Thermalright Ultima-90: Small Wonder?
by Wesley Fink on August 20, 2007 2:00 AM EST- Posted in
- Cases/Cooling/PSUs
CPU Cooling Test Configuration
The standard test bed for cooling tests is the EVGA NVIDIA 680i SLI motherboard. This is primarily based on the consistent test results on this board and the NVIDIA Monitor temperature measurement utility, which is part of the nTune program.
NVIDIA Monitor has a drop-down pane for temperature measurement which reports CPU, System, and GPU measurement. At this point reviews will concentrate on CPU temperature. In addition to the real-time temperature measurement, NVIDIA Monitor also has a logging feature which can record temperature in a file in standard increments (we selected every 4 seconds). This allows recording of temperatures during testing, which can then be reviewed when the stress tests are completed. There is also the handy reference of speeds and voltages in the top pane to confirm setup.
NVIDIA Monitor was compared to test results from the Intel TAT (Thermal Analysis Tool). Intel TAT CPU portions do function properly on the EVGA 680i motherboard, but the chipset-specific features do not operate as they should. Idle temperatures in TAT were in line with measured idle temps with NVIDIA Monitor. The CPU stress testing with TAT pushing both cores showed TAT stress temps at 80% CPU usage roughly corresponded to temps reported in our real-world gaming benchmark.
Other components in the cooling test bed are generally the same as those used in our motherboard and memory test bed:
All cooling tests are run with the components mounted in a standard mid-tower case. The idle and stress temperature tests are run with the case closed and standing as it would in most home setups. We do not use auxiliary fans in the test cooling case, except for the Northbridge fan attached to the 680i for overclocking. Room temperature is measured before beginning the cooler tests and is maintained in the 20 to 22C (68 to 72F) range for all testing.
Thermalright provides a small tube of premium thermal grease with the Ultima-90. However, for consistency of test results we used our standard premium silver-colored thermal compound. In our experience the thermal compound used makes little to no difference in cooling test results. This is particularly true now that processors ship with a large manufacturer-installed heatspreader. Our current test procedure uses this standard high-quality silver-colored thermal paste for all cooler reviews.
We test the stock Intel cooler at standard X6800 speed, measuring the CPU temperature at idle and while the CPU is being stressed. The CPU is stressed by running continuous loops of the Far Cry River demo. The same tests are repeated at the highest stable overclock we can achieve with the stock cooler. Stable in this case means the ability to handle our Far Cry looping for at least 30 minutes. The same benchmarks are then run on the cooler under test at stock, highest stock cooler OC speed (3.73GHz), and the highest OC that can be achieved in the same setup with the cooler being tested. This allows measurement of the cooling efficiency of the test unit compared to stock and the improvement in overclocking capabilities, if any, from using the test cooler.
Noise Levels
In addition to cooling efficiency and overclocking abilities, users shopping for CPU cooling solutions may also be interested in the noise levels of the cooling devices they are considering. Noise levels are measured with the case on its side using a C.E.M. DT-8850 Sound Level meter. This meter allows accurate sound level measurements from 35bdB to 130dB with a resolution of 0.1dB and an accuracy of 1.5dB. This is sufficient for our needs in these tests, as measurement starts at the level of a relatively quiet room. Our own test room, with all computers and fans turned off, has a room noise level of 36.4dB. Procedures for measuring cooling system noise are described on page five, which reports measured noise results comparing the stock Intel cooler and recently tested CPU coolers to the Thermalright Ultima-90.
The standard test bed for cooling tests is the EVGA NVIDIA 680i SLI motherboard. This is primarily based on the consistent test results on this board and the NVIDIA Monitor temperature measurement utility, which is part of the nTune program.
NVIDIA Monitor has a drop-down pane for temperature measurement which reports CPU, System, and GPU measurement. At this point reviews will concentrate on CPU temperature. In addition to the real-time temperature measurement, NVIDIA Monitor also has a logging feature which can record temperature in a file in standard increments (we selected every 4 seconds). This allows recording of temperatures during testing, which can then be reviewed when the stress tests are completed. There is also the handy reference of speeds and voltages in the top pane to confirm setup.
