AMD Radeon HD 4770 - Review

AMD Radeon HD 4770 - Review

AMD returns to the graphics card segment entry with the launch of the Radeon HD 4770. Behind a name for the least harmless, hides a surprising new DirectX 10.1 graphics chip burned for the first time in 40 nm!

AMD continues to innovate in effect on the manufacturing process since the graph of the subsidiary firm of Sunnyvale adopted before NVIDIA, the process of manufacturing 40 nm. The big question is of course, whether the user will benefit in any way for this new manufacturing process. And looking more closely at the specifications announced Radeon HD 4770, it is tempting to answer yes!

The new graphics chip from AMD is missing in fact it claims to offer a total of 640 stream processors while it has GDDR5 video memory, a first in this price segment! the Radeon HD 4770 is it the shadow of the GeForce 9600 GT and others? The answer in this test!

AMD Radeon HD 4770: the characteristics

Technically, the architecture used in the Radeon HD 4770 chip, code-named RV740 is not new. We find in fact that of Radeon HD 4000 launched last summer with native support for DirectX 10.1 and an arrangement is well known groups of 128 units each consisting of 5 processors flow. Thus, the Radeon HD 4770 a total of 640 stream processors. This figure must be compared with the number of units of the Radeon HD 4670 aka RV730, the Radeon HD 4770 is supposed to replace. Were found in the Radeon HD 4670 320 stream processors: the Radeon HD 4770 is therefore twice as many processing units. Moreover, the chip is at the same level as the Radeon HD 4830, a slightly simplified version of the Radeon HD 4850 and has 640 stream processors. For their part, the Radeon HD 4850, 4870 and 4890 include 800 stream processors.

In addition to its number of units, the RV740 chip behind the Radeon HD 4770 is of course its fine print. While the Radeon HD 4850 (and others) is recorded in 55 nm, the Radeon HD 4770 is first recorded in 40 nm TSMC. Less expensive to produce, the Radeon HD 4770 is supposed to consume less so that it can behave in the same area a larger number of units while heats less. Physically, the heart of the RV740 chip is incredibly small!


AMD also offers news from the side of memory, Radeon HD 4770 is the first graphics processor to the price level using the GDDR5. However, no miracle on the side of the interface, because unlike the Radeon HD 4830/4850 and 4890, the memory used in the Radeon HD 4770 is interfaced to 128 bits. The available bandwidth is therefore reduced, although the high frequency of GDDR5 is supposed to compensate for everything.

AMD Radeon HD 4770: features

We must admit, the similarities between the RV740 and RV770 are very numerous. So much so that one would be tempted to write that the RV740 is ultimately a die-shrink of the RV770 ... That is to say the same chip, engraved with a different manufacturing process. However, AMD has, as mentioned above, adjusted characteristics of the chip to match the target market. Memory bus reduced number of calculation units decreased while the side of operating frequencies, the RV740 chip is cadence at 750 MHz against 800 MHz for memory. This brings us to the following observation: even though the fineness of engraving could have higher frequencies of aggressive operation, the Radeon HD 4770 operates at the same frequency as the Radeon HD 4670. Of course, the number of transistors largest chip may not be alien to the stagnation of the frequency chips are actually 826 million transistors, almost as much as the RV770 animating its big sisters.

Moreover, with regard to memory, the operating frequency are added in any moderate narrow memory bus are the theoretical bandwidth is 51 GB / s. This is quite low compared to the Radeon HD 4850 which offers a bandwidth of 64 GB / s while the recent Radeon HD 4890 offers a memory bandwidth more impressive 124.8 GB / s! But it is almost two times more than what his time in the Radeon HD 4670.

Recall in passing that the RV740, like other chips on the Radeon 4000 family includes the ATI UVD engine that supports in hardware decoding of high definition video stream from including Blu-ray. Other AMD specificity, the presence of a 7.1 audio controller built into the graphics chip. This allows, when playing Blu-ray, broadcast on the HDMI audio from the film without going through additional connections and risky at the motherboard.

AMD Radeon HD 4770: map

Quite different from the Radeon HD 4670 it replaces, the Radeon HD 4770 is characterized by its dual slot design. It's quite unusual to emphasize since the graphics cards to less than Rs 9000 , are generally single slot. Another feature of the Radeon HD 4770, the presence of an electrical connector to the six pins PCI-Express. For the rest, we have a map rather classical interfaced in PCI-Express 16x, relatively long, with a red PCB of 21 centimeters.

The cooling system consists of a metal radiator fins based on the graphics chip and built into a shell of translucent red plastic at the end of which blows a fan squirrel cage type. Warm air is expelled out of the system. With 512 MB of Qimonda GDDR5 memory, the card has a specific radiator covers its memory chips. Overall, the noise generated by the card are limited unless it takes a sudden blow warm. In these cases, noise is just unbearable and recall the glorious days of the GeForce FX 5800. The side of operating frequencies, the card we tested is a reference design developed by AMD, it was no surprise that they are in accordance with the manufacturer to 750 MHz for the graphics chip and 800 MHz for memory.

Note the presence of a double tongue CrossFire to use the Radeon HD 4770 in CrossFire configuration, while the outputs offered by the card are more mundane. There are indeed two DVI connectors and an analog video connector to mini-DIN. Remember that manufacturers of graphics cards with the Radeon a DVI / HDMI to convert the DVI connectors HDMI connection.

