Titan X Launch

[Silverforce11]

Yeah I hope not. Don't want to create a lot of hype panic where there shouldn't be. Right?

Well, if you spend $1000 or more (SLI users) and the RAM is running well above manufacturer specs for temperatures, should you be concerned?

Should NV have taken better care to ensure their premium product has better quality control?

There's no hype panic but these are serious concerns, we all know running ICs above their temperature spec shortens lifespan. Maybe ask your connections at NV for us, perhaps they have fancy new GDDR5 with higher temperature specs that Hynix isn't aware of? If so, tell Hynix to update their data sheet.

 

[Silverforce11]

If the rear of the card behind the memory chip can reach 120C. What do you think the memory chip is at? 80C? The PCB isnt exactly known for its excelent heat transfer.

GDDR3 was a typo, was to be GDDR5.

PCBs are an insulator, it isn't an excellent thermal or electrical conductor. So yes, you are right, it doesn't transfer heat very well.

 

[ShintaiDK]

The memory already got an internal sensor as well. For Hynix its tTSEN.

5.22. TEMPERATURE SENSOR​

GDDR5 SGRAMs incorporate a temperature sensor with digital temperature readout function. This function​

allows the controller to monitor the GDDR5 SGRAM die’s junction temperature and use this information​

to make sure the device is operated within the specified temperature range or to adjust interface​

timings relative to temperature changes over time.​

The temperature sensor is enabled by bit A6 in Mode Register 7 (MR7). In this case the temperature readout​

is valid after​

tTSEN. Hynix applies 10us to tTSEN.​

The temperature readout uses the DRAM Info mode feature. The digital value is driven asynchronously​

on the DQ bus following the MRS command to Mode Register 3 (MR3) that sets bit A7 to 1 and bit A6 to 0.​

The temperature readout will be continuously driven until an MRS command sets both bits to 0.

 

[ShintaiDK]

PCBs are an insulator, it isn't an excellent thermal or electrical conductor. So yes, you are right, it doesn't transfer heat very well.

And by that we can conclude, if the PCB is 120C on the back, the memory chip behind the PCB must be 120C+.

 

[Silverforce11]

And by that we can conclude, if the PCB is 120C on the back, the memory chip behind the PCB must be 120C+.

Depends whats located on that rear PCB, are they resistors, small caps? :/

Clearly from looking at actual FLIR readings from review sites directly on the ram, GDDR5 chips have been running very cool from AMD/NV GPUs for many generations already.

 

[Keysplayr]

Well, if you spend $1000 or more (SLI users) and the RAM is running well above manufacturer specs for temperatures, should you be concerned?

You should be concerned IF and WHEN something happens. Attributing to this newly found hype train is pretty irresponsible.

Should NV have taken better care to ensure their premium product has better quality control?

Should they have? Did something happen? Why don't you save this hype until after something happens. If ever? Or we can wait for someone to hold a lighter on one of the RAM chips and melt it, then post pictures of it saying, "This is what happened to my Titan X memory in my well ventilated case with 17 fans blowing through it!!"

There's no hype panic but these are serious concerns, we all know running ICs above their temperature spec shortens lifespan. Maybe ask your connections at NV for us, perhaps they have fancy new GDDR5 with higher temperature specs that Hynix isn't aware of? If so, tell Hynix to update their data sheet.

The "concern" is only serious because that is what you are hyping it up to be. Serious. You are attributing to FUD. You should probably stop doing that.

 

[ShintaiDK]

Depends whats located on that rear PCB, are they resistors, small caps? :/

Clearly from looking at actual FLIR readings from review sites directly on the ram, GDDR5 chips have been running very cool from AMD/NV GPUs for many generations already.

Are you saying the images I posted are wrong? Its quite clear that VRAM have been running well over 100C for many years.

GDDR memory also constain its own sensor. How calibrated it is is anyones guess. But thats what they refer to.

 

[Silverforce11]

Are you saying the images I posted are wrong? Its quite clear that VRAM have been running well over 100C for many years.

Nope, not enough info from pics you posted since you lack direct ram temp measurements, you are relying on PCB temp which would be affected by many things including GPU core temp, VRM temps, resistors, capacitors, chokes and ram. While the PCB is a great insulator, given enough time, heat will spread over silicon.

So if you want to claim that GDDR5 has been running over 100C for many years, you'll need better proof. What proof we have, from FLIR measurements from various sites is that over the years, direct GDDR5 module temps have never been that high that I can recall.

 

[Silverforce11]

The "concern" is only serious because that is what you are hyping it up to be. Serious. You are attributing to FUD. You should probably stop doing that.

FUD would be me saying willy nilly that 85C is the rated spec without proof... but nope, its direct from Hynix.

