Somedays are good, Somedays are bad, and then you have Monday....
by Gary Key on November 29, 2007 1:00 AM EST- Posted in
- Gary's First Looks
We rarely discuss what actually occurs during the testing of a product. After all, those details hardly matter in the grand scheme of things. What matters is if the product meets or exceeds its specifications and more importantly, provides value to the consumer who purchases it. At times, we might express our frustrations when testing a product and even discuss a few problems we ran into along the way. Otherwise, it is business as usual on the publishing side and simply trying to tell it like it is when discussing a product’s merits, good or bad. We feel that most readers just want to know if the product works as advertised and not some editor whining about what they had to go through during testing.
We whine enough amongst ourselves and figure at some point during the day, you probably get a good dose of it as well, so we try our best to insulate it from our articles. However, as they say, times are a changing. Now, we have the ability through these blogs to discuss those items that do not make it into an article after our illustrious copy editor cuts it; provide quick first looks at a product, report exclusives and interesting news, or just plain give you a “tears in the beer” synopsis of today’s events in the computer world.
This particular editor ran into one of those perfect storms on Monday. After experiencing a particularly difficult day of testing, we normally just bite our lips, openly howl in an empty parts closet, throw Mr. Potato Head across the test room, and toss a couple back before Miller Time officially starts (illustrious editor – We do not condone this, nor has it actually ever happened, but we are looking for more Mr. Potato Heads). However, our favorite way to relieve stress is to go chase down the neighborhood cat (metaphorically speaking, editor – Or is it?) with the biggest vehicle we can find in the parking lot.
Even after checking off each item on the stress relief list, it was time to do something different, especially considering the cat was having no problem outrunning the Prius. That something was to take out a fair amount of frustration on a couple of favorite suppliers, not as a reviewer in private, but as an actual paying consumer of their products. That is an important fact as the cards in testing were not early production or first look samples; these cards were shrink-wrapped, on the shelf, and available for sale. The reasons for our high level of frustrations are not important, all right they really are, but a touch of guilt and remorse has set in so here is the abbreviated version of our Monday.
During testing of the 790FX and Phenom products (that is yet another story but one that cannot be printed on-line) , we decided to provide 3870 CrossFire numbers on the 790FX and compare them to 8800GT SLI on the 590SLI board with our retail Phenom 9600. At the same time, we were working on the AMD 690G/NV7050 Phenom comparisons with both on-board video and single card HD3870/NV8800GT results for those who like to game via AMD/NVIDIA. Of course, this means a lot of card switching, but considering just how limited the supply is of each video card, we were taking extra precautions with full ESD controls in place plus many prayers. Turns out, we could have just sprayed the boards with kerosene and lit a match to them since we were still going to have failures with or without ESD protection and hourly prayer services.
Within a few minutes of each other, we had a BFG 8800GT OC and HIS HD 3870 both fail. Both cards were in SLI and CrossFire configurations respectively when they met their maker in the great digital divide. We had already completed the single card and stock speed SLI/CF tests and were just doing a final run through on the cards while overclocked to ensure stability with the motherboards and drivers. We did not manually overclock these particular cards; we utilized the auto overclock capabilities of each driver set to determine the clock speeds. In fact the overclock results were only about 7% higher than stock. The 8800GT card failed while running SupCom at 1920x1200 and the HD 3870 card failed during Crysis at the same resolution. We administered emergency care immediately, but after a few minutes on life support, each card passed away. Now the fun starts.
We had purchased the BFG 8800GT at Best Buy and the HIS HD 3870 at Newegg. The first order of business was to return the 8800GT, get credit, or more importantly exchange it for a working product, as we did not have a backup. Why no backups, well these products are still experiencing shortage problems and the likes of ASUS, Gigabyte, and MSI are just now entering the market due to chipset allocations so our normal backup capability was gone. The other reason is this is not our first failure, nor second, or even third for that matter, but Anand will explain that side-story shortly. A quick trip to the local Best Buy netted a credit only situation as the shelves were bare, but a helpful employee told us another store was showing two units in stock. We hopped in the car and took off on a 30-minute jaunt across town. Sure enough, the store had two units, but they were both returns, for what reasons we do not know.