AMD Radeon HD 4770 : Performance

To test the performance of the Radeon HD 4770, we used a recent platform, details of which are shown below:

• Motherboard Asus Rampage II Extreme (BIOS 1204),
• Intel Core i7 920 2.66 GHz,
• 6 GB DDR3-1066 - OCZ,
• RAID 0 - Hard Drive 2x Western Digital Raptor 300 GB

Our system operates on Windows Vista Ultimate Edition with Service Pack 1. It was the 64-bit operating system. On this machine we will test logically the Radeon HD 4770 but it replaces the card, the Radeon HD 4670, not to mention the Radeon HD 4830 and Radeon HD 4850 512 MB NVIDIA side, we adopt the GeForce 9600 GT and that the GeForce 9800 GT, that being proposed at the same price as the new card while AMD offers features similar to GeForce 8800 GT

Drivers side, we have used drivers NVIDIA GeForce 185.68 beta and CATALYST 9.4 beta. Note that the test protocol was revised to go to the essentials while we retain resolutions lower than those usually used to cope sometimes reduced performance card entry.

3DMark Vantage
We begin our overview of the performance by the unmistakable 3DMark Vantage. Here the Radeon HD 4850 leads the way, followed in second position of the Radeon HD 4770. The new AMD chip directly wins the second step of the podium with 16% higher performance in face Radeon HD 4830 at least 1920x1200. Compared to the Radeon HD 4670, its predecessor, the Radeon HD 4770 is, at least, twice as fast, literally. Finally the comparison with the supply of NVIDIA does not benefit from the latter, the GeForce is significantly behind.


Call Of Duty 4 - v1.7
Call Of Duty 4 Modern Warfare, the duo Ranking consists of the Radeon HD 4850 and the second position of the Radeon HD 4770. In 1680x1050, the latest AMD is 29% more efficient than the GeForce 9600 GT and 4% faster than GeForce 9800 GT. Faced with the Radeon HD 4670, the performance gain was 58%! It should be noted in passing performance 9% above the Radeon HD 4770 Radeon HD face in 4830.



Crysis Warhead - v1.1 - 64 bit
For Crysis, maps AMD undoubtedly lead the dance. The first three steps of the podium is occupied by the Radeon and Radeon HD 4770 is once again in second place. Faced with the Radeon HD 4670, the performance of the Radeon HD 4770 are 57% higher. Note the absence of results of GeForce when the resolution exceeds 1280x1024.


Left 4 Dead
The recent release of Survival Pack Left 4 Dead has caused us many problems in less than 48 hours Valve has updated twice the graphics engine of the game forcing us to record our demo twice. Apart from this inconvenience, there are good benefits that are ahead of Radeon GeForce easily from 1680x1050. No change in the hierarchy, and the Radeon HD 4850 is 2% faster than the Radeon HD 4770 in 1680x1050. Faced with the GeForce 9800 GT, the Radeon HD 4770 wins with an advantage of 13%.


Far Cry 2 - v1.02
Surprise, in Far Cry 2, there is a change in the hierarchy, the Radeon HD 4770 matching the Radeon HD 4850. The two AMD cards occupy the top two places while facing the GeForce 9800 GT, its direct competitor, the Radeon HD 4770 is 16% faster in 1680x1050. Comparing face to Radeon HD 4670 allows to measure the magnitude of the difference between the two maps: 51% at the bottom word 1680x1050.


World In Conflict - v1.010
To finish this page with the strategy game World In Conflict. The Radeon HD 4850 takes a few colors, while the Radeon HD 4770 is at par with the Radeon HD 4830. In all cases, the offer exceeds the AMD and NVIDIA solutions Radeon HD 4770 is in 1280x1024, 9% faster than GeForce 9800 GT and 26% more efficient than the GeForce 9600 GT.



Call Of Duty 4 - v1.7 - FSAA 4x - AF 16x
Enabling filtering functions change somewhat the picture of our winners. The Radeon 4850 seems DH suddenly lose its superb when it displays the same performance as the Radeon HD 4770 in 1920x1200. That was enough for the GeForce 9800 GT rose to third position. In the same resolution, the Radeon HD 4770 offers identical performance to the GeForce 9800 GT when it is 24% faster than GeForce 9600 GT.


Far Cry 2 - v1.02 - 4x FSAA
Table inverse in Far Cry 2 where activation of the filter function takes the Radeon HD 4850 in mind. The Radeon HD 4770 keeps its second place performance with 19% higher than the GeForce 9800 GT in 1680x1050. In this resolution, the Radeon HD 4770 delivers a number of frames per second higher than the 10 Radeon HD 4830.


Left 4 Dead - 4x FSAA - 16x AF
Back in Left 4 Dead by activating the game since the filtering. The Radeon HD 4000 series are the leader with a Radeon HD 4770 which truste second place and is 10% faster than the GeForce 9800 GT and 35% more efficient than the GeForce 9600 GT in 1680x1050.


World In Conflict - v1.010 - 4x FSAA - 16x AF
We close the second page of the first tests with an overview of performance in World In Conflict. The results speak for themselves, the Radeon HD 4770 is very well located, always second, well before the Georce 9800 and 9600 GT. And while the gap between these three cards remain moderate in low resolutions, it is growing significantly 1920x1200. In this resolution, the Radeon HD 4850 is 9% faster than Radeon HD 4770.

Consumption

Burning in the 40nm Radeon HD 4770 requires a glance at the power consumption of various graphics cards needed. We conduct two ways, using a watt meter, of total electricity consumption of our system. Initially at rest, then load with a session in 3DMark Vantage 2560x1600.

At rest, the Radeon HD 4770 consumes more than its predecessor, the Radeon HD 4670, while consuming as much as the Radeon HD 4830. Its consumption is still at rest, more or less identical to that measured with the NVIDIA GeForce 9600 GT and 9800. Over, the Radeon HD 4770 consumes only 10 watts more than the Radeon HD 4670 and GeForce 9600 GT as it offers performance twice.

