Print 47 comment(s) - last by Darkskypoet.. on Feb 26 at 6:46 PM

RV740  (Source: Guru of 3D)

  (Source: Guru of 3D)
Benchmarks place it comfortably between the HD4830 and HD4850

Despite an uncharacteristically tight-lipped stance from ATI on its first 40nm parts, hardware review website Guru of 3D has managed to secure a sample of an RV740-based graphics card and put it through the paces. At first glance, the specifications suggest that it is little more than a shrunken RV770LE core, but a closer look reveals several improvements

According to the preview published today, the GPU boasts 640 shaders, 32 TMUs, and 16 ROPs; all the same as the RV770LE. However, the core and memory clocks both receive a significant bump, up from 575MHz to 650MHz and 1800MHz GDDR3 to 3200MHz GDDR5 respectively. The substantial boost in memory frequency works to offset the performance difference incurred by the smaller 128-bit memory bus on the RV740, which brings the reported math processing rate for the GPU up to 900 GigaFLOPS.

Every single benchmark result published in the review places the RV740-based card right between the existing Radeon HD4830 and HD4850 graphics cards, also from ATI. For example, at 8xMSAA and 16xAF in the popular first person shooter Left 4 Dead, the RV740 turns in 28 FPS, flanked by scores of 25 and 30FPS from the two cards at 2560x1600 resolution. The card does exhibit a significant performance drop at higher resolutions, likely the result of the 128-bit memory bus width and the relatively low 512MB of memory. Other reported benchmarks include Far Cry 2, Crysis WARHEAD, Call of Duty 5: World at War, Brothers in Arms: Hell’s Highway, and 3DMark Vantage.

MSRP for RV740-based graphics cards is expected to be $99 USD, which should make it a very attractive offering in the upper-mainstream segment. While the official name of the video card has not been released, the author of the article suspects it would be called the Radeon HD4750, which would be a logical name for a card with the given characteristics.

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

Bandwidth Misconceptions
By shabodah on 2/25/2009 5:04:49 PM , Rating: 5
I've been fed up with hearing about 128-bit memory "causing" issues since the 7600GT. A 128 bit card running 3200mhz GDDR5 has 77% of the bandwidth of a 256 bit card running GDDR3 at 1800mhz. Yet the this new card is still faster than the older one because the core is faster. Furthermore, 512 bit cards with 512MB memory are running out of steam at high resolutions just as much as the 256 bit and 128 bit ones are. Let's all just get bandwidth straight and quit blaming it when it simply is not at fault.

RE: Bandwidth Misconceptions
By davekozy on 2/25/2009 6:02:36 PM , Rating: 3
The 512 bit cards GTX 280 and GTX 285 have 1GB of memory not 512MB. Cards with only 512MB, none of which have a 512 bit bus as far as I know, don't perform well at high resolutions because they don't have enough memory. Not sure how much bandwidth affects performance at high res compared to running out of memory.
128 bit is probably sufficient if you run the memory fast enough. 512 bit 2500MHz DDR3 on the 285 is as much bandwidth as 10000MHz DDR5 at 128 bit!

RE: Bandwidth Misconceptions
By shabodah on 2/25/2009 6:43:52 PM , Rating: 2
Look into some older generation highend cards or workstation cards. Regardless, you are agreeing with me. Even the 2GB GTX285 card is showing significant gains at high resolutions over the 1GB GTX285 cards, and thats with no other variables to account for.

RE: Bandwidth Misconceptions
By The0ne on 2/25/2009 7:35:48 PM , Rating: 2
Or can head over to Anandtech to read one of their latest articles on video card. Multi-GPU and 1Gig+ memory :D

RE: Bandwidth Misconceptions
By oopyseohs on 2/25/2009 6:38:27 PM , Rating: 3
You're absolutely right. However, I suspect a similar card with 512MB of GDDR5 at 3200MHz pumping through a 256-bit bus would hold a pretty good advantage over this 128-bit one. Furthermore, that extra bandwidth would probably help to keep frames high at the larger resolutions. Certainly we shouldn't be "blaming" the bus width on this decrease, but you have to agree that it does have an effect.

RE: Bandwidth Misconceptions
By Alexstarfire on 2/25/2009 11:22:05 PM , Rating: 1
No..... just no.

RE: Bandwidth Misconceptions
By Darkskypoet on 2/26/2009 6:46:09 PM , Rating: 2
It doesn't matter the width of the memory channels in question if you have to hit main system memory because you run out of local Vid Memory! Ever hear thrashing hard disk sounds? You could have the worlds fastest 1gb of system memory... If you require 1.5gb your hitting the next slowest part of the chain. (ie. Hard drive thrash) PERIOD.

This chip is in place to replace the 4830, I bet you most 4830s out there are actually good chips that could be made into 4850s at least, if not 4870s. This chip is there to essentially stop them from having to give away good cores at such a deep discount, further the 40nm process with TSMC is leaky right now, so that's another reason why no point in going high end yet.

Go high volume, and thrash the ASP of Nvidia's upper mainstream. AMD did it to the 8800gt, 9800 gt, and 260. Why stop now? Very good performance, I'd imagine adequate overclocking, but more importantly CHEAP .

A Cheap as borscht 4830 replacement with an option to twin it with ddr3 for the 65$ range. why waste, or even mfg another 48xx core for a 4830-esque product, when you don't have to? There has been more then a few reviews pointing to the fact that the 4830 is great, if it were a little bit cheaper. Well, welcome to that part.

"We shipped it on Saturday. Then on Sunday, we rested." -- Steve Jobs on the iPad launch

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki