New Toys!

EVGA 8800GT Video Card,

Gigabyte GA-P35-DS3L Motherboard,

and Thermalright Ultima-90 Heatsink:

New Toys! Let's See How Well They Play Together

How I Ended Up with More New Hardware than I Expected

It all started innocently enough. After my review of the Asrock 775Dual-VSTA motherboard, I was pleased with my ability to combine the faster Core 2 Duo processor with my existing 6800GT video card, breathing new life into my system when it came to playing games. I anticipated that I eventually would upgrade to a faster video card, and I hoped to take advantage of the fact that the Asrock motherboard had a PCI-E video card slot, as well as an AGP slot (which my 6800GT used). In the meanwhile, I was waiting for a high performance video card to fall into the $250 range. At the time, the most appealing card was an 8800GTS 320MB, but it was selling for around $300.

In November, 2007, Nvidia released their 8800GT. This new video card is based on a refresh of the G80 chip, which is used on Nvidia's 8800GTS, 8800GTX, and 8800Ultra video cards. While this new chip, the G92, lacks some of the power of the prior one, it incorporates a die shrink from the G80's 90nm process down to a 65nm process. The 8800GT takes advantage of the ability of this new chip to run at relatively high speeds, and it makes up for a slight loss in rendering power in this way. The 8800GT also couples the G92 with high speed memory, which offsets its narrower memory bus. As a result, the actual performance of the 8800GT falls somewhere between that of the 8800GTS and the 8800GTX. Even more exciting is the news that these new cards are suppose to sell for under $250. (Anandtech has a review of a reference-design 8800GT, in which you can find technical details and performance comparisons, for those who are interested in an in-depth analysis of this card.)

However, this isn't exactly the way things worked out, at least not immediately. The high demand for these new cards pushed their availability down and their price up. It's only been more recently that the availability has increased to the point that the prices have settled into the $240-$250 range. After Christmas, I decided to go ahead and pick up one of these cards. At the time, the best price was on a stock EVGA 8800GT (i.e., not factory overclocked). Although I paid $260 for the card, it did come bundled with Enemy Territory: Quake Wars, which helped justify my paying more than $250. My new EVGA 8800GT is pictured below, along with the 6800GT it is replacing. The 8800GT is slightly longer than the 6800GT, but it is a single slot card. Its reference design cooler is much slimmer and lighter than the massive, all copper heatsink that Leadtek came up for the 6800GT.

Unfortunately, when I installed my new 8800GT, I discovered that it wouldn't work with my Asrock 775Dual-VSTA motherboard. The 8800GT is one of the first video cards designed around the new PCI-E 2.0 standard. While this card is meant to be backward compatible with PCI-E 1.0 slots (since at this time, almost all motherboards are still based on the older standard), it wasn't happy with the rather unique PCI-E slot that Asrock implemented on the 775Dual-VSTA. Apparently, in order to get both an AGP slot and a PCI-E, video card slot on the same motherboard, Asrock bent the rules a bit. Faced with this problem, I decided that my Asrock motherboard had served its purpose, and it was now time to move on.

Update: After I had gone ahead and purchased a new motherboard, as described in the next section, Evga made available a new BIOS for their 8800GT cards that addresses the issue of motherboard compatibility. No specific mention is made of the 775Dual-VSTA motherboard, but it sounds like this new BIOS should take care of the kind of problem that I ran into. If this BIOS had been available before I purchased my new motherboard, I probably would have tried it out. As I don't have much need for it, now, I've not put it to the test.

If you have an 8800GT from another manufacturer and are facing these kind of compatibility issues, check with your card's manufacturer to see if they have an updated BIOS available. If they don't and you remain determined to get your 8800GT to work on your not fully compatible motherboard, take a look at what is available at mvktech.net. They have a lot of video card BIOS files archived, and they are the home of the Nvidia Bios Editor or NiBiTor. Check their forums for more information.

From here on, all new EVGA 8800GT video cards include this new BIOS; so, this problem of compatibility with some motherboards should not affect future purchasers of the 8800GT. I assume that other card manufacturers are doing something similar, though I don't have any information on this.