You are saying running ICs well above temp spec is not serious? It's only serious when mass casualties of Titan X show up later on, preferably after the warranty expires? Okay..

Was bumpgate serious to you?

 

[Spanners]

And by that we can conclude, if the PCB is 120C on the back, the memory chip behind the PCB must be 120C+.

I wouldn't necessarily agree with that, I don't want to guess but if the BGA package is connected to some kind of heat-sink (like in the HD4870 example) then I don't see why the package temps wouldn't be much lower. The GPU itself, with a much larger output (and better heatsink) isn't 120C+ because it's on the other side of a hot PCB.

 

[Keysplayr]

FUD would be me saying willy nilly that 85C is the rated spec without proof... but nope, its direct from Hynix.

You are saying running ICs well above temp spec is not serious? It's only serious when mass casualties of Titan X show up later on, preferably after the warranty expires? Okay..

Was bumpgate serious to you?

What I am saying is, you are perpetuating FEAR, UNCERTAINTY and DOUBT.
Before anything ever happens. Might as well tell everyone you shouldn't fly in an airplane. It might crash. Just stop man.

 

[Keysplayr]

I'd also like to see the calibration records of each of these sites infra-red and FLIR devices.
I also have a non-contact laser thermometer around here somewhere. Have to look for it.

 

[Silverforce11]

What I am saying is, you are perpetuating FEAR, UNCERTAINTY and DOUBT.
Before anything ever happens. Might as well tell everyone you shouldn't fly in an airplane. It might crash. Just stop man.

I was actually seriously considering folking major $ for a Titan X, mainly due to future GameWorks title I want to play. This was the last straw. Edit: I mean GameWorks is why I wanted to switch to NV, paying a premium.

In your silly analogy, if said plane was running too many active hours between safe maintenance spec, or if its engine overheat above safe recommended spec... yes, I would certainly not fly on it.

 

[Keysplayr]

I was actually seriously considering folking major $ for a Titan X, mainly due to future GameWorks title I want to play. This was the last straw.

In your silly analogy, if said plane was running too many active hours between safe maintenance spec, or if its engine overheat above safe recommended spec... yes, I would certainly not fly on it.

Most of the planes up there, are probably doing just that.

So now you'll tell everyone who wanted to buy a Titan X, not to, based on the temperature of the memory chips. Amiright? Is that next?
Based on your own, "last straw" experience?

 

[MrK6]

NVIDIA cards are known for their shoddy/cut corner designs, and it's gotten worse the past years. Mobile GPU failures, GTX 480 failures secondary to heat, GTX 570 power regulation, GTX 970 memory configuration, I could go on. However, I would at least except a part they're asking $1,000 for to have a halfway decent cooler on it, which they seemed to have failed miserably. What benefits nvidia is that only a few will care in the year or two it takes for design failures to surface. I think they make throwaway GPU's, use power and clock limiting to try to mask it, and still try to charge a proper price for them.

 

[therealnickdanger]

I'd also like to see the calibration records of each of these sites infra-red and FLIR devices.

This. I don't need to see the actual logs, I'd settle with them even acknowledging that they know how to do it and that they do it regularly.

It's one thing to spend a lot of money on an expensive piece of lab equipment. It's entirely another to ensure that it works properly and is calibrated. I work with laser and radar survey equipment on a daily basis and we have to run calibration tests before each day's use. The technologies have gotten to the point where they are usually OK, but adjustments need to be made every couple weeks.

That said, it's unlikely that their equipment is off by more than 3%, so we're still talking about some high temperatures here. I think NVIDIA should strongly embrace CLC (or at least a WF design) going forward. Throttling during normal use on a $1,000 GPU is NEVER acceptable.

 

[wand3r3r]

The "concern" is only serious because that is what you are hyping it up to be. Serious. You are attributing to FUD. You should probably stop doing that.

What I am saying is, you are perpetuating FEAR, UNCERTAINTY and DOUBT.
Before anything ever happens. Might as well tell everyone you shouldn't fly in an airplane. It might crash. Just stop man.

Quoting datasheets which show the maximum operating temperatures that VRAM is spec'd to run at is FUD now?

Considering that it does appear that the titan x vram is 20C hotter than is specified to run at it is quite alarming, unless there is proof otherwise.

You can't stop people from noticing potentially fatal flaws, it would be more productive to prove that they are within spec.