Striking out on that front, we turned our attention to the web. After searching all of the major sites for the BFG card, we arrived at two options. The first was to purchase a replacement card (8800GT OC2) at BFG for $339.99 plus $37.97 next day shipping for a total cost of $377.96, well above the $279.99 we originally paid for our 8800GT OC card (note – OC card back in stock at BFG for $299.99 plus shipping today). Since neither BFG card was in stock elsewhere, our second option was to find a similar model from one of the other launch partners.
We located two cards (up to four-today) from XFX and EVGA in stock at TigerDirect starting at $299.99 and going up to $329.99 with next day shipping costs of $49.79 to $50.79, resulting in a total cost of $349.78 or higher. Not only were the cards more expensive but the shipping cost was ridiculous. That was our starting point and we continued to scour the web for better pricing and availability. Our next stop was Dell and they had the Superclocked EVGA 8800GT for $317.99 with only $16.00 being required for next day delivery, great, until we add in the required tax for a total of $361.54. We had the shipping cost where we wanted it; just the initial cost and taxes exceeded TigerDirect’s total cost at this point.
We ended up at Buy.Com and found the standard XFX 8800GT in stock for a price of $289.99, about $40 more than originally planned at launch, but after adding in $34.96 for overnight shipping, we had a total cost of $324.95. Figuring that was about as good as it will get and needing to take a break to find Mr. Potato head’s left ear , we hit the change order button (they do not show you shipping costs up front) to input our correct credit card information and then hit place order. Well folks, through the magic of real-time inventory systems, we got the dreaded, item no longer in stock message. We tore off Mr. Potato Head’s right ear, nailed the trash basket with it, and continued our hunt.
Our next trip took us directly to EVGA’s web store. We found the 8800GT SSC in stock for $299.99, we were joyous, and a guaranteed overclock to 700MHz on the core was worth the extra $20 over our original card cost while matching the slower cards at TigerDirect. We placed the item into the shipping cart, selected our zip code to get shipping costs, and then we passed out for a few minutes. No, we had not reached Miller Time yet; it was the thought of paying at least $91.24 for next day delivery with AM service coming in at a relative bargain of $100.33. Come on folks, it is not as if EVGA is a small time operation and does not get preferred shipping rates from UPS. This is highway robbery for a lack of better words. Very disgusted, we tossed what was left of Mr. Potato Head out the window (think we finally nailed the cat without trying – editor - just kidding we hope, but Sprinkles is nowhere around today).
Our final stop was at MWAVE, a quick search netted the EVGA Superclocked 8800GT being in stock for $289.90 with next day delivery at a more reasonable $24.32 for a total cost of $314.22. We whipped out the credit card, finalized the order, and woke up Tuesday morning to the sounds and fumes of FedEx delivering our new product. With it in hand, we resumed testing until today. Not being ones to let a sleeping cat lie (we know it is “dog” but in honor of Sprinkles we changed it), we decided to manually overclock our cards and try for a couple of Phenom based 3DMark records. Yep, another failure, not going to mention which card yet as it still might come around after a cool down period, but right now, our display resembles a nice snowfall screensaver at stock settings.
We might have had bad luck or were just plain stupid with the 12% overclock today, but something is not right in our opinion. The base fan speeds are extremely low at idle (26%~29%) and then do not ramp properly as the temperatures increase during load. Playing around with fan speeds utilizing Riva Tuner or nTune can solve this problem, but should a user really have to do this? We think not. As such, we are looking at other potential problem areas with cards that might be heat related. Of course, our card was cooking eggs at a temperature of 107C before the black screen and reboot so who knows, well Anand has a few ideas. He has a great theory as to what is happening that he will present in an upcoming 8800GT article. We would like to state that we could have had a card fail from any of the launch partners. The reason being is the same company manufacturers the cards. After sorting and binning, the cards are relabeled for the likes of XFX, BFG, or EVGA. If you happen to get a bad card (and it does happen at times), do not slam a particular supplier as they all have an equal chance of getting a dud.
Overall, not a good experience from a shopping viewpoint, something we are sure most of you have experienced looking for the 8800GT. Besides all of the prices being significantly above the $200~$249 promises made at launch (something we understand based on supply and demand but still do not like), we were just floored with the shipping costs at TigerDirect and EVGA, the others were somewhat within an average range for most e-Tailors, which is still high. The good news is that since yesterday, several sites from Newegg to Frys.Com now have 8800GTs back in stock, although, still at a premium cost. Regardless, the 8800GT is still a great product, it just costs a little more now.
In the case of our HD 3870, well we could not find any in stock on Monday although an alert at Newegg showed the HIS card back in stock but at a price of $269.99. We had just completed the RMA procedure and dashed off to replace this card at a hefty $50 premium over the original $219.99 price. Unfortunately, we were not quick enough on the keyboard (or maybe reading the email) and the units sold out by the time we reached the checkout confirmation. We had a bit of good luck in this area and our friends at Gigabyte were able to ship out two of their just arrived 3870s to us. With an extra 3870 in hand now, we are going to try an experiment tonight to see if we had a fluke or if pushing the card (even a little) will create the failure again while under load.
While we have heard of problems in the forums with the 8800GT cards overheating and a few DOAs, it was not until we actually started testing under a variety of conditions that we figured out our luck is running short, or there might be fire brewing under that smoke rising from our systems. It might just be bad luck considering the loss of the 3870 card at the same time, but until we finish testing, it's hard to pinpoint an exact cause. What was equally disappointing was the shopping experience in trying to procure a card on an emergency basis at a reasonable cost. Well there you have it, actually it’s not as bad as what we remember after writing it (illustrious editor – ask Sprinkles if he agrees), but Mondays are sometimes like that.....
Facebook
Twitter
RSS
27 Comments
View All Comments
jebo - Monday, December 3, 2007 - link
Fascinating to read about behind the scenes work of a reviewer. Do it more often!Kougar - Saturday, December 1, 2007 - link
Probably adding a bit more insult to injury, Saturday morning and Newegg alone has stock on six different 8800GT cards and four of both HD 3870/3850 cards, although all still at infalted prices despite the just rumored HD 38070/3850 price cuts.Always interested in a bit of what goes on behind the scenes, and very interested in experiences such as this. Will be looking forward to that mentioned 8800GT article! Of course in the 8800GT's defense, Anandtech's own silently cooled Sparkle 8800GT review does make these card failures seem unusual.
ocyl - Friday, November 30, 2007 - link
A large group of users are reporting failures of various video cards (not just a specific model) at places such as http://forums.nvidia.com/index.php?act=ST&f=33...">http://forums.nvidia.com/index.php?act=ST&f=33... and http://forums.nvidia.com/index.php?act=ST&f=33...">http://forums.nvidia.com/index.php?act=ST&f=33.... For those who would like to flame me simply upon seeing this, please browse through the above threads first before you decide to do so, because we all can use less of it on the Internet and also because I am posting this not out of manufacturer-favoritism but the fact that I am one of the affected customers.kilkennycat - Friday, November 30, 2007 - link
I got an error-message from the webmaster on my first reply-posting attempt, so I assumed the posting died...Anandtech, you are very welcome to delete one of my duplicate replies.
kilkennycat - Friday, November 30, 2007 - link
I acquired a (single) eVGA8800GT (675MHz GPU/975MHz mem) a couple of days ago (from Newegg). Here was the process I followed immediately after installing the board with driver version 169.09 beta (currently recommended for Crysis).1. Stuffed a temperature probe up its a**. Er, I mean I inserted a temperature probe deep between the heatsink fins on the exhanust side of the GPU-chip. (I am fortunate in having one of the Antec P160 cases with the continuous dual front-panel temperature display -the other temperature probe is firmly attached to the heat-sink base of my CPU)
2. Fired up nTune (ver 5.05.54) to enable the GPU temperature/fan-speed sensing and adjustment display in the nVidia control panel.
3. Started 3D stress-testing of the card while Alt-Tabbing frequently back to the GPU-core temperature and fan-speed readings. Noticed to my great horror that the fan-speed stayed rigidly at 29% while the GPU core marched past 90 degrees. BTW, my P160 heat-sink probe registered an expected offset of about 12 degrees @ 78 degrees C.
I am a professional design engineer with considerable experience of digital silicon design and reliability testing. I am well aware of the normsal rule-of-thumb regarding thermal reliability of commercial silicon LSI, where every 10 degrees beyond 80 degrees core-temp potentially halves the part-lifetime.... regardless of other effects like propagation-delay increase with temp, usually leading to operational flakiness and random lock-ups. Presumably nVidia has designed their GPU silicon using maybe 120 - 130 degrees C timing models in their simulations, so prop-delay issues with the GPU may not be too much of a concern - but memory prop-delays with temp-increase may still be an issue.)
Anyway, having seen the GPU temp march past 90 degrees with the fan speed not moving one iota, I decided to take matters into my own hands and I switched over to Manual fan-control using the nTune-initiated nV Control panel features. At 67% fan-speed and full torture-testing, the GPU-core temperature did not exceed 80 degrees in my particular system.
However, nVidia's nTune is a VERY flaky piece of software with an abysmal history - BEWARE the Manual fan setting is NOT remembered after a power cycle.
My Conclusion: nVidia needs to provide tooling to all users of the 8800GT and having the following fan-control features for 3D operation:-
a. A user-settable GPU core-temperature at which the fan speed will start to ramp up from the default 29% when exceeded. The range of this setting should start at 60 degrees C. The range upper end would be at nVidia's discretion - they know the temperature limits of their design.
b. A user-settable GPU core temperature at which the fan speed reaches 70% - the 'whine' noise-threshold of the default 8800GT cooler. The range of this setting should start at 75 degrees C. The upper end of the range again at nVidia's discretion.
c. The ramp in fan-speed between "a" and "b" should be roughly linear and should continue beyond "b" with the same slope to 100% speed.
d. The above settings for 3D operation must be "permanently" stored and recalled automatically after a power-cycle...a la motherboard BIOS settings.
Footnote: Immediately after successfully testing the 8800GT, I registered the product with eVGA. To qualify for the life-time eVGA warranty, you must sign-up within 30 days. Also, this entitles you to the 90-day step-up exchange program. Considering the tone of the above article and the witnessed failures,the eVGA step-up program might be a very useful escape valve. The eVGA800GTS Mark 2 based on the G92 chip with all stream-processors working is due within 90 days. Maybe by that time nVidia will have figured out a decent fan-speed control algorithm with user-adjustable parameters and implemented it on the G92 8800GTS even if they win't (or can't) offer it for the existing 8800GT. The 8800GTS Mark 2 WILL run hotter than the current 8800GT.
kilkennycat - Friday, November 30, 2007 - link
I acquired a (single) eVGA8800GT (675MHz GPU/975MHz mem) a couple of days ago (from Newegg). Here was the process I followed immediately after installing the board with driver version 169.09 beta (currently recommended for Crysis).1. Stuffed a temperature probe up its a**. Er, I mean I inserted a temperature probe deep between the heatsink fins on the exhanust side of the GPU-chip. (I am fortunate in having one of the Antec P160 cases with the continuous dual front-panel temperature display -the other temperature probe is firmly attached to the heat-sink base of my CPU)
2. Fired up nTune (ver 5.05.54) to enable the GPU temperature/fan-speed sensing and adjustment display in the nVidia control panel.
3. Started 3D stress-testing of the card while Alt-Tabbing frequently back to the GPU-core temperature and fan-speed readings. Noticed to my great horror that the fan-speed stayed rigidly at 29% while the GPU core marched past 90 degrees. BTW, my P160 heat-sink probe registered an expected offset of about 12 degrees @ 78 degrees C.
I am a professional design engineer with considerable experience of digital silicon design and reliability testing. I am well aware of the normsal rule-of-thumb regarding thermal reliability of commercial silicon LSI, where every 10 degrees beyond 80 degrees core-temp potentially halves the part-lifetime.... regardless of other effects like propagation-delay increase with temp, usually leading to operational flakiness and random lock-ups. Presumably nVidia has designed their GPU silicon using maybe 120 - 130 degrees C timing models in their simulations, so prop-delay issues with the GPU may not be too much of a concern - but memory prop-delays with temp-increase may still be an issue.)