NVIDIA Monitor was compared to test results from the Intel TAT (Thermal Analysis Tool). Intel TAT CPU portions do function properly on the EVGA 680i motherboard, but the chipset-specific features do not operate as they should. Idle temperatures in TAT were in line with measured idle temps with NVIDIA Monitor. The CPU stress testing with TAT pushing both cores showed TAT stress temps at 80% CPU usage roughly corresponded to temps reported in our real-world gaming benchmark.
Other components in the cooling test bed are generally the same as those used in our motherboard and memory test bed:
| Cooling Performance Test Configuration | |
| Processor | Intel Core 2 Duo X6800 (x2, 2.93GHz, 4MB Unified Cache) |
| RAM | 2x1GB Corsair Dominator PC2-8888 (DDR2-1111) |
| Hard Drive | Hitachi 250GB SATA2 enabled (16MB Buffer) |
| Video Card | 1 x EVGA 7900GTX - All Standard Tests |
| Platform Drivers | NVIDIA 9.53 |
| NVIDIA nTune | 5.05.22.00 (1/16/2007) |
| Video Drivers | NVIDIA 93.71 |
| CPU Cooling | Thermalright Ultima-90 ZEROtherm BTF90 Xigmatek AIO (AIO-S800P) Evercool Silver Knight Enzotech Ultra-X 3RSystem iCEAGE Thermaltake Big Typhoon VX Thermaltake MaxOrb Scythe Andy Samurai Master Cooler Master Gemini II Noctua NH-U12F Asus Silent Square Pro Scythe Ninja Plus Rev. B OCZ Vindicator Thermalright Ultra 120 Extreme Thermalright Ultra 120 Scythe Infinity Zalman CNS9700 Zalman CNS9500 Cooler Master Hyper 6+ Vigor Monsoon II Lite Thermalright MST-9775 Scythe Katana Tuniq Tower 120 Intel Stock HSF for X6800 |
| Power Supply | OCZ PowerStream 520W |
| Motherboards | EVGA nForce 680i SLI (NVIDIA 680i) |
| Operating System | Windows XP Professional SP2 |
| BIOS | Award P26 (1/12/2007) |
All cooling tests are run with the components mounted in a standard mid-tower case. The idle and stress temperature tests are run with the case closed and standing as it would in most home setups. We do not use auxiliary fans in the test cooling case, except for the Northbridge fan attached to the 680i for overclocking. Room temperature is measured before beginning the cooler tests and is maintained in the 20 to 22C (68 to 72F) range for all testing.
Thermalright provides a small tube of premium thermal grease with the Ultima-90. However, for consistency of test results we used our standard premium silver-colored thermal compound. In our experience the thermal compound used makes little to no difference in cooling test results. This is particularly true now that processors ship with a large manufacturer-installed heatspreader. Our current test procedure uses this standard high-quality silver-colored thermal paste for all cooler reviews.
We test the stock Intel cooler at standard X6800 speed, measuring the CPU temperature at idle and while the CPU is being stressed. The CPU is stressed by running continuous loops of the Far Cry River demo. The same tests are repeated at the highest stable overclock we can achieve with the stock cooler. Stable in this case means the ability to handle our Far Cry looping for at least 30 minutes. The same benchmarks are then run on the cooler under test at stock, highest stock cooler OC speed (3.73GHz), and the highest OC that can be achieved in the same setup with the cooler being tested. This allows measurement of the cooling efficiency of the test unit compared to stock and the improvement in overclocking capabilities, if any, from using the test cooler.
Noise Levels
In addition to cooling efficiency and overclocking abilities, users shopping for CPU cooling solutions may also be interested in the noise levels of the cooling devices they are considering. Noise levels are measured with the case on its side using a C.E.M. DT-8850 Sound Level meter. This meter allows accurate sound level measurements from 35bdB to 130dB with a resolution of 0.1dB and an accuracy of 1.5dB. This is sufficient for our needs in these tests, as measurement starts at the level of a relatively quiet room. Our own test room, with all computers and fans turned off, has a room noise level of 36.4dB. Procedures for measuring cooling system noise are described on page five, which reports measured noise results comparing the stock Intel cooler and recently tested CPU coolers to the Thermalright Ultima-90.
Facebook
Twitter
RSS
38 Comments
View All Comments
Wesley Fink - Monday, August 20, 2007 - link
Corrected.CZroe - Monday, August 20, 2007 - link
"The Ultima-90 is also the first Thermalright we have tested with four sets of fan wire mounting holes. This means with the right fan wires you can mount two fans in a push pull configuration."So why not test it in this configuration? You had both fans, and one of your statements even seems to say this when taken out of context:
"our test configuration needs to look at the Ultima-92 with both a 120mm fan and a 92mm fan."