Gigabyte GA-P35-DS3L (Rev. 2) Motherboard

I knew that I eventually would want to upgrade my Asrock motherboard. Although the 775Dual-VSTA filled its role as a stepping stone board admirably, its inability to overclock my processor and its lack of a full bandwidth PCI-E slot were two, significant, limiting factors. Thus, it was easy for me to reach my decision to look for a new motherboard. After a quick search, I settled on the Gigabyte GA-P35-DS3L.

This motherboard is the minimalist model in Gigabyte's lineup of P35 chipset motherboards. It lacks the features that higher-end motherboards can boast of, such as RAID and firewire, but it offers good performance at a relatively low price. (It is also distinguished by still having external serial and printer ports, which the other Gigabyte P35 boards have done away with.) I picked this board up from Mwave for $89, and the possibility of getting a high performance motherboard for under $100 is what sold me on it. In addition to being able to run my 8800GT video card, which was my primary motivation for buying a new motherboard, I knew that this motherboard would provide a much better opportunity for overclocking my Core 2 Duo E6600 processor. The Asrock motherboard never achieved more than 288 MHz on the frontside, which only allowed an 8% overclock, and I knew that this new motherboard was capable of much higher bus speeds. (If you are interested in a professional review of this motherboard, take a look at what X-BitLabs had to say about it. If you are thinking of overclocking a quad-core processor and applying voltages up into the 1.5 volt range, you will definitely want to see why their testing revealed that this board is not a good choice.)

In examining this board, one of the first things that I noticed was that it is a bit smaller than most full-size ATX motherboards. Looking at the picture above, the "height" (or distance along this motherboard's shortest side) is a over an inch less than Gigabyte's other P35 motherboards. I also noticed that the positioning of the EIDE connector seemed unusual. The green IDE connector is placed towards what would be the bottom of the case when it is installed, and the connector is oriented so that it would be parallel to the bottom of the case. Usually, these connectors are turned 90 degrees, which would match the orientation of the black floppy connector (seen next to the main power connector). In fact, placing the EIDE connector next to the floppy connector, and in line with it, would have been a better arrangement. The current location and orientation is less than ideal for hooking up an IDE optical drive in a tower enclosure. Since the IDE connector is relatively far away from the optical drive bays, in my case, I have to stretch the ribbon cable over the top of the intervening cards in order to barely reach the lowest bay.

The other thing that concerned me about the layout of this board is how close the DIMM slots are placed to each other. If you fill all four slots, your memory modules will be virtually touching each other.

It's not just this motherboard that has the issue; all the Gigabyte P35 motherboards have their memory slots laid out this way. Although I temporarily filled all four slots for a quick check of how well it would work, I only plan on using two memory modules. In dual channel configuration, the two slots that will be used are not adjacent to each other.

I also have a minor complaint about the BIOS. In order to get to some important BIOS settings, such as memory timings, you have to hit Ctrl + F1 at the main BIOS screen, before opening the relevant submenu. This is only documented in the manual; there is no mention of this within BIOS, itself. It is a bit annoying that you have to remember to take this step each time you enter the BIOS.

Another annoyance was that I was unable to get the AHCI setting to work. The Advanced Host Controller Interface is what makes possible some of the fancy SATA features, such as Native Command Queing and hot plug capability. After setting this to enabled in the BIOS, I loaded the Intel SATA drivers during the Windows XP installation (using the F6 procedure), but the operating system was unable to find my hard drive afterwards. I'm not sure what the problem was. Perhaps the fact that this chipset doesn't support RAID also interferes with its support for AHCI. I must admit that I don't have any real need for these features and I don't really miss them.

On the plus side, one nice feature of the Gigabyte motherboard BIOS (which I've not seen implemented on other motherboards) is how you can save a number of BIOS configurations as files. This makes it easy to change configurations. For example, I can have one configuration handy for everyday computer use (with power management and fan control turned on) and have another configuration to use as a starting point for overclocking and stress testing (with power management and fan control disabled).

The BIOS keeps track of a number of previously successful boot configurations, as well. So, even if you don't manually save your favorite configurations, you can make use of one of your previously successful boot configurations, if your BIOS tweaking gets you into trouble.

In general, this motherboard does a pretty good job of recovering from a configuration that won't POST. It will automatically set the CPU speed, front side bus, and memory timings to something close to the default configuration and reboot itself.

