HD 4770 Previews/Reviews

[Drunner611]

How else does everyone feel about the 128 bit bus? I don't know if I can get myself away from the 256bit cards.

- opinions?

[Tuvok-LuR-]

I second pinball with Alternate Extraball Rendering

[SKYMTL]

Once again it looks like ATI's PowerPlay is not working...

[SNiiPE_DoGG]

How else does everyone feel about the 128 bit bus? I don't know if I can get myself away from the 256bit cards.

- opinions?

they cant do 256bit because the chip is so small they are pad limited, but the 128-bit bus reduces internal latencies on the board so with the gddr5 it works out quite nicely.

I second pinball with Alternate Extraball Rendering

great stuff

[Lightman]

Once again it looks like ATI's PowerPlay is not working...

So you finally have the card?
When we can read your review?
I personally can't wait to read it! Always top notch analysis

[SKYMTL]

So you finally have the card?
When we can read your review?
I personally can't wait to read it! Always top notch analysis

No. To be brutally honest with you; ATI cut us out of this one for some reason. Be that as it may, I have cards winging their way over as we speak. The review just won't be up on launch date.

As for the PowerPlay issue, the vast majority of consumers and reviewers seem to forget about everything that was originally promised with PowerPlay. We were supposed to get great idle power consumption and temps through what ATI called Clock Gating. However, it seems that neither the dynamic clock speed adjustments nor dynamic voltage reduction has been happening on current ATI cards through the Catalyst drivers. Unless GPU-Z is wrong in the pictures above, it seems that when at idle there are very few clock speed reductions even with this new card.

If you need any more proof of ATI's issues with idle power consumption, take a look at any review out there that pits the higher-end ATI cards against their Nvidia competition. Just omit the 9800 GTX+.

[SNiiPE_DoGG]

No. To be brutally honest with you; ATI cut us out of this one for some reason. Be that as it may, I have cards winging their way over as we speak. The review just won't be up on launch date.

As for the PowerPlay issue, the vast majority of consumers and reviewers seem to forget about everything that was originally promised with PowerPlay. We were supposed to get great idle power consumption and temps through what ATI called Clock Gating. However, it seems that neither the dynamic clock speed adjustments nor dynamic voltage reduction has been happening on current ATI cards through the Catalyst drivers. Unless GPU-Z is wrong in the pictures above, it seems that when at idle there are very few clock speed reductions even with this new card.

If you need any more proof of ATI's issues with idle power consumption, take a look at any review out there that pits the higher-end ATI cards against their Nvidia competition.

well GPU-z is reporting 2000mhz core clock in those screens so I really wouldnt trust it right now

[SKYMTL]

well GPU-z is reporting 2000mhz core clock in those screens so I really wouldnt trust it right now

In the Sensors section, yes. However, the main screen seems spot on to me.

[SNiiPE_DoGG]

well why dont we wait until you get it? seeing as your power reviews are the best there are

its possible ATI said hey, this chip is so low power we dont need to downclock, but we will see, GGDR5 is power hungry....

[ToTTenTranz]

How else does everyone feel about the 128 bit bus? I don't know if I can get myself away from the 256bit cards.

- opinions?

You shouldn't be looking at the bus width but rather at total bandwidth. With GDDR5, the HD4770 will have as much bandwidth as a 9800GTX or a HD4850.

[Lightman]

No. To be brutally honest with you; ATI cut us out of this one for some reason. Be that as it may, I have cards winging their way over as we speak. The review just won't be up on launch date.

As for the PowerPlay issue, the vast majority of consumers and reviewers seem to forget about everything that was originally promised with PowerPlay. We were supposed to get great idle power consumption and temps through what ATI called Clock Gating. However, it seems that neither the dynamic clock speed adjustments nor dynamic voltage reduction has been happening on current ATI cards through the Catalyst drivers. Unless GPU-Z is wrong in the pictures above, it seems that when at idle there are very few clock speed reductions even with this new card.

If you need any more proof of ATI's issues with idle power consumption, take a look at any review out there that pits the higher-end ATI cards against their Nvidia competition. Just omit the 9800 GTX+.

From my personal tests with HD4870 I've concluded that Clock Gating is working as advertised.The problem lies in lack of ability to downclock GDDR5 memory on the fly.

I've did simple test back when the cards arrived. Using my Kill-A-Watt I've first measured idle power consumption of my system with card clocked in CCC 500/900 then I manually reduced core clock to 160MHz leaving memory at 900MHz. This saved me 1W off total system power consumption. On the other hand downclocking memory from 900MHz to 465MHz saved me 40W!

I think the problem with GDDR5 lies in it's specifics like clock retraining for different freq. and only newer memory controller will hopefully fix that. For now every time I'm changing memory clock on my cards screen flickers for a fraction of a second (probably retraining memory clocks during that time).

