10 minute read
GIGABYTE GTX460/GTS450
GTX 460OC-1GI
Advertisement
Recommended Award
GIGABYTE
GTX 460OC-1GI / N450OC-1GI
RRP: GIGABYTE GTX460OC-1GI: $229.99 | GIGABYTE GTS450OC-1GI: $129.99 Website: http://www.gigabyte.com/products/product-page.aspx?pid=3530#ov http://www.gigabyte.com/products/product-page.aspx?pid=3593#ov
Test Machine
• Intel Core i7 980X • Super Talent 2200 C8 • OCZ AGILITY 2 50GB SSD • Thermaltake Toughpower 1.5KW PSU • Windows 7 64-bit
After having run all the tests, and several others whose results are not presented here. We came away impressed with the GTX460 as it delivered better performance than we had expected. It seems NVIDIA did some amazing work with reworking the GF104 core. It’s not quite enough to outdo the Radeon HD5850 but it is certainly a better product than the HD5830 which retails for a similar price. When overclocked it sometimes delivers numbers better than those of the GTX470 and the Radeon HD5850 in selected benchmarks. At $229 this is incredible value. The Heaven Benchmark results were particularly impressive because the GTX460OC delivered a marginally better score than we recorded with a Radeon HD5870 (reference model).
In 3D Mark Vantage the score is equally impressive but certainly worth taking note of when the GPU and Memory have been overclocked. With a score above 18,000 it delivers numbers expected of graphics cards costing anything from $50 to $100 more. We do feel that GIGABYTE could have clocked this card a little higher, given the impressive cooler that the card comes with. Fortunately this is easily fixed with any of the available overclocking utilities and 850MHz/2000MHz setting is easily achievable with a simple slider.
As impressive as the GTX460 was, we couldn’t help but be a little underwhelmed at the performance of the GTS450. Granted it’s a $130 graphics card and meant squarely at 1680x1050 and lower, we can’t help but feel there’s some wasted potential in this graphics card, courtesy of the 128-bit bus. Against the Radeon 5770, the GTS450 struggles unless overclocked to at least 900MHz, where it makes up some ground. Given that this is the cards direct competitor, it’s puzzling why the numbers are closer to those of the Radeon 5670 instead. With that said, if the ability to use PhysX and 3D Vision is appealing to you, the GTS450 is the only DirectX11
N450OC-1GI
card that can manage all of this at this price point. For a playable frame rate in Just Cause 2 we had to disable the Bokeh-Filter, GPU Water Simulations and reduce the resolution to 1680x1050 at which point we were able to break the30fps barrier in this DirectX10 title. So if you keep the detail settings at medium to high and play at maximum resolution of 1680x1050, there’s some fun to be had with the GTS450.
Overclocking this card results in limited gains, but they are easy to take advantage off because the card overclocks so well. 950MHz was our target clock and we achieved that by merely moving a slider to the right. The memory while rated at 2GHz does overclock to 2,200MHz but it doesn’t mean much with such a narrow memory bus. With some voltage tuning we did manage 1GHz on the core, which further boosted frame rates, but not enough to make it worth the effort. The GTS450 isn’t a poor showing of the GF106 core, but rather an un-optimal one.
Notes on scores by others
Given that these are mainstream graphics cards, extreme overclockers are not flocking to these to break any world records, however there are a few scores on HWBOT where the a GTX460 has clocked as high as 1230MHZ on the core (Achieved by Team China’s Icw33). As the 450 is even newer and capable of even less, at the time of writing there were not any worthwhile scores we were aware of.
[ Neo ]
Benchmarks
We ran a small array of benchmarks that demonstrated performance of all aspects of the system and allow easy comparisons in future reviews. We performed these on the industry standard 3600 MHz on an un-optimized install of Windows 7.
Benchmark
Heaven Benchmark 2.1 (1920x1080, Normal, 4xAF)
3DMark06
3DMark Vantage
Just Cause 2 (1920x1080, Max Detail, 16xAF) – Dark Tower
Gigabyte GTX460OC-1GI
896
22888 marks SM2: 8928 SM3: 9967 CPU: 7790
P16322 (P18067) GPU: 13922 CPU: 33804
Average FPS: 34.24 (40.1)
Gigabyte GTS450OC-1GI
564
17823 marks SM2: 7047 SM3: 7019 CPU: 7726
P11036 (P12359) GPU: 9014 CPU: 33711
Average FPS: 21.28 (23.94)
Summary
From the results, we have to conclude that, the GF104 is a better balanced graphics core than the GF106. While it’s more than likely that there’ll be more GTS450 cards sold than GTX460 cards, the truth remains that the GF104 core is an unmistakable hit, while the GF106 seems to have been crippled in a way to make it less impressive than it could have been. We would have preferred a 256-bit memory bus, even if it made use of GDD3 as it would give a healthy boost to the frame rates and possibly give it the edge over the Radeon HD5770
As far as GIGABYTE’s offerings of these products, they may not vary from the reference design, but the WindForce coolers do make all the difference in delivering whisper quiet operation and amazingly low temperatures.
Which would you buy?
Where the GTX460 is concerned, it’s an easy card to recommend. The performance is good and there is plenty of overclocking headroom. 850MHz on the core for 24/7 operation gives some good performance and because of the WindForce cooler, it’s whisper quiet. The GTS450 on the other hand is a little harder to recommend, because it’s by and large slower than the Radeon 5770 which is almost a year old and also comes in at the same price. However with some overclocking is does manage to gain some ground and does become a useful little card.