A more serious question is whether the northbridge chip is adequately cooled. The heatsink used is not particularly big, and you can tell simply by touching it that it gets quite warm. Using an inexpensive thermal probe from Scythe, I measured the temperature of the heatsink, and it reached as high as 143 deg. F. (62 deg. C.) when stress testing the system. (I took my measurements with the case closed. The picture below is just to illustrate how I jammed the thermal probe between the fins on the heatink to get a temperature.)

Of course, the chip itself is going to be even hotter than than the heatsink's temperature. Although one option would be to place a small fan on the heatsink, I really don't like small, noisy fans; so, I looked into replacing the stock heatsink with something larger. I even picked up this one from Zalman, but after receiving it, I doubted whether it was be big enough to do better than what was already in place. (I suspect that something like this heatpipe model from Thermalright is what is needed to fully cool the northbridge chip in a passive manner.) In the meanwhile, I jerry-built a paper air-duct to re-direct air from the CPU's fan down onto the northbridge heatsink. The 120mm fan on my Ultima-90 heatsink sticks up above the cooler, and the duct captures the air flow from there and sends it down to the the northbridge heatsink.

Somewhat surprisingly, this seems to have worked out quite well. (MacGyver would be pleased.)

Perhaps at some point, I'll try out the Zalman heatsink, as well; however, this will require removing the motherboard from the case, in order to uninstall the current heatsink. I think that this upgrade will have to wait until I am ready to lap the Ultima-90 heatsink, at the same time.

I also took a quick look at how hot the 2x2 12 volt power connector gets on this motherboard, because Xbit's review suggested that this could be a problem under heavy load. They actually managed to get this connector hot enough to fuse together the plug and the socket while overclocking a quad core QX6700 processor. At the time, they had the core voltage at 1.5 volts, and they were running the frontside bus at 360 MHz. The fact that their Zalman CNPS9700 heatsink was unable to keep the core temperatures below 70 degrees C. indicates how much power the CPU was consuming under these conditions.

My testing circumstances were not so extreme, though they are about as far as I'm likely to push my hardware. I checked the temperature on the outside of the 2x2 12 volt socket while running my dual core processor at 3.4 GHz with the core voltage set to 1.39 volts. After running Prime95 for about a half hour, the temperature on the socket was around 117 degrees F. (47 degrees C.). While this is a bit warm, and the temperature inside the socket is no doubt higher, I don't think this is hot enough to cause a problem.

Thermalright Ultima-90 Heatsink

In anticipation of being able to see how far I could overclock my processor, I went ahead and invested in an after-market heatsink and fan. Although not the largest heatsink in Thermalright's lineup, I opted for the Thermalright Ultima 90. It performs nearly as well as the larger heatsinks, but it offers a more compact and light weight design. (Anandtech did a full review of this heatsink and compared it to several other high-end models.) Part of this heatsink's appeal is that you can mate it with a 90mm fan and fit it into smaller cases, but I went ahead and purchased a 120mm fan, which this heatsink also accommodates. I decided to use a 4-wire PWM model from Scythe; so, I could take advantage of the motherboard's PWM fan-control.

As you can see below, the Ultima 90 (especially after you put a 120mm fan on it) still is a pretty big heatsink.

I was aware of complaints that Thermalright heatsinks do not have flat contact surfaces; so, one of the first things that I did was to take a look at the bottom of the heatsink. Since it's not easy to eyeball the flatness of a heatsink, I tried a simple demonstration. I placed a drop of water on a pane of glass and then looked at the heatsink from below to see what occurred. Ideally the water should have spread out evenly in all directions from the center of the heatsink. Instead, the water filled a kind of channel running the length of the heatsink along one axis. If you look carefully, you can see this in the picture below.

The darker band reaching from one edge of the heatsink to the other, running in the same direction as the heatpipes, is where the water collected, indicating a concave, bowed aspect to the bottom of this heatsink.

You also can get some sense of how flat the bottom of the heatsink is by looking at how it reflects an image. The bottom of this heatsink isn't finely polished or mirror-like; nevertheless, you can make out the concave aspect by comparing reflections along the two axes.

In the top picture, lines in the reflection are running in the same direction as the concave channel in the bottom of the heatsink. Consequently, the reflection doesn't look too distorted. In the second picture, the heatsink has been turned 90 degrees so that the reflected lines cross the bowed area of the heatsink, resulting in a more distorted reflection. Ideally, both reflections should have been distortion free.