Conclusion

At the end of this article, it is inevitable to conclude on the excellence of the Radeon HD 4770. Because at least a INR 9000 performances are really there with more power consumption which remains quite reasonable. AMD drives the nail of his supremacy by offering its first graphics chip etched 40 nm, when NVIDIA is just beginning to decline, largely, his bid in 55 nm. And to stay in the industrial aspect of things, the choice of GDDR5 graphics card for an entry-level is perhaps not the most appropriate, for matters of cost and production volume. Ultimately the team this GDDR5 Radeon HD 4770 is only offset the limited features of the memory interface such as the danger is to meet supply problems.

In terms of product strategy, the arrival of the Radeon HD 4770 leaves us perplexed. For INR 9000, the card offers superior performance to a Radeon HD 4830 when it had just teasing the Radeon HD 4850 sold approximately Rs 10000 , in its 512 MB and just 4% faster. Indeed, AMD seems to make the graphics card market down, something that the consumer is not necessarily a bad thing, especially given the price / performance ratio of the Radeon HD 4770, but for the financial health of AMD ... that's another story!

Finally, the big problem of the Radeon HD 4770 is its very limited availability at launch, several partners AMD we have indicated that they have little or no stock. Too bad, because the Radeon HD 4770 has serious advantages as it has no real rival in the competition. At the same rate, the GeForce 9800 GT is simply not the weight. Remains a detail ... sound card that can quickly be unbearable at full load: AMD hope quickly revised his copy on this point.

Review - New Mac Pro 2009

Review - New Mac Pro 2009

In depth, the architecture of the Mac Pro. Deliberate choice or willingness to stick closely to technology from Intel, its main partner? The question could remain partially open ... but hardly compulsive-t-on technical details of the new monsters that question very quickly leaves the room for evidence. More matter for Apple to commit blunders like to remain powerless against its own ads ... and not being able to keep its promises. He will be remembered for the famous bar ever 3.0 GHz PPC with IBM.

lagship of the fleet from Apple, the Mac Pro has not seen the design change, or very little, since the appearance of the Power Mac G5. As usual, the revolution comes from within.

Today, things have changed. Intel provides Apple and until recently had claimed the preview on its Intel Xeon processors , its range pro server to Apple. So it is logical that the Mac Pro 2009 edition adopted the new batch of Intel Xeon chip code-named Nehalem EP that support the Hyper-Threading and now, the virtualization processor cores, all supported by technology TurboBoost. Advertised as very swift, the new stations made, are they really?

To find out, we tested the two basic models present in the catalog configuration and very strong at 2.93 GHz version Octo-Cores And powerful monsters such calculations are certainly!

External signs of change

External changes are discrete, simply look at the connection to this. Crane, the FireWire 400 port has given way to a second FW 800 and two USB 2.0 connectors and the traditional headphone jack. At the rear the changes are similar in all respects. There are three USB 2.0 ports, two FireWire 800 (400 lost), a headphone output with an analog mini audio input jack, an input / output optical S / PDIF (Sony / Philips Digital Interface) digital connections and two Ethernet 100/1GoBaseT.

Except the deletion of the IEEE1394 connector, no significant change in the back of Mac Pro.

DVD side, no changes, both in design and in embedded hardware. Supplied with a double-layer SuperDrive, you can opt for a second unit at the Apple Store. The DVD burner Apple has still not given way to a device as a Blue Ray. Too bad ... Apple probably reserve it for the launch of Snow Leopard, but for the price we would have hoped at least that kind of player since the SuperDrive now enjoy a SATA interface.

The SuperDrive finally adopt the SATA. Further obstacle to install any other type of player stations in the Apple

You have understood, the Mac Pro still retain the same look, and it does not seem to program Cupertino change its tune, at least until the next generation And again, nothing is certain .

A revolution from inside
In adopting the new Xeon server chips Nehalem EP Series W3500 (for the four single-processor socket), E5500 and X5500 (for versions 8-socket dual-core), represented by the Mac Pro 2.26 GHz and 2.93 GHz Octo, Apple had to revise AZ internal architecture of its office equipment business.

Indeed, the new Intel Xeon processors, which have a completely new architecture with integrated controllers, Apple needed to revisit the overall design of the case. On the other hand, if this meant to of Research and Development in collaboration with Intel, which adds additional cost to the firm, it will be quickly amortized. Now, all models and all frequencies share a common single case since the architecture of CPU Nehalem EP implies that the principal organs and controllers are grouped together and placed near the processors.

Results to production what distinguishes a Quad-Cores of an Octo-Cores, this is only her daughter board or module as the processor and RAM also intimately related to CPU. We could ascertain, we have just enough to switch modules (CPU motherboards) to transform our Mac Pro 2.66 GHz Quad in Octo and vice versa, because of our station CTO 2.93 GHz eight cores a simple quad processors. Single signal after a restart, Mac OS X informs you that the RAM has been reset and fully operational. This has allowed us to configure our willingness to test three models to achieve all of the lab bench.

Once deposited the hood, remove the CPU module housing is a real breeze. Suffice it to unlock all pressing the two small levers on both sides.

Here, motherboard dual socket Mac Pro 2.93 GHz.

But these are not the only changes. Apple took advantage of this major revision to correct a few small design flaws cited in the previous generation. Bays that accommodate hard drives have been improved. The disk media slide better and fit more easily.

In terms of PCI Express 2.0, net improved support for many cards. Thus, the Radeon HD 4870, like GeForce 8800 GT for Mac Pro (Early 2008), enjoys a very ingenious system of locking the rails. So that more or less juggling to insert a NVIDIA 8800 GT in a machine, the insertion direction is well.

The inclusion of graphic cards or other large-format no longer poses any concern ...

This is not the only change that has been made as regards the insertion or extraction of a card on PCI Express. Formerly, small plastic ERGOS in charge of blocking cards that were inserted, were quite difficult to access. This is no longer the case today as they were replaced by a metal bar to act as a lock. More accessible, it is enough to push it to the left to unlock everything.