[SKYMTL]

From my personal tests with HD4870 I've concluded that Clock Gating is working as advertised.The problem lies in lack of ability to downclock GDDR5 memory on the fly.

I've did simple test back when the cards arrived. Using my Kill-A-Watt I've first measured idle power consumption of my system with card clocked in CCC 500/900 then I manually reduced core clock to 160MHz leaving memory at 900MHz. This saved me 1W off total system power consumption. On the other hand downclocking memory from 900MHz to 465MHz saved me 40W!

I think the problem with GDDR5 lies in it's specifics like clock retraining for different freq. and only newer memory controller will hopefully fix that. For now every time I'm changing memory clock on my cards screen flickers for a fraction of a second (probably retraining memory clocks during that time).

Yes and no. The issue is that the Clock Gating was supposed to work for both the memory AND the GPU core. To be honest with you though, I have tried the same test with a clamp meter and a HD 4870 1GB plugged into a PCI-E daughter board and reducing the core clock with the 9.3 drivers (at the time) to 200Mhz reduced the power consumption by a good 25W at idle if I remember correctly.

I think these variances in readings are largely based on the movement from one driver to the next. It almost seems like ATI is still fiddling with the settings every now and then.

[Lightman]

Yes and no. The issue is that the Clock Gating was supposed to work for both the memory AND the GPU core. To be honest with you though, I have tried the same test with a clamp meter and a HD 4870 1GB plugged into a PCI-E daughter board and reducing the core clock with the 9.3 drivers (at the time) to 200Mhz reduced the power consumption by a good 25W at idle if I remember correctly.

I think these variances in readings are largely based on the movement from one driver to the next. It almost seems like ATI is still fiddling with the settings every now and then.

Could be! I will try again now on Cat. 9.4!
It would be strange if Clock Gating would work on some and didn't on other machines. I'm running AMD 790FX chipset. Stay tuned

[largon]

GPU-Z main window reports my HD4890 idles at 750 core 850 mem, eventhough core and mem are both at 250MHz while stock is the usual 850/975.

[SNiiPE_DoGG]

GPU-Z main window reports my HD4890 idles at 750 core 850 mem, eventhough core and mem are both at 250MHz while stock is the usual 850/975.

that's what I suspected but I was not sure because I am not at my computer atm

[W1zzard]

gpu-z realtime clock reading isnt working correctly. also i think the number of rops is wrong.

ati hd4xxx power management has a problem with changing memory clocks: it causes flickering. that's why all (?) production boards ship with memory clock set to the same frequency for all power states

[Wadkiller]

Do you guys think the DD Maze 4 will fit on the 4770 with the stock moun ting mech that came with the Gpu block?

[largon]

The image flicker, is a problem with GDDR5 memory or 700 series GPUs themselves? Does HD4850, too, flicker when memory is downclocked?

I don't think it's a big deal since there's only a fraction of a second lasting image corruption the moment memory is downclocked. It's not like the image would be glitched constantly when memory clocks are lower than stock, no, it happens just for a quarter of a second. Maybe AMD should've just ignored it and set the cards to idle at low mem clocks to cut idle consumption, flicker or not...

Would ofcourse prefer that they'd not flicker.
:P

[Shintai]

I would prefer the flicker when chaning modes rather than high power consumption. Looking forward to the reviews with a good focus on power consumption.

[Mungri]

This looks like a nice card!

Too bad I just scrapped my budget building, but this card looks good to go with low cost dual cores for cheap gaming PCs.

Crossfire is amazing. 16x adaptive AA in everything at 1920x1200 is sweet. And just about every modern game does support it.

[netkas]

http://en.expreview.com/2009/04/24/f...9-monster.html

[Shintai]

GDDR5 seems to kill idle power numbers yet again. And HD4850 looks quite abit better in performance.

[Wadkiller]

WOW, they got teh core to 900mhz without to much hassles, which gave them almost 10000 performance vantage marks. Not to shabby. Just a pity that the mounting holes is spaced 43mm from each other & not 53 like the 48xx series cards, so i won't be able to watercool them.

[Mungri]

GDDR5 seems to kill idle power numbers yet again. And HD4850 looks quite abit better in performance.

But 4770 is cheaper and lower specification, so it is kind of supposed to be slower than the 4850?

Just a pity that the mounting holes is spaced 43mm from each other & not 53 like the 48xx series cards, so i won't be able to watercool them.

That will be a problem, that current aftermarket GPU coolers wont be compatible either

[Tuvok-LuR-]

good article, was a nice read
9800GT is just blown away
Anyway nVidia may still compete well by lowering 250 price a bit (performance wise only, tecnological gap would be quite big in favor of 4770)