Anyway, having seen the GPU temp march past 90 degrees with the fan speed not moving one iota, I decided to take matters into my own hands and I switched over to Manual fan-control using the nTune-initiated nV Control panel features. At 67% fan-speed and full torture-testing, the GPU-core temperature did not exceed 80 degrees in my particular system.
However, nVidia's nTune is a VERY flaky piece of software with an abysmal history - BEWARE the Manual fan setting is NOT remembered after a power cycle.
My Conclusion: nVidia needs to provide tooling to all users of the 8800GT and having the following fan-control features for 3D operation:-
a. A user-settable GPU core-temperature at which the fan speed will start to ramp up from the default 29% when exceeded. The range of this setting should start at 60 degrees C. The range upper end would be at nVidia's discretion - they know the temperature limits of their design.
b. A user-settable GPU core temperature at which the fan speed reaches 70% - the 'whine' noise-threshold of the default 8800GT cooler. The range of this setting should start at 75 degrees C. The upper end of the range again at nVidia's discretion.
c. The ramp in fan-speed between "a" and "b" should be roughly linear and should continue beyond "b" with the same slope to 100% speed.
d. The above settings for 3D operation must be "permanently" stored and recalled automatically after a power-cycle...a la motherboard BIOS settings.
Footnote: Immediately after successfully testing the 8800GT, I registered the product with eVGA. To qualify for the life-time eVGA warranty, you must sign-up within 30 days. Also, this entitles you to the 90-day step-up exchange program. Considering the tone of the above article and the witnessed failures,the eVGA step-up program might be a very useful escape valve. The eVGA800GTS Mark 2 based on the G92 chip with all stream-processors working is due within 90 days. Maybe by that time nVidia will have figured out a decent fan-speed control algorithm with user-adjustable parameters and implemented it on the G92 8800GTS even if they win't (or can't) offer it for the existing 8800GT. The 8800GTS Mark 2 WILL run hotter than the current 8800GT.
JarredWalton - Friday, November 30, 2007 - link
While being able to manually control the fan speeds might allow users to avoid the GPU failures, more to the point is that this sort of thing should have never happened in the first place. I have had quite a few GPUs fail during my time at AnandTech (X1950 and 7800 GTX most recently, plus some flaky 7600 GT cards). The GPU BIOS seems to be set WAY too conservatively on most of these cards.My feeling is that fan speed should ramp quickly past a certain threshold - if a GPU gets above 70C and continues to increase in a rapid fashion, the fan speeds need to increase a lot sooner. We can't just have "60C = 20% fan, 65 = 30% fan, 70 = 40% fan, etc." because if a GPU transitions from 60 to 70C really fast the GPU is likely going to hit 90C before the fan is doing its job.
Really, the bottom line is that hardware shouldn't be failing under normal use, especially not within a few weeks. Low noise levels are great and all, but sometimes you need a fan to move a lot of air - and who cares about noise levels!
strikeback03 - Tuesday, December 4, 2007 - link
I have a 7600GT that is overly aggressive in turning the fan up. Scroll down a web page quickly and it kicks into screamer mode. Start a movie and it kicks into screamer mode. Start my favorite RAW converter and it takes off. I added a fan on the side of the case blowing on it, and while according to nTune the temps dropped 3-5 degrees, the fan behavior was unaffected. Unfortunately nTune does not read the fan speed correctly, so I can't even try to shut it up. Probably going to end up just buying a passive heat sink.kilkennycat - Friday, November 30, 2007 - link
Jarred, I agree entirely, but I think that the lack of (safe-limited) user-control of the fan speed AS WELL AS decent built-in limits and ramps in the BIOS is the way to go. All of my previous experience of nVidia boards and temperature-limits indicate that the nVidia engineers are not sufficiently CONSERVATIVE in the default fan-settings. I really do not want nVidia making the final decision on fan-speeds and temperature profiles. I would like to be able to make even more conservative settings (that stick after a power-cycle) should I so choose. Without relying on that crippled piece of software junk called nTune.kilkennycat - Friday, November 30, 2007 - link
oops, sorry, I meant the "presence of" instead of the "lack of" in the first line of my above reply.(er, DailyTech and Anandtech really need the edit-update facilities provide by "the Tech Report" for their user-comments )