A push to the left enough to release all the maps that can be installed in Mac Pro ...

More ventilated, and the new chassis allow better air circulation. Nevertheless, the Radeon HD 4870 heat considerably and generate much noise, unlike the GeForce GT 120, very discreet on this point.

Mac Pro Quad 2.66 GHz and 2.93 based on chips from the Nehalem family W3500. Feature of these processors in 45 nm engraved, they can only work on motherboards single socket but consume relatively a lot. The 2.66 GHz which is a Mac Pro Xeon W3520 not least 130 Watts. It is lot, but it has the advantage of cheaper and ensure good performance. His hanging from 2.93 GHz, a Xeon W3540, also equipped with 8 MB L3 cache (or 2 MB per heart), do not consume over 130 Watts as his little brother.

Mac Pro Octo 2.26 and 2.93 GHz (from our lab) embark on together Xeon quad-core from the X5500 family which is itself divided into three categories of processors. There is a E (Enhanced improved) and the X and L series. Each of these categories requires motherboards dual socket but may work on a solo card dual socket unlike the W3500 series of mono quad-core processors. Another advantage is that these CPUs consume much less per unit.

On modules bi-socket processors, everything is doubled. Memory controller, RAM and consumption. Thus, the Xeon E5520 2.26 GHz consumes 80 Watts twice and has two banks of DDR3 ECC as its counterpart at 2.93 GHz X5570 Octo whose consumption blithely climbs to 190 Watts (95 Watts twice).

Besides simplifying the architecture for Apple, using the new Xeon can switch to a type of RAM much cheaper than previous models. DDR3 SDRAM ECC (ECC for error correction) costs about three times less than the memory used before and does more parity. Another major change induced by Xeon Nehalem, the integration of the controllers I / O memory is that allow direct access between the processor and RAM. The bandwidth is considerably increased and the phenomenon of latency reduced by approximately 35%. This also entails the disappearance of the famous Frontal Bus or FSB (Front Side Bus) which stood at 1333 MHz on the older models.Other important thing, Nehalem introduce technologies to the Hyper-Threading and TurboBoost.



Hard drives

Supplied as standard with discs Western Digital Caviar Black 640 GB , the same team that the latest iMac, including the 2.93 GHz, the Mac Pro can benefit from a 1Tera, Western Digital always, available on the AppleStore with 90 €. One option charged dearly, but certainly seduce ...

Especially since these discs, once configured RAID in N (N ear R edundant A rray of I ndependent D isk), better known as JBOD (J ust a B unch o f D isk), all data is concatenated and indexed sequentially, allows for excellent performance.

Certainly, without the redundancy of RAID 1 (or mirror mode), the overall capacity is divided by two. Nor the constraints of RAID 0, very fast, which lists the data - not sequenced - on all hard drives configured in RAID. But if a disk has to let go, the whole set of data is compromised. So, the solution N RAID is the most interesting in terms of performance, which is why we opted for this mode for our tests (see below).

However, there are many manufacturers (Hitachi, Seagate, Western Digital, Samsung, Fujitsu, Toshiba) hard drives on the conventional market. We will not talk here of DSS. It becomes quite dangerous to choose the right disk as needed: speed, security / reliability and storage capacity.

As you can see, if the WD Velociraptor 300 GB are considered swift, the Caviar Black 1TB delivered as an option on the store can be just as attractive and efficient while being less expensive.

Graphics

despite the power available, some games under Windows are able to challenge our favorite Mac Pro: The 2.66 GHz Quad. Crysis obviously like Lost Planet - Extreme Condition, but James Bond - Quantum of Solace (the demo), Crysis Warhead, Medal of Honor Airborne, Call Of Duty: World at War etc. The GeForce 120 GT is not really suited especially when all the rule parameters and the maximum resolution of 1920 x 1200 LED Cinema Display. Rest by taking the option Radeon HD 4870, with good charged all together, we get the station working to do everything. It will then reduce its cost and, above all live with noise generated by fans of the cooling system with both processors than ATI video card very poorly run under Windows XP, in the namely the Professional version that were used in tests.

Crysis Warhead really need to have a Radeon HD 4870, the GT 120 is not enough, unless you invest in a Quad 2.93 GHz.

While a 3.06 Ghz iMac with a GeForce 130 GT can play without saccade to Quantum of Solace, on the Mac Pro input range, the 120 GT fishing.

Here, the cooling system of the Mac Pro Quad 2.66 GHz. The radiator fins conceals imposing a fan who is responsible for circulating the air ...

At its opposite, the coolant tubes of the system. Here, the thermal paste in abundance on the top, carries part of heat on the wings.

On the Mac Pro 8-core, the heat radiators are twofold. Shorter as they dissipate heat in the same way as on Mono, except that the thermal paste is in abundance.But the Xeon E5520 and X5570 are consuming less and the dissipation is also lower. Integrated with the radiator fans are 30% smaller than the 2.66 GHz Quad.The transition to Nehalem chip has two big advantages. First, the integrated memory controller allows direct access between CPU and memory while reducing latency. On the other hand, DDR3 ECC employee no longer works parity, is much less expensive and easier to obtain.

Another advantage of the design of the Mac Pro 2009: in case of malfunction or breakdown, it will be quick and easy for Apple, including the AppleCenter authorized to repair or change some organs. The processor cards are removable and CPU coolant systems as easy to disassemble.

We have mentioned the ease of extraction of the card processor. Its locking system is equally important. All in solid aluminum.ike the Power Mac G5 Dual Core, the casing of the Mac Pro was particularly ventilated. The items are very easily accessible. The two big fans, here both sides are responsible for circulating the area horizontally.the majority of taxable items (plastic etc.) or clipsent is set by sliding. The screws are also much more accessible.Able to accommodate up to four cards GeForce GT120, this generation has only two PCI Express ports (2.0) including a 16x already monopolized for video and can accommodate a RAID card on the fourth location.