I decided to go ahead and make use of this heatsink as it was; instead of going through the process of lapping the bottom to make it flatter. I may do that in the future, but I wanted to see what kind of performance it offered straight out of the box.

The process of installing this heatsink went smoothly for such a large cooler, and the final result felt quite secure. I oriented the heatsink and its fan so that it was blowing towards the case's rear exhaust fan. (No, that rear fan is not a 80mm fan; it actually is a 90mm fan. However, it definitely is dwarfed by the 120mm fan on the heatsink.)

As you can see, with the 120mm fan installed, this heatsink takes up every bit of my case's 8 inch width.

As best as I can tell, the fan mounting wires don't actually make contact with the side of the case, when it is closed up, but there obviously is not a lot of clearance there, either. I think that I can safely say that an 8 inch wide case is the bare minimum for installing this heatsink with a 120mm fan attached.

So, how well does this Thermalright heastsink work? I made some comparisons, below, between the stock Intel heatsink and the Ultima-90. I have the fans set to run at full speed for these measurements, which is around 1700 rpm for the Intel heatsink's fan and 1190 rpm for the 120mm Scythe fan. For these comparisons, the processor is running at it's default 2.4 GHz and the voltage setting is the default 1.2625 volts. I used SpeedFan (v4.33) to check the temperatures of the CPU's two cores.

Unfortunately, I didn't get core temperatures running the Intel heatsink under stress; so, I can't directly compare it to the Thermalright heatsink under these conditions. Instead, I can only show how the Ultima-90 behaves while increasing CPU's stress load. For my purposes, the most interesting question is how well does this heatsink control temperatures under full load, as the CPU clock speed and the core voltage are increased during overclocking efforts. (Two threads of Prime95 v2.55 were run for at least 30 minutes to get a stable core temperature.)

Although the highest temperature achieved of 52 degrees C. is well within this processor's operating range, I think it is noteworthy to look at how quickly the temperatures start to rise once we get beyond clock speeds that can be run at the default voltage setting. With each increase in speed beyond 3.0GHz, a bump up in the core voltage is required to keep the processor's operation stable. As a result the temperatures start to climb significantly with each small increase in performance. For those who are intent upon wringing more performance out of their CPU's, by increasing the voltage, a good cooling solution is essential to doing so safely.

Overclocking the E6600 CPU

It's the need for more voltage that limits what each processor can achieve when it comes to overclocking. As we can see below, once we get beyond 3.0GHz, which is about as fast as this processor will go with stock voltage settings, the need for more power increases with each small increase in CPU speed. Once we reach 3.2 GHz, each 3% increase in speed requires a 4% increase in power to the CPU.

For me, this chart is very informative, as it indicates the diminishing returns in processor performance, once we start to increase the voltage over the stock setting to achieve higher clock speeds. Although Intel's specification sheets list an operating range from 0.85 volts to 1.5 volts, the factory default voltage setting is much less than 1.4 volts. After looking at the above results, I concluded that 3.2Ghz was the optimal overclocking goal for my particular processor. This is achievable with only a small increase in voltage. Pushing the voltage up into the 1.5 volt range just doesn't offer that much more performance to justify the increased wear and tear on my CPU.

Memory Settings and Performance

Since this motherboard offers much higher memory speeds than the maximum of DDR2-667 available on the Asrock motherboard, it seemed worthwhile to see if these greater memory speeds had much impact on performance. Interestingly, the pattern that emerged with the Asrock motherboard continues with this Gigabyte motherboard. The Core 2 Duo processors are just not hungry for memory bandwidth. The onboard memory cache and the "look ahead" memory algorithms keep latency issues from interfering with performance, as well. While faster memory speeds and tighter memory timings do yield more memory bandwidth and lower latencies, the actual benefits in application performance are relatively small.

Below, we can see how the amount of memory bandwidth scales with different memory speeds and settings. It is interesting to note how the 400 MHz bus results in better bandwidth, even though the actual memory speed is less (2.4x356=854 MHz versus 2x400=800 MHz).

Memory latencies seem to be follow the same pattern.

