$99 MONSTER!!!
source: expreview.com
GPU production nodes surpasses CPU cycle for the first time
It seems that the conversion of GPU’s manufacturing process has been lagging behind CPU cycle, but this history is going to be rewritten soon, as AMD debuts the world’s first graphics processor, RV740, which takes advantage of 40nm processing, which means GPUs will overtake CPUs for the first time in history of IT in terms of production nodes.
Back in June 2007 when the process was about to see the transition from 65nm to 45nm, Taiwan Semiconductor Manufacturing Company (TSMC) announced that they would kick off mass production on 32nm in 2009. Naturally, TSMC should have manufactured 40nm products as the transition between 45nm and 32nm sometime in 2008, but driven by the pressure from the competitors such as AMD, IBM and Intel, TSMC later claimed that they would skip the 45nm and offer 40nm technology as the first one.
As one of the most adventurous partner working with TSMC, AMD has been brave enough to take the lead in implementing 40nm technology, though the real product comes a bit later than expected.
According to TSMC’s internal roadmap, the company will offer three different production nodes. CLN40G is the general purpose process and will be used by the GPU manufacturers. CLN40LP is the low-power process and will be used for the production of notebook derivatives of GPUs manufactured in the general purpose process. The third 40nm node is CLN40LPG, which is targeted at manufacturing chips for handheld devices. So expect Nvidia’s Tegra chips to shrink all the way down to 40nm.
TSMC has spent $590 million on their equipment since the end of February to expand the 40nm node at its 12-inch facility, and they would kick off mass production on 40nm nodes in the second half of this year at Fab 12.
Though the manufacturing process is just one of the decisive element of GPU technology, but it does mean a lot for AMD to break Intel’s dominant position in semiconductor in production technology for years, in which RV740 has played a principal role.
The final product based on RV740 is branded as Radeon HD 4770 which claims to be the fastest graphics card under $100. You can expect the performance edition which is named as Radeon HD 4750 later. According to AMD’s roadmap, RV740 will replace the existing RV770LE (Radeon HD 4830).
Following RV770’s structure, RV740 actually features the same stream processor count as RV770LE. RV740’s 640 stream processors consist of 128 Stream Processing Units (SPU), and each 16 SPUs form a SIMD (Single Instruction Multiple Data). Therefore, there’re 8 SIMDs in total and each SIMD contains 80 stream processors.
In order to reduce cost, AMD has slashes the memory interface of RV740 to 128-bit, while adopts GDDR5 memory to make up for the bandwidth loss.
GPU-Z fails to detect the correct memory clock, making the memory bandwidth higher than actual.
Thanks to the application of GDDR5, the data rate achieves 3200MHz though the actual memory clock is only 800MHz. Radeon HD 4770 delivers 51.2GB/s memory bandwidth, compared with 57.6GB/s on HD 4830. HD 4770 features GPU clock of 750MHz, making its computing speed achieve 960GFlops, seeing 30% increase on HD 4830 and almost catch up with 1000GFlops on HD 4850.
RV740 comes with 16 (4 groups of) Render Back-Ends Units. According to previous structure, each group of RBE is bundled with a 64-bit memory controller, so we’re confused to see the 4 groups of RBE paired up with 2 memory controllers only. We’ll try to figure it out and let you know if we find something.
Radeon HD 4770 supports DX10.1, PCI Express v2.0, UVD2 and so on.
What does 40nm processing bring?
Smaller die size and lower production cost
40nm processing means higher-density transistors on the die. Theoretically, the size should be at least 40% smaller than 55nm, and the production cost for each unit also drops.
RV740 measures 12.3*11.5mm (140mm2), much smaller than 260mm2 of 55nm RV770. The reduction of transistor count also contributes to the decrease of die size, but the technology improvement obviously counts more.
The transistors connect more closely due to the new technology, and thus the power consumption and heat reduce. The maximum power consumption of Radeon HD 4770 is 80W according to AMD, compared with 110W on Radeon HD 4830 (RV770LE).
40nm GPU can work at higher clocks than 55nm, so Radeon HD 4770 sees performance improvement.
RV740 adopts 2nd-generation GDDR5 memory
AMD launched Radeon HD 4870 last June, making it to be the first one to adopt GDDR5 memory on PC graphics card. And earlier this month AMD rolled out with Radeon HD 4890 which also features GDDR5. Both HD 4870 and HD 4890 are AMD’s flagship products, so it’s amazing to see HD 4770 which is aiming at low-end and mid-segment also uses GDDR5.
GDDR5 costs more than GDDR3, but GDDR5 can simplify the PCB wiring and etc, so it doesn’t necessarily costs more in general view.
What benefits does GDDR5 bring?
With 1.5V voltage, GDDR5 causes less power consumption than GDDR4/GDDR3. The FBGA 170 packaging helps to reduce die size and lower the heat.