 

[n0x1ous]

GTX 480 failures secondary to heat

Hadn't heard this one. I have a reference launch day 480 that's been running for what 5 years now....gets up to low 90's gaming......never had an issue other than burning my finger touching the front of the card after a session

 

[RussianSensation]

NVIDIA cards are known for their shoddy/cut corner designs, and it's gotten worse the past years. Mobile GPU failures, GTX 480 failures secondary to heat, GTX 570 power regulation, GTX 970 memory configuration, I could go on. However, I would at least except a part they're asking $1,000 for to have a halfway decent cooler on it, which they seemed to have failed miserably. What benefits nvidia is that only a few will care in the year or two it takes for design failures to surface. I think they make throwaway GPU's, use power and clock limiting to try to mask it, and still try to charge a proper price for them.

IMHO, it feels like NV has not put 100% effort into the Titan X and are holding something back. They are waiting for AMD to launch R9 390X so that they can work with AIBs to launch faster clocked GM200. What are the chances we will not see MSI Lightning, Asus Matrix, EVGA Classified GM200 cards this generation as we are probably 18-24 months away from Big Pascal?

NV is repeating the Kepler strategy, except this time the Titan X is the full GM200. However, in terms of hierarchy I expect there will be a faster GM200 card at some point before Pascal launches as I don't believe NV will let AMD have the performance crown.

NV also will need to figure out what to do with GM200 chips that are not fully yielding. I bet even if you take a slightly cut down GM200 (2816 CUDA cores over 22 SMs), give it the MSI Lightning treatment and OCed it should outperform Titan X air OCed.

Also, the reason I criticized the original Titan's $999 pricing was that to me a $1K single-GPU card should have the single GPU performance crown for the entirety of that generation. The Titan didn't as 780Ti and 290X beat it. Can NV guarantee that the Titan X will be the fastest single chip GPU until Pascal/14nm GPUs launch in 2016? I have my doubts. At least the original Titan had DP performance which made it somewhat of a prosumer card. With that now being taken away, the Titan X feels like a 980TI/1080GTX with a $300 pricing premium for 12GB of VRAM....

This card will still sell to early adopters, the benchmarking crowd and researchers interested in SP CUDA performance for say deep learning. However, for gamers I feel the Titan X hasn't been able to capture the original Titan magic. Even the stock performance of Titan X over 980 is less than the original Titan had over a 680:

43% faster than a 680

vs.

30% faster than a 980

Finally, what makes it worse is that the original Titan was fairly close to a $1000 GTX690 and $1000 HD7990 in performance. Today an R9 295X2 is $660 on Newegg and is 20% faster than the Titan X at 4K making the situation far more difficult than when choosing $1000 7990/690 vs. $1000 Titan. Without DP performance, it's difficult to say with as much confidence that the Titan X will retain its value as well as the original.

Therefore, in relative terms while the Titan X costs similar to the original Titan, its performance leap over NV's 2nd fastest card (980) is less and its relative price/performance standing against the fastest card (R9 295X2) is FAR worse. If NV priced this card at $700, I bet a lot of gamers wouldn't even bother waiting for 980Ti/1080GTX or a 390X. Right now it seems like a straight up Brand Value/marketing play and a money grab. Shockingly, the original Titan actually looks better in comparison to its historical GPU market vs. how Titan X looks today. Ironically the original Titan's price was heavily criticized but it seems there is not as much backlash for this one....

 

[Silverforce11]

@RussianSensation
There's no need to criticize price or perf/$. That does not apply to NV. People either choose to pay extra for NV or they don't. It's clear in AT's article, NV is pushing their top tier GPUs as premium luxury items. As such, mundane metrics like value or perf/$ don't apply.

I mean, do you go around criticizing why a Patek Phillip watch is $20K or more? Or $30K hand-bags? Nope, luxury items are excluded from rational & logical decision making.

 

[SlowSpyder]

For $1000 it should have had a better cooler. When AMD released the R9 290X, it was cheaper than Titan and the GTX 780, but people still complained about the reference cooler (but at least with the AMD cards non-reference designs were coming, with Titan that may not be the case). But, at this price I'm not sure most people buying one wouldn't just buy an aftermarket cooler anyway. I know I'd have mine under water if I owned one.

 

[SexyK]

Well the card should be here Saturday and the water block is right behind it. Looking forward to seeing what it can do under water. Now just debating whether I should upgrade the whole rig to Haswell-E while I'm tearing down the loop anyway. Probably not worth it but some of the platform upgrades are quite nice.

 

[n0x1ous]

(but at least with the AMD cards non-reference designs were coming, with Titan that may not be the case).

Looks like they are coming

http://www.inno3d.com/products_detail.php?refid=192

 

[SlowSpyder]

Looks like they are coming

http://www.inno3d.com/products_detail.php?refid=192

That's a good move on Nvidia's part in my opinion. Open it up, let the board partners push this thing and build them with a proper cooler. :thumbsup:

 

[Headfoot]

Opening it up to aftermarket coolers is a good move, but they need to unlock voltage and power capping too so people can actually push the chips with the new cooling