However, when we look at whether faster memory speeds, greater memory bandwidth, and lower memory latencies make much difference in actual performance, we see that the gains are relatively small. A benchmark using the number crunching program Prime95 illustrates this point. (Lower times means better performance.)

Consequently, I can't see a good reason to spend much time tweaking the memory in order to find the fastest speed and the lowest timings that it will run at. Simply setting it to run at its default DDR2-800 speed and 4, 4, 4 timings seems to be good enough.

8800GT Video Card Performance

The purchase of this new motherboard led us through a digression, as we took some time to look at what it had to offer. In particular, I found out that I my CPU would comfortably overclock from 2.4 GHz to 3.2GHz. Now, we can pick up where this story began. Did the purchase of a new 8800GT video card deliver the kind of game performance that I had hoped for?

Aquamark 3 is an older benchmark and doesn't provide an opportunity to test out the latest graphics features; however, I like it because the results correlate fairly well with the performance I see in the popular Battlefield 2 series. (Note: The overall Aquamark3 score is actually the average frame rate multiplied by 1000; so, a score of 73,033 translates to an average of 73.0 frames per second. I'm not sure how the Graphics and the CPU scores are derived.)

The new video card almost doubles the performance that the 6800GT was capable of on this benchmark. We also can see how the low resolution (1024x768) that this benchmark is run at keeps the 8800GT from fully stretch its legs; increasing the CPU speed from 2.4GHz to 3.2GHz delivers a further 20% increase in video rendering performance.

Company of Heroes is a popular real time strategy game, and it includes a performance test built into the game. The first graph, below, uses the settings at which I was able to play the game with the 6800GT video card (a 1024x768 resolution with a mix of high and medium graphics settings).

Company of Heroes is based on a newer game engine than the one used in the Aquamark 3 benchmark, and it makes heavier use of advanced graphics features; consequently, the difference in performance between the 6800GT and the 8800GT is even more remarkable. The average frame rate is four times greater using the 8800GT. Once again, the fact that increasing the CPU's speed results in even better performance suggests that the 8800GT is not really being taxed at the above settings.

Below, I've taken a look at how the 8800GT's performance scales with higher resolutions, and I again compare the results between a 2.4GHz and a 3.2GHz processor. All the graphics features are set to the highest settings (except for leaving anti-aliasing off). The two resolutions that I used were 1280x1024 and the game's highest resolution, 1600x1200.

At these higher resolutions and higher graphics features settings, we can see that increasing the CPU's speed doesn't result in quite as much performance improvement as we saw at 1024x768 with lower graphics settings. At this point, the video card finally is beginning to work at its limit.

Although there is good reason to question whether the 3D Mark series of benchmarks really tells us much about actual game performance, these benchmarks remain popular and easy to use.

The older 3DMark03 presents a picture that is fairly similar to what we have seen with Aquamark3. The 6800GT is capable of handling the subtests within this benchmark; however, the 8800GT nearly triples that level of performance.

The 3DMark05 and 3DMark06 benchmarks essentially repeat this same pattern of results, though 3DMark05 seems to show a bit more influence from the CPU speed for some reason. On these tests, the 6800GT is struggling to attain reasonable frame rates, while the 8800GT is in its element.

Overclocking the 8800GT

We've overclocked the processor to see what might be gained from this; so, what sort of benefit might come from overclocking the video card? In addition to the reference version of the 8800GT, which I'm running, Evga also sells three factory overclocked models. The stock model runs at 600MHz/900MHz (GPU core/Memory speed); the SC model runs at 650/950MHz; the KO model runs at 675/975MHz; and the SSC model runs at 700/1000MHz. I thought that it would be interesting to see if my stock 8800GT could also run at these factory overclocked speeds and to see what might be gained by this, if it could.

I ran the Company of Heroes Performance Test at 1600x1200 with all the settings at their highest, except for anti-aliasing, which was left turned off.

We see that we get some modest gains in performance with overclocking, about 16%, comparing the stock settings to the SSC speeds.

I tried the same sort of comparison using the benchmark that is integrated into the Crysis Demo. While my video card seemed to have no trouble with the higher timings when running Company of Heroes, I found Crysis to be much less forgiving. A 671MHz core overclock and a 973MHz memory overclock were the highest speeds that I could set and still complete all four passes of the scripted benchmark demo.