Simple and refined, the GeForce 120 GT is especially known to be very discreet. Its totally recessed fan is barely audible when the computer is turned on.

Design after all the classic card entry which, if not very powerful, is also charged 135 €. However, it delivers a power greater than that of the iMac 24 2.93 Ghz even mounted on a Quad 2.66 GHz.

Unlike the GeForce, HD 4870 occupies the equivalent of two ports. It is also equipped with a Dual DVI connector and a Mini DisplayPort now de rigueur on the new machines.

Unlike the GeForce GT 120 which draws its power from the port where it is connected. The ATI card requires that you connect the two power ports on the motherboard.


The use of aluminum and galvanized steel is the main asset of this beautiful box, if its lines evolve much, retains all its qualities almost. Robust structure, easy access and installation of cards additives or additional hard drives in less than five minutes silent ventilation, etc., Are of particular interest. But we regret, despite the serious improvements and the care taken in manufacturing that Cupertino has not fully reviewed the design of its product to reduce the weight and thus sizes.




Conclusion :

We must not forget that these computers are aimed primarily at businesses and independent professionals. The new Mac Pro computing power has unprecedented and thanks to the arrival of Intel Xeon chips Nehalem EP capable of simulating up to sixteen eight virtual cores with Hyper-Threading. However, despite the sharp increase in power, including with respect to the previous generation 8-core launched in January 2008, these new machines are expensive, even prohibitive, especially the very high end, even in terms of future potential.such an investment can not be truly sustainable, at least we believe, once Snow Leopard ( Mac OS 10.6) released in the shelves of Apple Resellers. One aspect that must take into account today. Especially as you have seen, whether on Mac OS X or in some applications, however well-developed by Apple or third-party publishers, we can not say that the whole of the sixteen hearts (eight hearts physical) is particularly well managed.

Review - Gigabyte GV-N96TSL-1GI Geforce 9600GT

Review - Gigabyte GV-N96TSL-1GI Geforce 9600GT

When a new graphics card out, especially high-end models, the cooling system is the same for all references of manufacturers. Each brand does indeed apposed its label to its colors on the cooling system of the reference map. The latter is sometimes as effective and silent on many Nvidia cards, sometimes noisy and less efficient as on some ATI cards. This allows manufacturers to come out then cards with alternative cooling systems as do Gainward, Asus, Gigabyte, etc.Some systems use known as Gigabyte that uses Zalman solutions, while others use systems developed just for them, often by them. In addition, the DIY (Do It Yourself) is quite active and responsive by offering alternative solutions for cooling.

When Gigabyte does not learn from its mistakes

History to help you in any decision to buy a card with a system or a cooling system that would replace your original system, we start testing cooling of graphics cards with a new protocol and a new test. The first to open the ball is the Gigabyte 9600 GT Silent Cell recently announced version 512 MB overclock.
There is almost a year, we were testing a graphics card Gigabyte GeForce 9600 GT which happened to be equipped with a passive cooling system and at the same time overclocked. At the time, we found that the cooling system had its limits and that the error was Gigabyte to offer a 9600 GT fanless which generates more heat that a GPU clocked at frequencies standards. If it is often said that one learns from its mistakes, Gigabyte does not seem to know that saying since recidivism Taiwanese firm with two other fanless GeForce 9600 GT, the GV-N96TSL-1GI and GV-N96TSL-512i. The first is equipped with 1 GB of GDDR3 memory and 512 MB of seconds if the 1 GB version works with basic frequency of 9600 GT to 650 MHz for the GPU, 900 MHz memory and 1625 MHz for the shader core, carrying the 512 MB of memory has its core clocked at 700 MHz, the shader core and 1800 MHz while the frequency of the memory remains at 900 MHz. If the value of having 1 GB of memory on a midrange card, expired as the 9600 GT is more than questionable, what about overclocking, especially with a fanless design. This is a curiosity for a map if "old", which is exceeded by many maps today. Side frequencies, Gigabyte has been wiser than the model GV-NX96T512HP since the GPU from 650 to 700 MHz and 720 MHz. A public version overclocked with 512 MB of memory would have been more appropriate given that the audience for such a card is even more fans Media Center a year ago. The players themselves prefer a more powerful card.
Damage that Gigabyte did not adopt the "green" of the 9600 GT to inaugurate its fanless cooling system Silent Cell. As a reminder, the green is happening on the power connector PCI-Express, PCI-Express slot and the 75 watts it can deliver is sufficient for the operation of the card. The advantage of this "green edition" is obviously heat revised downwards, as all frequencies. That would have been even more useful than the 9600 GT does not benefit from lower frequencies in 2D. There are however ways to define via Riva Tuner and normally through the tool Gigabyte HUD Lite. Unfortunately, we tried to do so with HUD map test, without success. Less than one software update compatible with the new Gigabyte card, it will simply frequencies 3D idle unless playing with RivaTuner.

New with the Silent Cell

The main novelty of this card is obviously the passive cooling system called Silent Cell and strangely represented by an octopus futuristic. It covers the entire map but comes into contact with the GPU via a large copper base topped by a piece of aluminum. From this basic starting three copper heat pipes running through the fins in three different points: two at the front of the card and one rear. The fins are long and the number of 26. Eleven of them are a little longer than outside the enclosure via the air vent of the PCI bracket.

