Important: Ati 4850 Temps Too High?

[wevsspot]

Angry,

 

When you expecting your Accelero S1 rev2 to arrive. I'm looking forward to some reports!

[ir_cow]

these stock fans suck, at 60% it sounds like a cd being burned at 40x and 100% sounds like a airplane. just need a few more dollars and i'll get my DUOrb.

[Bubbaxm2]

EDIT: Found Solution for both cards...

 

Now its getting this whole bit to work with Automatic instead of Manual...

 

EDIT Again: Finally was able to get both cards to work with Automatic, and once CCC starts up both cards run at the desired speed, without selecting the profile.

 

 

Awesome fix. :thumbs-up:

Well what was your fix to get them to run faster in automatic? I just got both of mine hooked up and edited the xml file. Damn these things are loud!

[Waco]

The cores will not use a bridge chip to communicate, so there will be no microstuttering

 

They share a memory cache, so frames are delivered smoothly - even using AFR rendering.

Uh, what?

 

The 4870X2 does indeed use a bridge chip and the GPUs definitely do not share memory.

 

I'd attribute any "microstuttering" as you call it to CPU, driver inefficiencies, or lack of RAM before blaming the architecture of the cards...

[suedenim]

So anyone got a cooler modified 4850 running at 4870 speeds yet?

[watsonte]

Well what was your fix to get them to run faster in automatic? I just got both of mine hooked up and edited the xml file. Damn these things are loud!

 

When originally editing the XML file you have to set it to Manual, once enabled, meaning the fans are running at the desired speed, go back into the XML file and set it to automatic.

 

At that point you will hear the ramp up in speed if necessary...atleast thats what Im hearing. Although its only been one time. For the most part at 38% Speed my cards never hit over 79-81C which is what they originally idled at.

 

I really have no idea what the limit is on these cards but it must be high

[watsonte]

So anyone got a cooler modified 4850 running at 4870 speeds yet?

 

Are speeds all that relevant considering one is using GDDR3 and the other GDDR5?

 

Can you really make up memory bandwidth by increasing speed?

[suedenim]

Are speeds all that relevant considering one is using GDDR3 and the other GDDR5?

 

Can you really make up memory bandwidth by increasing speed?

 

So now we have 2 questions.

 

1) Any 4850's at, or near 4870 clocks?

2) Does GDDR5 beat clocks?

[Angry_Games]

GDDR5 doesn't have more memory bandwidth, that's all up to the memory interface (ie 128-bit interface isn't nearly as good as 256-bit, etc).

 

GDDR5 does have lower latency and higher attainable clocks though.

 

If you can get GDDR3 to run at the same clock speeds as 5, then you could make up a good amount of performance, but it would still depend on whether the DDR5 was running on a higher bus (like 384-bit, or 512-bit, etc).

 

 

as for #1, haven't even put my Accelero cooler on yet. Getting there today as got to get a customer's rig built which means 1/2 of my rig = gone (sold half my rig)

[InCrYsIs]

I'm just using a couple of Vantec slot coolers. No heat problems at all.

[wevsspot]

In the process of reviewing an Abit motherboard, I picked up a Gigabyte 4850 to use as the test card. This card is stable at 740/1025 using the OEM fan. I used RBE version 1.12 to adjust gpu and memory clocks, fan speed profile and 3D voltage, saved and flashed the modified BIOS. GPU voltage is set to 1.20v for 3D operation. GPU idles at 45C and the hottest I've seen it during a one hour run of RTHDRIBL is 71C. I know that 3D marks don't really tell us much, but this card scored 13006 in 3DMark06 at these settings.

 

One thing I've found for sure is that you need a very capable processor or you WILL bottleneck this video card. With the E2180 at stock clocks 3DMark06 scores were paltry at the 8700 point mark. Once I bumped the processor to 3.0ghz and above the 4850 took off like it had wings. Currently the E2180 is running at 3.2ghz and that's about as high as I can get it OCCT stable on this Abit board.

 

Memory bandwidth on this card isn't anything to write home about, limited by the 256mb bus. Here are the performance numbers of the card running at 740/1025;

Pixel Fill Rate = 11.8 GP/s

Texture Fill Rate = 29.6 GT/s

Bandwidth = 65.5 GB/s

 

As reported by GPUz version 0.2.5

 

Just for grins and giggles I ran the GPUz numbers on my 8800GTS 640MB card and here is what GPUz tells me (this card is clocked at 600/1800)

Pixel Fill Rate = 12 GP/s

Texture Fill Rate = 28.8 GT/s

Bandwidth = 72 GB/s

 

Now I'm not saying that the GTS is anywhere near the card that the 4850 is because I haven't had a chance to put the 4850 into my main rig and run gaming numbers between the two cards. So don't start blasting me about how much greater the 4850 is. I'm only reporting the numbers that I've had a chance to put together so far. The results will all wash out in the end one way or another, because to me the GPUz or 3DMark numbers don't mean a hill of beans. I'm interested in the gaming frame rates.

[Waco]

GDDR5 doesn't have more memory bandwidth, that's all up to the memory interface (ie 128-bit interface isn't nearly as good as 256-bit, etc).

If you're comparing GDDR5 versus GDDR3 on a 256 bit bus (like the 4870 and 4850 have) the GDDR5 does indeed have a MASSIVE advantage in memory bandwidth.

 

Hell, the 4870 has damn near as much bandwidth as my 2900XT 1 GB. Of course, unlike my 2900XT...it can actually make use of it.