By: David Hess (davidwhess.delete@this.gmail.com), August 14, 2019 5:30 pm
Room: Moderated Discussions
Gian-Carlo Pascutto (gcp.delete@this.sjeng.org) on August 14, 2019 12:43 am wrote:
> David Hess (davidwhess.delete@this.gmail.com) on August 13, 2019 1:51 pm wrote:
> > I cannot say that I have ever had that problem but for good reason. Operation of the CPU at the
> > thermal throttling temperature is not conducive to reliable operation so I have never allowed
> > it. If the peak temperature is above about 65C Tj, then the heat sink is not large enough.
>
> It's typical in every laptop ever to hit TJmax all the time at load, and modern AMD CPUs will
> boost till they either nearly hit the thermal limit (95C Tj) or their power limit, the latter
> of which can be increased with PBO. In this case it's going to be "rather difficult" to stay
> below 65C without water cooling and radiators. They'll be stable nevertheless.
It sure is but I haven't owned a laptop for years now because I found them to be unreliable.
> > As far as the noise from the CPU fan, unless it is a small
> > whiny one, its noise should be insignificant even
> > at high speed because it is naturally mechanically well isolated from the case which also shields it.
>
> When I said earlier it's easy to build a *somewhat* quiet system but hard to build a *really*
> quiet one, this is exactly what I had in mind. Even extremely quiet fans like a Noctua will
> end up being the loudest component if running at anything close to full speed.
I ran across these passively cooled cases recently:
https://www.hdplex.com/
> If you can't hear your CPU fan at full speed, it just means the rest of your system is loud.
Not always. The NoS system I built for evaluating Windows 10 not long ago started with no moving parts except the CPU fans. The Intel stock heatsink was completely unaccaptable and could not even maintain Tj below 90C. But I had a really big vertical Hyper-somethingorother heat pipe cooler where the fans never get past idle speed with the Tj held below 65C under worst case conditions. And that was silent with the case open.
I tried running it without the fans at all but the thin closely spaced fins intended for forced air cooling just could not cut it.
> > Even more often, they leave out temperature measurements. As above with CPUs but even
> > more so, my experience is that GPUs which operate above 65C Tj have poor reliability.
>
> I absolutely cannot second that observation. Note that most GPU have a fan curve that will put them around
> 70-80C under full load. If they are not reliable at that point I RMA them because they're broken.
Nvidia ended up defending against several lawsuits over unreliable GPUs after they raised their "standard" operating temperature.
What I suspect happened with both Nvidia and AMD and to a lessor extent Intel was that they decided that the old standard of 20 years was not appropriate for the consumer market. For Intel this led to consumer processors operating at slightly higher clocks than Xeon processors and for Nvidia and AMD, they just pushed it as far as possible for benchmarks while not having too many problems with warranty returns.
> As you can see, the idea that a modern GPU runs below 65C at full load is an illusion.
Some currently produced GPUs run below 65C but you have to look for detailed reviews to find them. Of course they are not the highest performance ones.
> David Hess (davidwhess.delete@this.gmail.com) on August 13, 2019 1:51 pm wrote:
> > I cannot say that I have ever had that problem but for good reason. Operation of the CPU at the
> > thermal throttling temperature is not conducive to reliable operation so I have never allowed
> > it. If the peak temperature is above about 65C Tj, then the heat sink is not large enough.
>
> It's typical in every laptop ever to hit TJmax all the time at load, and modern AMD CPUs will
> boost till they either nearly hit the thermal limit (95C Tj) or their power limit, the latter
> of which can be increased with PBO. In this case it's going to be "rather difficult" to stay
> below 65C without water cooling and radiators. They'll be stable nevertheless.
It sure is but I haven't owned a laptop for years now because I found them to be unreliable.
> > As far as the noise from the CPU fan, unless it is a small
> > whiny one, its noise should be insignificant even
> > at high speed because it is naturally mechanically well isolated from the case which also shields it.
>
> When I said earlier it's easy to build a *somewhat* quiet system but hard to build a *really*
> quiet one, this is exactly what I had in mind. Even extremely quiet fans like a Noctua will
> end up being the loudest component if running at anything close to full speed.
I ran across these passively cooled cases recently:
https://www.hdplex.com/
> If you can't hear your CPU fan at full speed, it just means the rest of your system is loud.
Not always. The NoS system I built for evaluating Windows 10 not long ago started with no moving parts except the CPU fans. The Intel stock heatsink was completely unaccaptable and could not even maintain Tj below 90C. But I had a really big vertical Hyper-somethingorother heat pipe cooler where the fans never get past idle speed with the Tj held below 65C under worst case conditions. And that was silent with the case open.
I tried running it without the fans at all but the thin closely spaced fins intended for forced air cooling just could not cut it.
> > Even more often, they leave out temperature measurements. As above with CPUs but even
> > more so, my experience is that GPUs which operate above 65C Tj have poor reliability.
>
> I absolutely cannot second that observation. Note that most GPU have a fan curve that will put them around
> 70-80C under full load. If they are not reliable at that point I RMA them because they're broken.
Nvidia ended up defending against several lawsuits over unreliable GPUs after they raised their "standard" operating temperature.
What I suspect happened with both Nvidia and AMD and to a lessor extent Intel was that they decided that the old standard of 20 years was not appropriate for the consumer market. For Intel this led to consumer processors operating at slightly higher clocks than Xeon processors and for Nvidia and AMD, they just pushed it as far as possible for benchmarks while not having too many problems with warranty returns.
> As you can see, the idea that a modern GPU runs below 65C at full load is an illusion.
Some currently produced GPUs run below 65C but you have to look for detailed reviews to find them. Of course they are not the highest performance ones.
