4870 Overclocking Thread

[dejanh]

Yeah buddy, I have an agenda. That's why I had the 4870 in my hands at Fry's ready to buy before finally deciding on the 260. The 4870 is a great card....I may go for a crossfire setup some day. Like I said in the other thread, an overheating card would be defective, which is a QC issue. That's different than comparing two cards that are known to be manufactured properly. Nvidia, like ATI are always quick to fix their QC issues. Get your logic straight.

As much as this hurts being that I own a 4870 he's sort of right about this. I do wonder why 4870 VRMs are so hard to cool and why a better solution was not devised. 120C+ temps on VRM are not exactly nice.

On a different note, I asked a Diamond rep what do their engineers use to stress test their cards and his answer was "a couple of loops of 3DMark06". Ouch

My 4870 ran at 1100MHz on stock voltage and stock cooling, looping various benchmarks overnight. Never once had an issue with it the 2 weeks that I owned it. As I stated before, the only way to kill memory is to either cause it to overheat or by increasing the voltage, thus causing it to overheat. Either way, it's not happening without modding the video card which is my point all along. It was GAR and a few other that clearly know nothing about memory that claimed the memory degraded simply by overclocking it. It was BS and I was simply pointing it out.

And for what it's worth, EVGA and XFX are rumored to be leaving Nvidia.

Whenever I hear "overclocking" for me that really implies not just upping the frequencies, but also upping voltages. But that's just because of my typical process in overclocking I do stand by the fact that if you up your core to about 800 and your memory to 1100 you will artifact in ATITool in 45min+ runs and FurMark will likely cause your card to crash. I was able to run looped benchmarks THAT WERE NOT ATITool, FurMark, or 3DMark06 TBTs for hours on end without issues, but as soon as you push the card to the max with one of those three the story changes. Anyway, I told people before to try it out if they do not believe it

One further thing worth noting is that different people have different standards on what they consider stable. For me I want 100% stability, all the time, in anything. Any less is not good enough especially if the product being tested is running at stock specifications (e.g., faulty 4870 XOC BEs).

In terms of the XFX/EVGA, those are just rumors IIRC. I remember reading that AMD will not partner with them for the fear of alienating all of their current manufacturers and making their products essentially obsolete. You can see this happening too because current ATI manufacturers do not come even close to the level that XFX and EVGA are at today.

[Blkout]

As much as this hurts being that I own a 4870 he's sort of right about this. I do wonder why 4870 VRMs are so hard to cool and why a better solution was not devised. 120C+ temps on VRM are not exactly nice.

On a different note, I asked a Diamond rep what do their engineers use to stress test their cards and his answer was "a couple of loops of 3DMark06". Ouch

ATI guarantees their cards not to fail with the stock cooling solution. I fail to see how you can hold that agains them. They engineered and designed it. Are you saying that you know more than the ATI engineers?

[Blkout]

Yeah buddy, I have an agenda. That's why I had the 4870 in my hands at Fry's ready to buy before finally deciding on the 260. The 4870 is a great card....I may go for a crossfire setup some day. Like I said in the other thread, an overheating card would be defective, which is a QC issue. That's different than comparing two cards that are known to be manufactured properly. Nvidia, like ATI are always quick to fix their QC issues. Get your logic straight.

No sir, an overheating card because of a poor design is most certainly not a QC issue. Nvidia has trouble getting good yields. It's rumored to be less than 40% from each wafer. My understanding is that the "fix" right now is that each Nvidia partner is to test each card before it leaves their facility. This is an effort to weed out the poor design flaws from Nvidia.

[dejanh]

ATI guarantees their cards not to fail with the stock cooling solution. I fail to see how you can hold that agains them. They engineered and designed it. Are you saying that you know more than the ATI engineers?

Ok, well that's great. Can you tell that to Diamond then for me please? Because I tried and tried to tell them this, and tried, over and over again. And at the end after going through 5 of their cards with same results myself I decided to give up. Something about keeping the fan at 100%, my room sounding like an airport runway, and still not being able to stabilize their stock XOC BEs did not appeal to me after all those tries...

