cherry picked chip. Most of the time you'll hear about a golden bin more than just a golden chip, if one chip happens to be unusually amazing (in terms of ocing and needed voltage) then in general the rest of that week's batch will be around that same level of goldenness

[gurusan]

the RV770XT chips get about .11V more than the RV770Pros...so this leads me to believe that no they are not speed binned.

[Big.Wayne]

Alright gurusan, fancy bumping into you here in America!

If you have time can you post up some of your HD4850 results here please!

[Jakalwarrior]

Since they aren't able to make enough 4870s because they don't have enough memory my guess would be that most cores are never tested to see if they can hit 4870 speeds, so the 4850s should have a good chance of being great instead of being picked for suckiness.

[Miwo]

thats a good point Jakal, DDR5 is in short supply. I suppose it wouldnt make business sense to keep a whole bunch of XT cores waiting around in inventory while waiting for the DDR5 to be in supply. They could be used down the road in the 4870x2, but those cards make up very few % of the market. 4850s are by far the biggest seller

Though I would expect that there has got to be some testing / slight binning involved. Only the chips that can do 750MHz at the lowest voltage should make their way into the 4870. But hey, you never know - I'm certainly not complaining if my RV770Pro chip is just as good as an RV770XT chip

[Big.Wayne]

Thanks for good replies all!

I'm sure it will become apparent as more people do some testing and share their results, hopefully RivaTuner or similar overclocking tool gets an update soon to make life easier!

[AliG]

Since they aren't able to make enough 4870s because they don't have enough memory my guess would be that most cores are never tested to see if they can hit 4870 speeds, so the 4850s should have a good chance of being great instead of being picked for suckiness.

yeah hopefully, I just ordered a 4850 and can't wait for it to come in, and honestly I don't think the extra bandwidth is doing all that much, with the 3800 series it was proved voltage really was the only difference between the rv670xt and rv670pro (the pro with vmods could hit 1ghz) and would perform the same at the same core clock (which meant gddr4 was useless). I'm hoping that once the new rivatuner comes out and we can raise the voltage a bit I can get my 4850 to 800 core, with better cooling of course

[gurusan]

you should have no problem reaching 800 core. I'm currently at 800 stable with 1.284v

Trying to see how much I can push the card on lower voltages (under 1.35ish)

[Big.Wayne]

Hey I just had a thought!

The HD4850 is powered by a single 6-pin PCI-E power lead and the HD4870 is powered by two 6-pin PCI-E power leads. Could this mean that depending what extra volts the GPU needs to overclock madly that the card featuring two separate power leads will do better?

I know you can volt-mod the HD4850 but the card can only be supplied so much power due to having one connector? I read somewhere that when the pre-overclocked HD4850's start rolling out they will also feature two 6-pin PCI-E power leads so the card is getting enough juice to run the overclock?

And to follow this train of thought, if indeed the cards receiving the most power are the ones overclocking the best, does this mean that HD4850 owners using a full PCI-E 2.0 spec motherboard have a slight overclocking edge on those using older PCI-E spec boards?

I believe the PCI-E 2.0 spec increases the amount of power the card can draw from the actual PCI-E slot (is it 75w?).

[alexio]

A 6-pin power connector can only supply 75w officially, but in reality it will supply about as much as the PSU can. I'm not sure if the PCIe has a hard limitation, but I doubt it, so in reality it should be able to supply over 75w as well. The difference between the RV770XT and RV770Pro PCB is merely that the XT PCB has better (beefier) components that transform the 12v input to vGPU/vMEM/etc values. Keeping the temperature down of the PWM components on the HD4850 cards should ensure that up to 1.5v is pretty safe. You could add a few capasitors to get even cleaner power but it probably makes hardly any difference.

[Big.Wayne]

A 6-pin power connector can only supply 75w officially, but in reality it will supply about as much as the PSU can. I'm not sure if the PCIe has a hard limitation, but I doubt it, so in reality it should be able to supply over 75w as well

Hey alexio, thanks for reply. I have no engineering knowledge but I'm good at asking the right questions!

If what you say is true then why does the HD4870 need two six-pin PCI-E power leads? Are they playing safe? kinda keeping within an agreed spec?

Keeping the temperature down of the PWM components on the HD4850 cards should ensure that up to 1.5v is pretty safe

Again I am not sure what the standard voltages are for this GPU? are you saying that 1.5v is gonna be as much juice as the RV770 cores need to really fly? and this can be achieved with soft-mods/hard-mods on the HD4850 cards?

[alexio]

Hey alexio, thanks for reply. I have no engineering knowledge but I'm good at asking the right questions!

If what you say is true then why does the HD4870 need two six-pin PCI-E power leads? Are they playing safe? kinda keeping within an agreed spec?

Yes it's the official spec. And of course in the states you wouldn't want to risk being sued if someone burns his house down because he has a PSU with really crappy wiring.

Again I am not sure what the standard voltages are for this GPU? are you saying that 1.5v is gonna be as much juice as the RV770 cores need to really fly? and this can be achieved with soft-mods/hard-mods on the HD4850 cards?

Stock is 1.16v. If you keep the PWM components cool they can easily supply much more than this. For 24/7 use I wouldn't use 1.5v, but it probably wouldn't do any harm to the card as long as the components stay cool.

[Big.Wayne]

[sound of brain clicking]

Ok great thats useful info and it does make the choice harder for people between the two cards.

Do you happen to know how its possible to increase the voltage to the GPU core on the HD4850 cards? I read some people are doing a pencil mod? and if so what kinda extra volts does this bring please?

[alexio]

HD48*0 vmod thread

[AliG]

When can you raise the voltage through software? I want to do some serious ocing (I always have), but I'm not risking my investment by doing a vmod, not yet at least, maybe I'll practice first and then try it, but I only upgrade my gfx every 3 or so years since I don't have enough money to justify on gaming