ATI RV740 review/preview

Well, a good starting point would be why the 4870X2s have no issues with down-clocking GDDR5, wouldn't it?
 
IMO they could, theoretically, retrain for each of their GPUs and VRAM-segments individually while the respective "other" GPU takes over Desktop.
 
My 4670's RAM clock locks as soon as I turn on a second display. 3850 doesn't mess with the RAM clock at all normally, but if I edit the BIOS so it does change the RAM clock, the second display flickers. These are GDDR3 cards.

Have to say that GDDR5 seems to make things even worse. 4770's idle power is pretty ugly stuff IMO. I was kinda hoping for miracles on the idle power front with 40nm and a GPU that's basically double the frugal 4670. Seems like GDDR5 is most of what messes that up.
 
mczak said:
I guess retraining could be an issue. Does the card actually "redraw" as some mention or just flicker? I'd guess that if retraining would take too long you'd get display buffer underrun (hence screen blanks) since memory isn't available during that time, but I can't see the need for rerendering. Also, display buffer underrun could be avoided by a larger onchip buffer, not sure why AMD wouldn't simply have done that (unless we're talking a long time for retraining - I've no idea how long this takes actually, of course if we're talking tenth of a seconds larger display buffer would be unpractical).
Click to expand...

I've noticed some odd things sometimes...

When downclocking memory for the first time it may take about half a second to a second for it to blank out and redraw everything.

But later when I uplock it back to standard frequency it's more like a flash, IE like it's a split second blank and redraw.

Sometimes it just seems to randomly do the longer change and sometimes it does the faster change.

In all cases however, it's noticeable to some degree whenenver changing GDDR 5 memory frequency for the 4890. I no longer have the 4870 installed to test that.

The length of pause might also have something to do with what I'm running at the time possibly, I really haven't done any extensive tests on it.

Regards,
SB
 
fellix said:
Radeon HD 4770 Media Edition vs. Retail Edition

(a.k.a., you won't get what you see!) :LOL:
Click to expand...

The "media edition" card is the one selling for €109, the "retail edition" is the one selling for €99.
Some reviews clearly say that "you won't see this card under $109"
http://www.hexus.net/content/item.php?item=18202&page=6

and besides the marketing pictures used in the reviews, most sites (like hexus) used the retail card.

The part numbers on the card are also very different 102b/74302 for retail cards and 102b/74305 for the "media" version.
Supposedly that's the card+cooler that comes out once the 40nm production is actually at full tilt.
 
fellix said:
Radeon HD 4770 Media Edition vs. Retail Edition

(a.k.a., you won't get what you see!) :LOL:
Click to expand...
Not exactly news. To summarize, differences I can see:
9 vs 18 solid caps in the power section
8 vs 12 transistors (looks like 2 instead of 3 transistors per phase)
there's some weird looking thing near those green caps in the retail edition I can't identify on those pics (up/right and bottom/right)
Also, fan header is only 2-pin instead of 4-pin (I guess no pwm control then, and I wonder how it's monitored)

These changes could potentially lower OC potential, and power draw could be different as well.

Interestingly, the retail edition is actually the same as that in the sneak preview from guru3d from 2 months ago (save the dip switches on the back and the cooler). I didn't really believe that would be the final board back then, and it still seems a bit odd to me (with all those unpopulated pieces, looks like it could be made at least half an inch shorter easily).
Not really sure what's the point of the reference card actually. Maybe it overclocks better but most review sites didn't bother to exceed overdrive's limit anyway.
 
Last edited by a moderator:
mczak said:
Not exactly news. To summarize, differences I can see:
9 vs 18 solid caps in the power section
8 vs 12 transistors (looks like 2 instead of 3 transistors per phase)
there's some weird looking thing near those green caps in the retail edition I can't identify on those pics (up/right and bottom/right)
Click to expand...
Different inductors... I think.
mczak said:
Also, fan header is only 2-pin instead of 4-pin (I guess no pwm control then, and I wonder how it's monitored)
These changes could potentially lower OC potential, and power draw could be different as well.
Click to expand...

On the other hand, the reail version has 3 more parts located right from the row of memory modules.
 
Can anyone tell from the pics if Tech Report got a 'retail edition'? The Expreview link won't work for me so I don't know what to look for. They could only raise the GDDR5 a measly 40MHz, and the core only 70MHz - not so bad but poor compared to what others have been showing..

Ahah! Expreview finally loaded. TR's card was in fact a retail board........
 
homerdog said:
Can anyone tell from the pics if Tech Report got a 'retail edition'? The Expreview link won't work for me so I don't know what to look for. They could only raise the GDDR5 a measly 40MHz, and the core only 70MHz - not so bad but poor compared to what others have been showing..

Ahah! Expreview finally loaded. TR's card was in fact a retail board........
Click to expand...

Using AMD's Overdrive Auto-Tuning Tool....
That's not oveclocking.
 
homerdog said:
Can anyone tell from the pics if Tech Report got a 'retail edition'? The Expreview link won't work for me so I don't know what to look for. They could only raise the GDDR5 a measly 40MHz, and the core only 70MHz - not so bad but poor compared to what others have been showing..

Ahah! Expreview finally loaded. TR's card was in fact a retail board........
Click to expand...

techreport has BOTH cards :D

http://techreport.com/articles.x/16820/11
 
Last edited by a moderator:
From my own experience using a 4870 and a 4830, the auto tune O/C utility always goes to the limit and then backs off 10Mhz. So the fact that they aren't at the allowed limit doesn't mean it won't go higher.
 
LordEC911 said:
I have never hit my actual OC limits with ANY sort of Auto-Tune/Overclock Utility.
Click to expand...

:oops: When I read this sentence:
Finding the card's top GPU and memory speeds took some time, but when the dust had settled, we had a GPU clock of 820MHz and a memory speed of 840MHz.
Click to expand...
I thought they had spent some time actually trying to find the limit. Upon second reading I see they did no such thing. :cry:
 
40nm Yield Issues confirmed

http://www.xbitlabs.com/news/other/...ion_Technology_Has_Yield_Issues__Company.html

”There have been difficulties with the yields. 40-nm is a difficult technology to manufacturer. We understand the root of the problem,” said Rick Tsai, president and chief executive officer of TSMC, reports EETimes web-site.
Click to expand...
It's also interesting that that article says that 65nm (which I presume also includes 55nm) production accounts for 23% of wafer revenue. Implying that ATI and NVidia graphics chips are no more than 1/4 of TSMC's revenue.

Jawed
 
You must log in or register to reply here.

Similar threads

CarstenS
AMD R9 Nano official specs (and later, reviews)
3 4 5
Replies
90
Views
29K
Deleted member 13524
D
I
Sandy Bridge review thread.
3 4 5
Replies
86
Views
29K
swaaye
S
A
Phenom II reviews
3 4 5
Replies
95
Views
17K
John Reynolds
J
Unknown Soldier
AMD Officially Announces 790GX Chipset
Replies
3
Views
2K
ShaidarHaran
S
Back
Top