I think your first sentence should say

ATI guarantees their cards not to fail with the stock cooling solution [at ATI defined stock specifications]

Also, I am not sure what I am "holding against [ATI engineers]". I do not even remember mentioning them. In terms of knowing more than they do, maybe, you never know. I am an engineer too actually and I can guarantee you without a shadow of a doubt that if you think all engineers know their jobs you are sadly mistaken, and on top of that yes we too make mistakes Anyway, who even mentioned engineers...

Oh and I had to add this...

ATI only really engineered and designed the RV770. The card is made up of a crapload of components that ATI had nothing to do with from conceptualization right up to the finished product. They just chose off the shelf components to use to build and run the circuits on their cards. They came up with the reference specification, but that does nothing for the quality of particular components especially in large batches where issues are more prone.

[mega_slayer82]

No sir, an overheating card because of a poor design is most certainly not a QC issue. Nvidia has trouble getting good yields. It's rumored to be less than 40% from each wafer. My understanding is that the "fix" right now is that each Nvidia partner is to test each card before it leaves their facility. This is an effort to weed out the poor design flaws from Nvidia.

Wrong. If it was poor design, then ALL the cards would be overheating which simply isn't true. You're filled with all sorts of faulty logic and FUD, aren't you?

[L-MoS]

Blkout, let me make something clear because I want to put this memory discussion to rest...

Running the memory higher than its maximum specifications will NOT pose a risk to the memory unless you happen to be feeding the memory more voltage as well. However, with increased frequencies you need to up the voltages too to achieve 100% stability in most cases which means that you are now both pushing the memory over specifications in terms of speed AND you are feeding it more volts as a result. Now, it is the "more volts" part that will kill your memory, and depending on how much voltage you are feeding this can happen sooner than later. What I was getting to before is that for 100% stability in every tool you will not be able to simply OC the memory to 1100. You will have to up the voltages as well, which then brings us back to what I was saying about voltages killing the ram.

Here are the Qimonda specs...

http://www.qimonda.com/download.jsp?...rev100_www.pdf

Critical voltage is 2.00V, maximum frequency 4GHz, but you cannot run the ram even close to this without it failing very fast (speaking of volts specifically, read the specs). Probably the maximum safe voltage is 1.75V - 1.8V, but I would not be comfortable running it at even this voltage if I am hoping to have the use of my card for more than 1-2 years.

Is this clear now or am I going to have to repeat myself again? I think people misunderstood what I was saying before but I am pretty sure that even my original post stated speed + more voltage = more heat...implicitly this was supposed to state that such a combination may and likely will result in broken ram. This is XS after all, and most people that post here in overclock and voltmod threads will be posting results and clocks of volt modded cards. This is not a "mellow-systems" forum with a "let's see what our stock card can do" motto. The first thing I did after I got my cards was solder VRs on them so I can adjust voltages



In relation to this, I do not think that NVIDIA has better build quality per-say. However, they do have better partners like EVGA and XFX. This has probably been beaten to death before in so many discussions. I think what makes the NVIDIA guys better is the CS, warranty backing their cards, and looser restrictions on what voids your warranty. They are just more enthusiast friendly basically. That's all. I've had my share of faulty NV cards as well.

Dejanh, I apologize for that buddy... I wasnt directly calling you a moron. It was more of a generalization directed toward the masses that have been making statements like such on topics that are unproven and based more on opinion and saying it like they are fact... You have to realize that people who are more or less new at this are goggling 4870 overclocking and coming here then seeing those commits and scares the out of them... But I'm a stranger here and that just irked me So I'm sorry if you felt disrespected. Hope that's settled then.

Now as you mentioned, In some cases you do need to increase voltage at a point to maintain stability and if you do push it too far, Yes while stable even the memory *can* degrade and fail prematurely, But this is a touchy subject because its difficult to determine for sure if your RAM failing 2 years down the road was because you had it 500 mV out of spec and in the case of the 1100 on GDDR5, *MOST* ppl are not volt modding that do not plan on going beyond 800 while running at 1100, So that statement you made seemed more toward folks using CCC at 790/1100 and were time bombing their card. So I think it's safe to drop this now. You know where I'm coming from and I know where your coming from. Again no hard feelings earlier? I'll admit... that was a bit out of line back there.