If one could think that it is to vent the heat outside the box, Gigabyte has a completely opposite reasoning. For the manufacturer, this is the place where fresh air will return. The hot air from its side will be evacuated by the curved fins at the end of the card before getting up the case and be removed by the fan housing. In our case we find it hard to be convinced that the heat will follow this path, especially when a fan front will bring fresh air from the front of the case. The fins espouse a particular form to maximize the surface of dissipation, but does not affect components of the card such as capacitors. Note that this heatsink is attached to the card via four screws housed in standard mounting holes around the GPU. This system is very well done and has a production quality copy. For those who have not noticed, the system transforms the 9600 GT card double slot while the reference model is a single card slot.
Configuration and testing protocol

We have chosen it because it is a housing means for cooling tower basic and classic with a 140 mm in the front intake and a 120 mm in the back extraction. It is devoid of any other fan or grid evacuation of air, making it an ideal candidate to better control the flow of air. The installed configuration is as follows:

Casing Lian Li PC-7F
Motherboard Asus Striker II NSE (NVIDIA nForce 780i SLI)
Intel Core 2 Duo E8500 (3 GHz)
2x1024 MB Corsair DDR3-1333
Heatsink Noctua NH-U12P cooled 2x120 mm Noctua NF-P12
Food Seasonic M12 650 watt
HDD Samsung 250 GB


We use the Scythe to control two temperature probes and four fans, two of Noctua heatsink fans and the two boxes. This brings us to detail our testing protocol. The first step is to run both fans at speeds boxes constituting a reasonable compromise between noise and cooling performance. Thus, the 140 mm front turns 800 rpm while the 120 mm rear running at 1000 rpm. The two 120 mm fans Noctua heatsink of the processor are set both to 800 rpm, which is more than enough given the quality of the radiator and the Austrian double airflow that we apply.

This is done, we face many temperatures, leaving the graphics card to rest without activity in the operating system, Windows Vista SP1 here. The temperature readings are:

* Ambient temperature 15 centimeters to the cabinet to 12 centimeters of soil to be at the height of the front fan.
* Temperature extraction of the graphics card to the back of the case
* Temperature in the food, under the fan on the graphics card
* Air temperature fan cooling system of the graphics card
* Temperature above the map sensor located in the space between the CPU heatsink and the PCB from the 3D map.
* Temperature of the GPU via RivaTuner
* Temperatures and other information available on the model of the card, via RivaTuner


After the idle steps, we identified the same session after a 45-minute test of OCCT GPU, especially stressful graphics cards. The second step is to consider the worst case by cutting the full breakdown of the case and leaving only the two CPU fans run at 800 rpm. This helps push the cooling system of the card into a corner and see the speed at which the fan turns, if any. During this second phase, we note the same temperature in idle mode and 3D, thanks to OCCT.

Finally, we measure the noise of the fan on the graphics card at the speeds we measured during different tests. These measurements are made with all the fans off, only the discrete fan Seasonic power remaining in operation.

Cooling performance

We compared the temperatures obtained with the Gigabyte 9600 GT Silent Cell temperatures measured with a reference 9600 GT. Temperatures are expressed in "delta" that results from the subtraction of the ambient temperature of the GPU. The goal is to have comparable results not influenced by ambient temperature of the room.
The graph is quite explicit, it shows that the Gigabyte solution is necessarily a less efficient solution ventilated, including a breakdown of housing. If the difference is very marked during a test as unrealistic OCCT stress that 100% permanently map, during a scene of realistic play in 3DMark06 running in a loop, the difference fades from a differential 25 to 10 degrees when the breakdown of housing is up. If you decrease the speed of the fan housing, the Gigabyte is a gap distance with a negative 24 degrees in 3DMark06 and 32 degrees with the OCCT test. Without ventilation cabinet, Silent Cell reached its limits and the GPU exceeds 100 ° C as you detail the tables below detailing our measurements. Note that the GPU has reached values of 104 to 106 ° C, temperatures causes of lags in OCCT, evidence that the GPU has reached its limits.

It should be noted that the breakdown of housing a maximum was not enough to keep the GPU temperature under 100 ° C for the OCCT test. To achieve this, we were forced to put both Noctua cooling system processor to their full potential, or 1300 rpm.

Analysis of noise

Remember, these measures concern only the fan on the graphics card, the other fans being cut to isolate the noise of the card only. Only the fan stays on the diet but it is more discreet than in our block Seasonic. RPM observed during the tests enclosed are reproduced via the Riva Tuner software that allows you to manually adjust the fan speed. The meter is placed in front of the card, open box, 20 centimeters in front of her and her height. Obviously, the 9600 GT Silent Cell has constant values of noise in the absence of a fan.

In our graphs, we postpone noise expressed in dB / A, but also the corresponding temperature of the GPU (the Delta to be exact). This allows you to see which solution offers the best cooling / noise.
In idle, the 9600 GT from Gigabyte fanless did not dramatically out of the 9600 GT with the standard cooling system that once the drivers responsible is silent. It is a little more audible, but in return offers a better cooling.
When testing a simulated game in 3DMark06, we see that the standard 9600 GT offers better cooling but at the cost of a surge of noise, especially in the absence of ventilation of housing. With conventional ventilation, the noise level is acceptable but it is far from the silence offered by the 9600 GT fanless which in return has its GPU much heat while remaining at 92 ° C maximum, which is acceptable.
Sub OCCT test very demanding unrealistic limit, the 9600 GT collapses in terms of heat but obviously retains a noise level low. the reference 9600 GT for its part, is just the opposite with a good cooling but noise high or very high without ventilation housing.

Conclusion

Leaving again overclocked fanless card, Gigabyte committed on the road maps for benefits contradictory. Silent heater too but when it is overstretched, the 9600 GT 512 MB Silent Cell is considered to place a card in a PC or a HTPC in which either we do not play games too demanding, or with d ' good ventilation of housing. For as often, a fanless system can not possibly go without a good flow of internal breakdown. If you want a card that heats less, use the 1024 MB of memory or lower frequencies with Riva Tuner, which will settle a little but not entirely, since you can not lower the voltage a GPU, more high on the overcloked. In summary, this is an interesting card, well finished, offering interesting outputs (HDMI / VGA / DVI) but not to stress with games greedy and in a poorly ventilated under threat of overheating. You also accept a higher price than 9600 GT standard fare you would access the superior in terms of performance.