The Score
The pricing for NVIDIA’s latest mainstream GPUs is really aggressive and the GPUs themselves offer great performance (at least where the GF104 is concerned). The There are plenty of faster GTX460 cards out there, but nothing an overclocking utility can’t fix. The GTS450 on the other side is a tougher sell but overclocks just as well if not better.
GTX460OC-1GI: 8/10
N450OC-1GI: 7/10
Benchmarks
We performed these tests on the industry standard 3600 MHz on an unoptimized fresh install of Windows 7.
SuperPI 8m: 3m 51.255s 3DMark06: 20803 Marks SM2: 7607 SM3: 9951 CPU: 6423
3DMark Vantage: P17423 (GPU Score: 17056, CPU Score: 18625) GT1: 53.39 GT2: 46.41 CT1: 2420.63 CT2: 28.62
Everest read: 10314/MBs Everest write: 9353/MBs Everest copy: 13819/MBs Everest latency: 45.4/ns
AMD Phenom II X6 1090T
Recommended Award
Test Machine
• ASUS M4A89TD Pro • 5870 stock using Catalyst 10.2 driver • Corsair GTX2 2250 and
G.Skill Flare 2000 • WD VelociRaptor 600GB • Antec 1200 OC • Windows 7 32-Bit
Intel has had the performance market cornered for a while, and unless AMD can do something dramatic, it looks to stay that way for a while still. Not oblivious to this fact, AMD has been doing what they can to compete, targeting performance at specific price points over outright performance. The AMD Thuban core fits into this criteria ad understandably so because not many end-users can spend $1000 on a CPU just to benchmark, let alone for gaming. The 1090T comes in at around 30% of that price and this is great news for everyone, in particular enthusiasts. We have said it before and will say it again, AMD is a genuinely fun platform to overclock and the fact that the products are so affordable provides an excellent excuse for any overclocker to get involved with AMD.
AnAlySIS
The first thing we can determine from comparing 2D results directly against Phenom II X4 (Deneb) is that there is no tangible change in performance. Thuban and Deneb results are identical (clock for clock) in single threaded benchmarks but that is not the case when we take look at some multithreaded benchmarks such as 3DMark 06 and Vantage. 3DMark06 scores 20803 on the Phenom II X6 1090T up from 20003 marks on the Deneb core. We saw a huge boost in the CPU score by moving from four to six cores. Vantage was similar in scaling, the GPU scores remained near identical but the CPU score increased by 6000 points increasing the overall score by over 2000 marks, a massive increase. It is evident the IMC (Integrated Memory Controller) on the Thuban core is much improved over the one in on the Deneb core. The improved IMC affected the NB (North Bridge) frequency as in addition to the memory frequency. Looking at the NB frequency on 1090T, we could quite easily surpass 3000 MHz using air-cooling, comfortably running LinX at 3200 MHz, in contrast to the Deneb core, where if you could run above 2900 MHz your NB would have a very strong North Bridge. This might not seem like a big shift, but in 32M, a performance gain of 300MHz can mean a reduction of 10 seconds. Moving onto memory clocking, this is where the Thuban core began to shine. Chew publically took the record for highest memory frequency on Deneb last year
with a frequency of 1946 MHz, which at the time was quite amazing on an AMD platform.
Now with Thuban, we are already looking at frequencies in excess of 2400 MHz and we are sure these results will get better as time goes on and better CPUs surface. Our experience with the CPU did not stop there, as it also seemed much easier to achieve higher LinX clocks on air-cooling, even though there are an additional 2 heat producing cores. We were able to run 20 loops of LinX with our Thuban at 4200 MHz using the reference cooler, which is approximately 200 MHz over what a good Deneb based CPU can achieve. When cooling the CPU to subzero temperatures benchmark scores over 6400 MHz seem to be possible and are a regular occurrence at HWBOT. It does seem though, that Deneb CPUS still have the edge in the 2D benchmarks such as SuperPI 32M, where there is about 200MHz over the top Thuban frequencies. This might just be a matter of time until a great CPU is found, or it may be due to the extra heat put out by the 2 addtional cores and the inability of LN2 to achieve the temperatures required for Thuban to beat Deneb.
Notes oN scores by others
Extreme overclocker Chew seems to have found some amazing CPUs; and has put up a SuperPI 32M score at 6550MHz and managed to run 3DMark Vantage at nearly 6100 MHz. We used the 1090T in our labs to push our G.Skill Flare ram kit to nearly 2400 MHz and only a few days later Mad222 pushed past the 2400 MHz barrier with a set of “Geil ONE” modules.
Summary
This CPU is leaps and bounds ahead of Phenom X4 in memory and north bridge clocking. This translates to better performance in 2D and 3D benchmarks and a step closer to Intel in the memory world. There are also big benefits for enthusiasts that wish to stay on air or watercooling, with big gains to be had in maximum stable frequencies. Overall a very good step forward by AMD. Fingers crossed that next time we see a significant performance jump in performance so the competition does not leave them behind completely.
Would you buy it?
For a workstation or gaming machine, we would be considering this platform and this CPU. Unless you have money to burn, the 1090T really is at a great product providing all the power you need for any power workstation or gaming machine. It is worth keeping in mind that to take proper advantage of this CPU you will need to get a board based on the 800 series chipset.
The Score
This is about the best score any current generation AMD part can score, and near impossible to give a higher score than this considering that Intel is a long way in front. However, the competitive price and the clock speed gains over the previous generation CPUs are more than enough to get a strong recommendation from us.