I noticed your an Engineer too, Maybe that's why I bumped heads with you earlier! Well I'm an Engineering Technician, low on the totem pole in the engineering dept. But still in the family! And yea your right about engineers not knowing they're job... Our director of engineering is clueless to say the least!

[xMrBunglex]

what happened to this thread? why is Nvidia being mentioned at all in this thread? we shouldn't let arguments from other threads slip into this one and ruin it.

[Blkout]

Wrong. If it was poor design, then ALL the cards would be overheating which simply isn't true. You're filled with all sorts of faulty logic and FUD, aren't you?

Poor design was choosing to use a 1.4 billion transistor chip with low yield rate. This is what causes so many failures.

Give it up man, you have no leg to stand on.

[Blkout]

Ok, well that's great. Can you tell that to Diamond then for me please? Because I tried and tried to tell them this, and tried, over and over again. And at the end after going through 5 of their cards with same results myself I decided to give up. Something about keeping the fan at 100%, my room sounding like an airport runway, and still not being able to stabilize their stock XOC BEs did not appeal to me after all those tries...

I think your first sentence should say

Also, I am not sure what I am "holding against [ATI engineers]". I do not even remember mentioning them. In terms of knowing more than they do, maybe, you never know. I am an engineer too actually and I can guarantee you without a shadow of a doubt that if you think all engineers know their jobs you are sadly mistaken, and on top of that yes we too make mistakes Anyway, who even mentioned engineers...

Oh and I had to add this...

ATI only really engineered and designed the RV770. The card is made up of a crapload of components that ATI had nothing to do with from conceptualization right up to the finished product. They just chose off the shelf components to use to build and run the circuits on their cards. They came up with the reference specification, but that does nothing for the quality of particular components especially in large batches where issues are more prone.

You went through 5 4800 series cards in just over a month? The RMA process at Diamond must be exceptional. That's sarcasm, I'm just not buying your story.

The difference between you, if you're even an engineer, and the ATI engineers is that they have a team of people that designed the video cards and I'm quite certain they get paid more than you do because they are better at what they do. Don't get upset, you know it's true also. I don't think they made a mistake. The 4800 series is an overwhelming success that seems to be a real thorn in your side. Either way, you'll have to get over it and accept the fact that ATI created a brilliant product this generation and Nvidia didn't. Even Nvidia's CEO admitted they underestimated ATI this round. If he can admit it, maybe you should too considering that he probably know a little more about this business that you do.

[B.E.E.F.]

Reported. This is the HD4870 overclocking thread only. No other hardware except the HD4870, and maybe the HD4850 should be mentioned here.

[dejanh]

You went through 5 4800 series cards in just over a month? The RMA process at Diamond must be exceptional. That's sarcasm, I'm just not buying your story.

The difference between you, if you're even an engineer, and the ATI engineers is that they have a team of people that designed the video cards and I'm quite certain they get paid more than you do because they are better at what they do. Don't get upset, you know it's true also. I don't think they made a mistake. The 4800 series is an overwhelming success that seems to be a real thorn in your side. Either way, you'll have to get over it and accept the fact that ATI created a brilliant product this generation and Nvidia didn't. Even Nvidia's CEO admitted they underestimated ATI this round. If he can admit it, maybe you should too considering that he probably know a little more about this business that you do.

I went through 5 4870s in 15 days by returning them to NCIX and not Diamond. Diamond simply said "send it to us for testing and we will get it back to you". Not good enough. I only have 15 days to get a full refund so that's not going to fly.

In terms of them getting paid more than I do and on whether I am an engineer, and/or whatever else you said...