Source: Techarena

Review - NVIDIA GeForce GTS 250 Graphic Card

Review - NVIDIA GeForce GTS 250 Graphic Card


The beginning of March was marked, like each year or almost since good a long time, by the behaviour of the annual edition of CeBIT. In the past living room impossible to circumvent for new technologies, CeBIT in speed, is wedged in the calendar between an international show, and a major European event, the IFA. This phenomenon is not exactly new and the world economic crisis did nothing but accentuate it this year with a particularly morose edition. For proof, the only advertisement of importance, in the graphic map productions, was the exit of GeForce GTS 250, the card which occupies us today.

So formerly CeBIT was used as platform of launching to the manufacturers for new products, it is not more the case today and GeForce GTS 250 is, to be clear as much as of the beginning, only one famous version of GeForce 9800, itself an updated version of GeForce 8800 GTS. It was necessary for us thus to give a progress report on this at the very least unexpected graphics card whose name change leaves think that it is about an innovation in the family of the GeForce chips 200, what is alas not the case. With it all the same an interest?
NVIDIA GeForce GTS 250: new name of G92

We wrote it and repeated, the too ambitious architecture of the chips GeForce 200 does not allow NVIDIA to decline them in versions more economic to produce, as it was for example the case with preceding architectures. It is indeed of use, in the manufacturers of graphics card, to conceive an architecture and to decline it on a whole range of product by adapting various parameters in order to obtain the best compromise performance/price. Only here, with 1.4 billion transistors, the chips GeForce WP 200 are too difficult mammoths with to decrease (note: any analogy with famous remarks would be fortuitous).

Whereas NVIDIA thought its gaining strategy, the renewed and unexpected aggressiveness of AMD with the range of Radeon HD 4000 took the firm with the chameleon of short. The departments research and development of NVIDIA company on the future graphics chip DirectX 11 of the firm, but, while waiting, it is necessary to occupy the ground not to be too much let outdistance in the sales by AMD. For that, NVIDIA appealed, once again, in G92! It is indeed a graphics chip built on famous G92, itself an optimized version of G80 equipping first GeForce 8800, which finds with GeForce GTS 250.

NVIDIA GeForce GTS 250: characteristics and 55 Nm!

You will have included/understood it, no new technical thing is not to be awaited of GeForce GTS 250, if it is not that it is a question of a new supposed name of making it possible NVIDIA to align its quotation with the remainder of its current range. To return from there to the chip, this one benefits from architecture DirectX 10 of GeForce 8 with a total of 128 processors of flow. In order to maximize profitability, one will announce all the same the change of manufactoring process. Whereas the G92 chips animating GeForce 8800 WP, 8800 GTS 512 Mo, 9800 GT/GTX+ are engraved in 65 Nm, always by TSMC, G92 which one finds within GeForce GTS 250 is engraved in 55 Nm. It should thus less heat and less consume that its elder, the surface of the die being indeed reduced.

Graphics chip NVIDIA G92 in 55 Nm: B1 revision

Side of the frequencies of operation, the chip GeForce GTS 250 benefits from frequencies raised vis-a-vis the first models based on G92. One thus finds oneself with a chip operating with 738 MHz for processors of flow turning to 1836 MHz. Side report, we are always in the presence of a bus 256 bits with a frequency of 1100 MHz for the chips GDDR3 which equip the cards of GeForce GTS 250. Small innovation for this model, increase in the memory quantity with 1024 Mo of memory for the model of reference. Partners NVIDIA remain free to propose models of GeForce GTS 250 with 512 Mb or 1 Gb of video memory.

GeForce GTS 250 made by Gigabyte: GV-N250ZL-1GI

Gigabyte is one of the first manufacturers to have been able to forward to us a graphics card based on the processor GeForce GTS 250. Equipped with its own PCB, blue color please, the card moves away very largely from the design of reference suggested by NVIDIA. Exit indeed the radiator double-stage which proposes NVIDIA and place with a cooling system personalized and signed ZALMANN. If the graphics card is adjusted in a simple square, it will occupy all the same two sites in your computer, device ZALMANN being rather cumbersome. Resting on the graphics chip, and only on this one, the system is composed of a broad aluminium tube radiator in the center of which breath a ventilator.

GeForce GTS 250 made in Gigabyte: GV-N250ZL-1GI

Interfaced in the NCV-Express train 2.0, the card embarks 1 GB of memory GDDR3 distributed on components of Hynix mark. Its power supply is ensured by the bus NCV-Express train, but also by a connector six pins. Equipped with a connector SLI, the card in multi-GPU mode with another card GeForce GTS 250 but also with GeForce 9800 GTX+ can be used, like specifies it the documentations transmitted by NVIDIA.

As for the frequencies of operation, they are strictly identical to those recommended by NVIDIA with 738 MHz for the graphics chip, 1836 MHz for the processors of flows and 1100 MHz for the memory. Let us finish by the exits which one at the very least finds on this card since they are one nothing unusual. Gigabyte indeed proposes a connector DVI, a connector VGA and a connector HDMI. Of what to be able to connect any screen, or almost, without passing by an adapter.

Remain a problem to be underlined, as you can see it on the photographs, the total completion of this card which seems rather average. The cooling system is assembled twisted, whereas the pressure which he exerts on the PCB seems to twist this last. Difficult to say if it is of an exclusive problem to our sample of test or a more general phenomenon.