INDEED! The Ultra-120 Extreme could mount two fans with a second set of clips and rubber, so I always thought that it was a shame that they weren't included. Sure the clips were designed to use the same holes (they didn't have to be), but one could easily be forced into the second set of holes from the top and the other could be forced into the second set of holes from the bottom.
Anyway, I noticed that there were no views of the underside or mention of the finish. I assume that it's identical to the Ultra-120 Extreme (which I am regretting buying after reading this ;)).
Also, the comparison showing both 120 and 92mm fans mounted shows the rubber strips mounted incorrectly for the 92mm fan. Does the final retail version include two more of these for a push-pull config?
Also, your cooler comparisons had me strongly considering a Cooler Master Hyper 6+, but it is 100% unavailable. I tried my darndest to buy one. Cooler Master does not offer it, so it only makes sense to remove it from the comparison. I know that it fills a unique position, but in light of availability, it really needs to go (especially seeing how crowded the chart is getting).
Wesley Fink - Monday, August 20, 2007 - link
The Ultima-90 includes one set of rubber strips, one set of 120mm fan clips, and one set of 92mm fan clips. Since the height is different than the Ultra-120 series those fan clips will not work and extras for the Ultima-90.Therefore, we did not have the clips for testing push-pull with matched fans. We will try to test this when we do a retest of some of the top units.
The finish is about the same as the Ultra-120 eXtreme, and Thermalright has written many buyers who questioned the curvature that the " . . . the convex surface is made that way to work well with the bolt down retention mechanism."
CZroe - Wednesday, August 22, 2007 - link
I know that the clips on the Ultra-120 Extreme are different, so I wasn't suggesting/considering using clips from one on another (what a waste of an Ultra-120 Extreme ;)). Just testing the included clips with the two fans you've been using simultaneously would be enough.I certainly wasn't suggesting matched fans. I've been told that push/pull configurations only help with a lower CFM fan pushing and a higher CFM fan pulling or else you get no increased CFM or decreased noise. This was the consensus when I had to quiet down my Shuttle SN45G XPC (SFF) a few years ago with a fan duct and some case modding.
Thanks for the info about the finish. Considering the top-end performance of the Ultra-120 Extreme, I don't think anyone will question their decision to do that (it certainly didn't impact performance!).
Egglick - Monday, August 20, 2007 - link
This is an excellent article, along with what sounds to be an excellent heatsink. However, I must echo some of the same feelings that other posters have been expressing:Fix the test system's setup.
If we want to get a legitimate idea of how loud a cooler is, you're going to have to do better than 38db @ 24in for a noise floor. Swap out the videocard for a fanless model, and if necessary switch to a quieter PSU. Hell, go with an 8500GT if you need to. We're not testing the framerate of the videocard, we're testing the noise/performance of CPU coolers.
Let readers judge the heat/noise of videocards on their own merit.
yacoub - Monday, August 20, 2007 - link
Totally agree!!Tiamat - Monday, August 20, 2007 - link
I know its tough to show all the data, but I was wondering if you could include insets that show just the heatsinks in the neighborhood of the performance of the reviewed specimen. When displaying all of the data at the same time with much of the graphs overlapping, its very difficult to find the appropriate plot. In fact, I gave up trying after 3 seconds (standard attention span when looking at graphs) -- this kind of renders the plots useless! Usually the take home message should be as quick in the plots as it is in the text, if not quicker!My suggestion would be to show only 5 trends in the graph (including the intel retail for reference, two above, two below) while keeping all of the data in the table for those who need the ancillary information.
Thanks and I hope this helps!
Keep up the good work!
