ATI RV740 review/preview
- Thread starter LunchBox
- Start date
Jawed
Legend
NVAPI allows for the same G-buffer and Z-buffers to be used as the 10.1 path, as far as I can tell. So that bit's accelerated (not having to write a separate Z). I'm not sure if the sampling is accelerated - I don't know if NVidia has a fast fetch like gather in 10.1. I haven't signed an NDA for NVAPI.
But I think the debate was over whether the shadowing IQ is the same on both, rather than performance. Degustator says the shadows are identical on NVidia 10.0, presumably with NVAPI, as on 10.1.
Jawed
But I think the debate was over whether the shadowing IQ is the same on both, rather than performance. Degustator says the shadows are identical on NVidia 10.0, presumably with NVAPI, as on 10.1.
Jawed
Well, for the shadows it's all possible in plain old DX10.0. You just need to take 4 times as many samples. So I don't see the relevance of NVAPI in this discussion. I don't think it has anything to offer to improve the situation. AFAIK they lack a Gather fetch in hardware.
DegustatoR
Legend
No, i'm saying that DX10.1 does nothing in SCS to improve shadow rendering quality - the same quality on Radeons is avialable in DX10 renderer.
Quality of shadows between R and GF are a bit different, but you really need to know where to look to find it (worse filtering of penumbras on GFs is 'hidden' in scene textures most of the time).
Itocp has images of a HIS Radeon 4770.
Looks similar to the engineering sample shown in guru3d, except someone taped a cylon basestar to the top of the cooler for some reason.
Re the GT214, was redone as the GT215.
Looks similar to the engineering sample shown in guru3d, except someone taped a cylon basestar to the top of the cooler for some reason.
Re the GT214, was redone as the GT215.
I *think* I first spread the leak that to Chiphell, which subsequently caught on to mainstream Chinese sites before going full circle to TPU's "I-have-to-make-everything-simple-but-long-for-elementary-students" coverage. 
The same exact 2 pics originated from Lowyat.net's Garage Sales where the 4770 went on sale with a pre-list (rip-off) price of 430 Malaysian Ringgits (120USD).
The same exact 2 pics originated from Lowyat.net's Garage Sales where the 4770 went on sale with a pre-list (rip-off) price of 430 Malaysian Ringgits (120USD).
Oops sorry should have tried a little harder to work out the original source, sorry!I *think* I first spread the leak that to Chiphell, which subsequently caught on to mainstream Chinese sites before going full circle to TPU's "I-have-to-make-everything-simple-but-long-for-elementary-students" coverage.
The same exact 2 pics originated from Lowyat.net's Garage Sales where the 4770 went on sale with a pre-list (rip-off) price of 430 Malaysian Ringgits (120USD).
Looking like they seem to only be releasing the 4770, can't seem to find out anything about the 4750 which seems strange. Guess maybe that SKU with DDR3/GDDR3 is up to the partners to do with their own board design etc...so maybe a month or so before the second parts introduction.
Tried to figure how many units ATI will need for launch, would need at least 200k, probably more likely 500k units?
Working backwards from $99 RRP implies wholesale round $75 taking out shipping packaging(ie box cd-rom cables) of around $10 leaves $65. Now 8 layer board say another $8, cooler $5, board components say another $8 leaves $54 for the chip and the memory. The memory 8x64mb chips @$2.50 each gives $20 for the memory leaving $34 for the gpu and profit.
Looking at TSMC prices, seems to be ~$7k for a 40nm wafer...so say you wanted your chip to be around $20 and at 137mm2 ~500 would fit on a wafer therefore they need around 350 good dies to hit the target. Or a yield of 70%
At 70% or 350 chips per wafer implies need around 1500 wafers or 30% of some 5000 per month they are currently ordering...
(For that size chip 70% yield seems low, but 40nm is a new process and rumors strongly hinting it is bad so its best to be conservative...)
(Above is off the top of my head, feel free to correct, but please dont take too seriously, i only spent a couple of mins writing it)
Here are the latest "leaks". I'm assuming the amount of ROPs in the GPUZ screenshots are off by 8?
A 1Gb version of this card would be nice.
/edit/Overclocking. Not too impressive given the constraints of CCC, but hopefully we see more shortly. When is the release date for reviews etc?
A 1Gb version of this card would be nice.
/edit/Overclocking. Not too impressive given the constraints of CCC, but hopefully we see more shortly. When is the release date for reviews etc?
Last edited by a moderator:
Jawed
Legend
http://en.expreview.com/2009/04/24/first-review-of-radeon-hd-4770-99-monster.html/9
75W less than HD4850 under Furmark. That's pretty remarkable.
Overclocked to 850MHz+. That's pretty nice!
Jawed
75W less than HD4850 under Furmark. That's pretty remarkable.
Overclocked to 850MHz+. That's pretty nice!
Jawed
Yeah no kidding. This is one tidy little package. Should be a screamer on the sales charts. Assuming AMD takes its marketing department down from the attic.
The mobile variant has to live up to a higher performance-number: HD 4860. *joinsfellixrunning*
Similar threads