GDDR5 highlights
GDDR5 combines three concepts including Data eye optimization, Adaptive interface timing and Error compensation. Data eye optimization has key features such as data/address bit inversion, adjustable driver strengths/voltages/terminations, and it brings fast interface tweaking, relaxed transimission lines and reduced PCB costs. Adaptive interface timing can ensure no trace length matching and stable system operation, while Error compensation can detect errors and do real-time error detection.
Radeon HD 4770 vs. Radeon HD4850
When challenging with Radeon HD 4850, HD 4770 is 5.66% behind, with honor.
Radeon HD 4770 vs. GeForce 9800GT
Radeon HD 4770 outperforms 9800GT in most of the items with clear advantage, by 54.11% in ETQW 1920*1200/8xAA/16xF.
Radeon HD 4770 vs. GeForce 9800GT OC Edition
GeForce 9800GT Overclock Edition still appears weak when compared with Radeon HD 4770.
Radeon HD 4770 vs. GeForce GTS 250
Actually, this comparison is just listed for your reference. But Radeon HD 4770 still wins a narrow victory in a few items.
Conclusion
RV740: Ordinary appearance while extraordinary performance
The fastest graphics card under $100
Our review proves again that we can never make judgements just by appearance. Yep, HD 4770 looks a little ugly (some of you may don’t think so) as we said here, but it’s highly cost-effective.
AMD claims Radeon HD 4770 is the fastest graphics card under $100, which is quite true. It delivers 95% gaming performance of HD 4850 which is $40 more expensive.
The excellent performance of Radeon HD 4770 in power consumption, overclocking, temperature and noise control owe to the 40nm manufacturing process and what the new processing brings. Actually, Radeon HD 4770 is the biggest surprise that AMD brings us since HD 4800 Series.
AMD still has lots of RV770 stock, and as we said, 800SP Radeon HD 4830 was created to pave the way for HD 4770. But now, the strong HD 4770 puts HD 4850/4830 in an awkard position. The MSRP of Radeon HD 4770 is set as $99, so we think AMD will reduce the price of HD 4830 soon.
Radeon HD 4770 has undoubtedly given NVIDIA tough time, as NVIDIA doesn’t have a good plan to fight back so far. ATI was faster to transition to 65nm, 55nm and now 40nm, and NVIDIA will have to wait for 32nm to try to beat ATI.
However, AMD can’t relax either, because RV740’s excellent performance mainly owes to 40nm technology. The advanatge may disappear when NVIDIA also pushes out 40nm products. There is another problem: Qimonda has announced bankruptcy, so we don’t know if the productivity of GDDR5 can be ensured in a long term.
Anyway, AMD’s sucessful transition to 40nm will enable the whole product line to move to new processing soon.
Crossfire Benchmark:
Benchmark Platform And Settings
Considering HD 4770 only features 512MB memory size, it’s unfair to compare it with HD 4890 under the items such as GTA4 and FarCry2, so we only kept the benchmarks under five items including 3DMark Vantage, and set the imgae quality at the highest level.
Radeon HD 4770 VS. HD 4770 CrossFire
Amazingly, HD 4770 CrossFire has outperformed HD 4770 single by by over 100% under Call of Duty: Modern Warfare, and the advantage is more than 80% in average.
Radeon HD4770 Crossfire VS. HD4890
HD 4770 CrossFire has outperformed HD 4890 by nearly 20%, with fewest difference under Left 4 Dead.
Power Consumption Benchmark & Conclusion
According to our test, HD 4770 CrossFire brings more than 80% 3D performance improvements, superior to HD 4890 nearly 20%. When loaded by 3D tasks, the CrossFire platform sees about 80-100W power consumption increase compared with single HD 4890, but it still consumes less power than HD 4890 does.
On the other hand, PowerPlay function disables the automatic decrease of GDDR5 memory clock, so HD 4770 CrossFire brings a bit higher power consumption than HD 4890 when idle or playing videos. We believe that the upcoming non-reference cards will see an improvement in this.
According to Newegg, the MSRP of HD 4770 has improved from the claimed $99 to $109.99, while HD 4890 is currently sold at $249.99, which means HD 4770 CrossFire still costs less than HD 4890. Besides, thanks to the advanced 40nm technology, HD 4770 comes with less power consumption and more room of core/memory clock improved, so HD 4770 X2 will hopefully become a deadly weapon to confront with NVIDIA’s GeForce GTX 285.
Of course it’s not easy to make a HD 4770 X2, as our experience tells that X2 graphics cards are usually less effective than CrossFire platform with two cards. Besides, the HD 4770 X2 might have a bottleneck when playing anti-aliasing games with inadequate memory size. 1G memory is quite essential for the multi-card platform. As you know, CrossFire doesnt double the memory. The last but not least, it’s still uncertain whether the yield rate of 40nm RV740 is enough or not.
How do you think about AMD making an HD 4770 X2? Please feel free to leave your thoughts.
Posts
No comments:
Post a Comment