Final Hamlet - Monday, August 20, 2007 - link
I don't get the attraction of max overclocks... are they any good, except ego-pushing?What I would be really interested in is how these coolers do in a stock PC without any fans. That would be _really_ interesting... otherwise I don't see any point in buying these things or overclocking 10MHz higher (oh, great, you did it...).
coolerman - Monday, August 20, 2007 - link
Most of the good air coolers seem pretty comparable. The 30-90 MHz overclocking differences are laughable, and while the temperature varies a bit most coolers fall into one of two categories: top out at ~3.9GHz and 37-45C, or top out at ~3.8GHz with slightly higher temperatures at the lower clock. The Ultima-90 is nice in that it weighs less, but the Tuniq 120 does manage to best it in several tests - i.e. with 92mm at top OCs - and technically costs less once you add in shipping and fans. Speaking of which, am I the only one that finds the $20 fan with a $50 HSF to be humorous? Fan makers must make a decent profit, compared to the poor heatsink blokes!Also, do you think you could possibly cram any more entries into the scaling charts? Perhaps sort them by some reasonable criteria while you're at it - say, temperature for instance? The upper portion is useless, as it's just a cluster of illegible lines, and the lower table is almost sorted alphabetically, which means if I'm looking for similar coolers I have to stare for a long time. And those lovely colours… the large images help some, but fundamentally there's just too much data there now I think.
Anyway, seems like even though Thermalright got a glowing review, the iCEAGE, Hyper 6+, Vindicator (with SilenX), Ninja (with SilenX), and Tuniq 120 are all in the same ballpark. Of course, some of those are tough to find (iCEAGE anyone, or Tuniq across the pond?), but for the price the Ultima-90 really doesn't look that special. Fan + shipping means you'll pay probably $65-$70, which while better than the $85 you'd pay for Ultra-120 eXtreme is still not exactly a steal. Ninja is $45 shipped, Tuniq is about $55 shipped. $55-$60 shipped, plus a fan? You'll have to do better than that! As it is, I'll buy whatever is cheapest within reason - which is often Scythe, for the money. (Let's not even talk UK/European prices on most of this stuff, as I'm sure the majority of you don't care.... I'll just avert my lustful eyes from Newegg.com, Xoxide, etc.)
Now, my real question is how a true water cooling setup compares to all of these coolers. Fundamentally, air and water cooling are still limited to room temperature. Water simply cycles all the liquid around to a potentially larger radiator (and reservoir). I'm sceptical that most water cooling solutions will really do much better than something like the Ultra-120 eXtreme, but I'd like to see some results using your own testbed. If you ever do water, though, please run it at full load for 8 hours or so. Some water setups depend on the water starting at a low temperature, and as it heats up they can't maintain cooling efficiency. I had one that would crash after about two hours of intense gaming every time unless I turned the overclocks down.
Last but certainly not least, the "we're thinking about a new cooling test system so that our noise testing might actually have merit" talk is getting a wee bit old. It's about time to actually move on I'd say. I'd like a more moderate (fanless!) GPU in there in terms of noise, if only to allow us to see how quiet (or not) the coolers really are. Your PSU testing seems to have a bit better equipment for noise analysis, so maybe you can get some help with that area. I don't think X38 is necessary, just like 680i isn't necessary for the most part, but I suppose getting more or less useful temperature readings might be needed. Just toss everything that doesn't make the top 5 and move to the new testbed! (Easy for me to say, since I'm not doing all the work!)
Cheers,
Charles
Wesley Fink - Monday, August 20, 2007 - link
While the frequency difference may appear minor at the top, the wattage dissipation is not. For the last two reviews I have also been quoting wattage in the review commentary for comparison.A stock X6800 represents 75 watts, while that CPU at 3.83Ghz is at 150W - or double the heat to dissipate. At 3.90Ghz at the voltages required the requirement is 160 to 161W, while at the Thermalright Ultra-120 eXtreme maximum stable speed of 3.94 GHz the wattage is around 166W.
Many tests only consider air cooling results to 150W, but most readers buy secondary cooling to extend their overclock. Since the C2D overclock extremely well the wattages at some of the very high overclocks are very demanding.
We have a water-cooling review in process for comparison with these results. Once that is done we will choose a few top performers for retest and start testing with the new cooling test bed. The new test bed will require retesting of all coolers included so that is why the shift when a section like top air-coolers is completed.
Thr 92mm fan costs just over $6, while the 120 was $20. We used the Scythe S-Flex again because that was the fan used in testing the Thermalright Ultra-120 and the Ultra-120 eXtreme. We wanted to keep as many variables the same as possible.
Looking at output and noise you should be able to select a value 120mm fan below $10 that meets your specs. The Yate Loon medium output fan, for example, is very popular as a value case fan. The Yate Loon D12SM-12 cost $6.99 retail and has specified output of 70.5cfm at 33db at 1650rpm.