(1) It's foolish of you to assume that you know how much I get paid...
(2) I'm sure they are better at designing video cards. I don't design video cards, but I am both a software engineer and a hardware engineer (major/minor)
(3) 4870 is not a thorn in my side. Only the Diamond XOC BE is a thorn in my side and maybe the overheating VRMs. That's it. Otherwise, 4870 is an exceptional card. In fact, the whole 4800 series is exceptional.

I think you lost track of your opinions somewhere. I feel like you are having an argument with me but I am not sure over what exactly...

[mega_slayer82]

Poor design was choosing to use a 1.4 billion transistor chip with low yield rate. This is what causes so many failures.

Give it up man, you have no leg to stand on.

Which is why there are more 260s than 280s. 280s are low yeild, yes, and many of them have to be dumbed down to 260s. But that doesn't mean the card is going to fail after you get it. If it works, it works....but if it's defective, RMA it. I see your point though...

All: Sorry for yet ANOTHER off topic post. Report me if you wish...

[0ut0fstep]

All of the 4870 threads have somehow been derailed and turned into non-informational pages of ati vs nvidia crap.

[dejanh]

I'll try this again...let see if we can steer things on course this time...

Anybody got their hands on a PowerColor 4870 PCS+ yet? They seem to have modified the power supply system on the card by using 4+2 VRMs as opposed to stock 3+2. Do these play nice? I heard that the initial tested version was blowing up in FurMark just like it siblings do...apparently the VRMs were getting too toasty.

If anybody has the card, what were your results so far in trying to OC it?

Let's please stay on topic now...

[M-XXXX]

27-28C IDLE, 31-32C Load

My card is flying under water. I had problems with 820 mhz with stock cooler, now I dont have any problems with 860 mhz on core in 3DMark06. Problably time modding soon

[dejanh]

Now that's some home-brew cooling Balkan style

Vrlo fino (nice)

[M-XXXX]

I not worry about looking, if it works its fine for me (you should see my bench table)

BTW we are not on Balkan, we are south east Europe

EDIT: What isaverage gain of voltmod?

[mega_slayer82]

All of the 4870 threads have somehow been derailed and turned into non-informational pages of ati vs nvidia crap.

Thank you, Captain Obvious...

[dejanh]

I not worry about looking, if it works its fine for me (you should see my bench table)

BTW we are not on Balkan, we are south east Europe

EDIT: Was in average gain of voltmod?

In terms of what the voltmod will give you, it won't be much. The core will likely not go any higher than 890, and that is if you are very lucky. If you keep the memory lower then the core can be stabilized around 880-890 with a voltmod. Anything over 1.35V - 1.37V on the core will cause the card to become unstable however.

[|Oc|REvO_HsArC]

hi guys

how are these hd4870 watercooled temps? ok?
full load furmark



those vddc temps are pretty high.
I'm using the EK full cover block. The voltage regulator cooler provided kind of sucks.

[xMrBunglex]

hi guys

how are these hd4870 watercooled temps? ok?
full load furmark



those vddc temps are pretty high.
I'm using the EK full cover block. The voltage regulator cooler provided kind of sucks.

it depends on which driver you're using. i think ATI gimped Furmark for the 8.8 beta driver so it wouldn't run as fast because people keep overheating with it. if you're running Furmark at full speed and your VRM's are at 80°, i'd say the VRM cooler is doing a great job.

the easiest way to find out if you're running Furmark at full speed is to rename the .exe and check the framerate. if your framerate is the same with a different .exe name, all is well.

[|Oc|REvO_HsArC]

I'm on 8.7

so, I'm good? great

[M-XXXX]

hi guys

how are these hd4870 watercooled temps? ok?
full load furmark



those vddc temps are pretty high.
I'm using the EK full cover block. The voltage regulator cooler provided kind of sucks.

If you ask me your temparatures suck

[xMrBunglex]

If you ask me your temparatures suck

first of all, there's a difference between Furmark and ATiTool. a big one. second of all, you're using GPU-Z 0.2.6, which doesn't show VRM temps. download 0.2.7 and use that from now on.

[Xcel]

OCed mine some more and seems furmark sable at 825/1025 but only time will tell if it really is stable.