Performance

To test the performances of GeForce GTS 250, we used a recent platform whose detail appears below:

* Motherboard Extreme Asus Rampage II (BIOS 1204),
* Processor Intel Core i7 920 2.66 GHz,
* 6 Gb of memory DDR3-1066 - OCZ,
* RAID 0 - 2x Disks hard Western Digital Raptor 300 Gb


Our system operated under Windows Vista Edition Integral with Service Pack 1. It was about the version 64 bits of the operating system. On this machine we will extremely logically test GeForce GTS 250 but also GeForce 8800 WP, GeForce 8800 GTS 512 Mb but also GeForce GTX 260. We will also test Radeon HD 4850 and Radeon HD 4830, in version 512 Mo, without forgetting Radeon HD 4890 which is used to us as Master standard.

Side drivers, we had resort to drivers for NVIDIA GeForce 185.63 beta and the CATALYST 9.4 beta. It should be noted that the protocol of test comprises the last plays inaugurated at the time of the test of Radeon HD 4890: Tom Clancy' S H.A.W.X fascinating charges with it DirectX 10.1 and BurnOut Paradise history not to make the dead end on the plays of race.


3DMark Vantage

First test impossible to circumvent of our arsenal, 3DMark Vantage gives its preference to Radeon HD 4890 follow-up of GeForce GTX 260. GeForce GTS 250 which interests us today locates at semi way between Radeon HD 4830 and Radeon HD 4850. In 1680x1050, GeForce GTS 250 is 16% faster than Radeon HD 4830 but it is made precede by more powerful Radeon HD 4850 5%. Vis-a-vis GeForce 8800 GTS 512 Mb, GeForce GTS 250 are faster 8% in 1920x1200.


Call Off Duty 4 - v1.7

Celebrate FPS, Call Off Duty 4 Modern Warfare gives its preference to recent Radeon HD 4890. GeForce GTS 250 finishes here in third position, and this whatever the resolution. In 1680x1050, the card of NVIDIA is 8% more powerful than Radeon HD 4850 whereas it exceeds GeForce 8800 GTS 512 Mb of approximately 7%.


Crysis Warhead - v1.1 - 64 bit

Extremely greedy, Crysis complicates gives it. Cards of NVIDIA provided with 512 Mb of video memory are not able here to follow since the resolution is higher than 1680x1050. If GeForce 8800 are thus exceeded, GeForce GTS 250 remains third with an unquestionable advantage on Radeon HD 4850. In 1920x1200, the last card of NVIDIA is 16% faster than Radeon HD 4850.


Left 4 Dead

For the zombies in madness, not of doubt, Radeon HD 4890 remains the most powerful graphics card. If GeForce GTX 260 preserves its second position, GeForce GTS 250 also remains in third place. However, the last reference of NVIDIA is only 3% swifter than Radeon HD 4850 in 1680x1050. One will note however that the more the resolution increases, plus this variation increases. Thus in 2560x1600, GeForce GTS 250 are 12% faster than Radeon HD 4850.


Far Cry 2 - v1.02

No notable change under FarCry 2 since our trinity is always composed of Radeon HD 4890, GeForce GTX 260 and finally of GeForce GTS 250. In 1920x1200, the advantage of this last vis-a-vis Radeon HD 4850 climbs at 13%. Compared with GeForce 8800 GTS 512 Mb, GeForce GTS 250 is up to 30% more powerful here. It should be said that GeForce GTS 250 profits from a quantity of video memory twice higher than GeForce 8800 GTS.


World In Conflict - v1.010
Let us finish this first page of tests by a glance with the performances obtained with the RTS Wolrd In Conflict. There is hardly upheaval in the classification, if it is only GeForce GTS 250 is made precede by Radeon HD 4850 in the resolutions of 1680x1050 and 1920x1200. Vis-a-vis GeForce 8800 GTS 512 Mb, GeForce GTS 250 are 4% more powerful in 1680x1050.

Consumption

As for each test of graphics card, we had in heart to check the electricity consumption of our various protagonists. With this intention we measure, by means of a wattmeter, the total consumption of the system. Initially at rest… then in load during a test 3DMark Vantage in 2560x1600.



One immediately notices the fall of electricity consumption in favour of GeForce GTS 250 which, thanks to its higher intricacy of engraving, consumes at rest on average 25 Watts less than GeForce 8800. In load, GeForce GTS 250 is also the graphic controller who less consumes with GeForce 8800 WP.

Conclusion

The frequencies of operation of the chip evolve/move indeed in a very moderate way with the result that in final GeForce GTS 250 are hardly 5% faster than GeForce 8800 GTS 512 Mo under certain conditions. Admittedly, in other cases of figure, the gigabytes of memory allows more important profits, in particular under Crysis where one observes deltas of 12% between the two cards.

But the commercial name of this card remains a true problem is quite simply misleading! Because finally, GeForce GTS 250 does not have anything familiar with the chips GeForce 200 which one finds with GeForce GTS 260,275,285 and 295. Not, NVIDIA re-elects an existing product, already old, while trying to make it pass for nine. It is alas nothing and GeForce GTS 250 is only the consequence of the choices operated when designing the GeForce chips of series 200. By making a too ambitious chip, because too complex, NVIDIA deprived itself of the possibility of declining it. Blow, GeForce GT200 remains confined on the high-end whereas the firm with the chameleon can propose only chips based on architecture GeForce 8, alias G80, on the entry and the midrange. The strategy could have functioned … but it was without counting on the renewed aggressiveness of AMD. AMD which proposes graphics chips to him exploiting DirectX 10.1, a specificity always absent from chips NVIDIA.

Sold in 150 euros including all taxes, GeForce GTS 250 in its version 1 Gb better than Radeon HD 4850 provided with 512 Mb makes certainly, but it is also more expensive. Consequently, if the card of NVIDIA seems positioned well in term of performances, its price appears one to us nothing too high for really convincing.

Source: techarena