Apparently, Evga must be doing some selective binning of their GPU and memory chips in order to guarantee that their KO and SSC models will run at their advertised speeds. Does it make sense to spend more to get one of these factory overclocked 8800GT video cards? Below you can see how Evga lists the prices for these different models on their website. (It should be noted that only the stock version is shown as actually being in stock, at this time; all the overclocked versions are back-ordered.)

If Evga sticks with this pricing scheme, the prices scale in a reasonable fashion compared to the gains in performance. The SSC model sells for 11% more than the stock model, and you can expect maybe a 16% gain in performance. However, since these overclocked models seem to be in short supply, the prices you can actually find on these overclocked cards might be higher than those shown above. If that seems to be the case, you might want to stick to the stock model, or wait for the prices on the overclocked models to drop back into line. (Also, I would use whatever actual prices you can find on the stock model as the baseline for determining how much more you should pay for the overclocked models. For example, Newegg has the basic model on sale for $240 with free shipping, at this time; given that option, I'm not sure that I would want to pay more than $265 for the SSC model.)

8800GT's Heatsink and Fan

Before wrapping this up, I wanted to say a few words about the cooling that the 8800GT comes with. Apparently, the model that I received includes the updated fan and heatsink. You can distinguish the newer cooler by the larger fan opening (compare the picture of my 8800GT, at the top of this page, to the one in Anandtech's review of the 8800GT). Reportedly, this fan is quite a bit quieter than the one that came with the earlier 8800GT models. However, some of this quietness may be at the expense of some rather high GPU temperatures. In a cool room of around 65 degrees F. (18 degrees C.), my video card idles at a temperature of around 49 degrees C. Playing Crysis for 45 minutes and recording the GPU temperatures with Nvidia Monitor, the GPU temps peaked at around 85 degrees C. They then dropped back down to 78 degrees C., before settling into the low 80's range.

These temperatures sound pretty high (85 degrees C. is equivalent to 185 degrees F.), but they appear to be within this GPU's operating range. I think the GPU doesn't start to throttle back to lower clock speeds until it reaches 100 degrees C.

I considered whether these relatively high temperatures might have some impact on how well I could overclock the GPU. I manually set the fan speed to 60%, which dropped the maximum temperature seen while playing Crysis from 84 degrees to 63 degrees C., but I still could not finish the Crysis benchmark with a 679 MHz core speed. At least with my 8800GT, better cooling doesn't look like it will help me achieve any better overclocking results.

I mentioned earlier that Evga made available a new BIOS for their 8800GT video cards. In addition to addressing the compatibility issue that a few motherboards had with this card, the new BIOS changes the manner in which the fan operates. It begins to spin up earlier, as the GPU temperature rises, and it keeps the peak temperature about 10 degrees cooler than it was before. I'm not sure these changes are really necessary; so, I've still not upgraded the BIOS on my video card.

Conclusion

Overall, I'm quite pleased with how my new hardware works. The new Gigabyte motherboard provides a nice platform for getting the full potential out of my CPU and memory. My CPU is now overclocked by 33%, and my DDR2-800 memory, finally, is running at its rated speed. My new 8800GT video card is able to take full advantage of this, since it has a 16X PCI-E video card slot to work with (instead of the 4X PCI-E slot that the Asrock board offered). At only $89, the GA-P35-DS3L provides a lot of performance for a modest amount of money.

The Thermalright Ultima-90 heatsink is very capable of keeping my E6600 cool, even when overclocked. As an added bonus, the 120mm fan that I'm using can provide some extra cooling for northbridge heatsink. The bottom of this heatsink isn't as flat as I might have expected, especially on a relatively expensive heatsink ($46); however, I can't say that this has significantly affected its performance. Perhaps, I could have gotten away with a less expensive cooler, such as the Arctic Cooling Freezer 7 Pro (about $25, including fan). (See Anandtech's review of this heatsink for more details.)

Purchasing the 8800GT video card is what led to my additional hardware upgrades; yet, I have no complaints, as the end result is quite satisfying. The 8800GT greatly surpasses what my 6800GT was capable of. On older games, it at least doubles the 6800GT's performance, and on newer games the advantage is even greater.

[Back to Sequoyah Computer Home]

[February 5, 2008]