That’s no point really discussing about general concepts in semiconductors, and applying that to criticise a very specific technology and process node. Because in-between, you miss and don’t understand lots of things. Semiconductor manufacturing, and reliability modelling of silicon are very specialised field. I hope you realise that at least, and let’s stop right here.
ASIC physical designer here, worked with all major foundries and latest nodes. Electromigration is doubling every 10C. 105C to 125C would mean reducing 24/7 operation from 10 years (industry non automotive standard) to 2.5 years. Also the Tj is at the junction, not at the package level. There are lots of consideration for local hotspots too. Tjmax 95C does not mean package at 95C.
And yes, passing EM at a specific temperature with a 10 years projected life is a foundry requirement for manufacturing.
6 Likes
You demanded I respond with proof and then suspended my account for doing so. That’s literally entrapment. Why on Earth would I want to be part of a community with a mod so toxic as to do this, all because they don’t want people talking about their negative experiences of their favourite billion dollar company? It’s childish – particularly now that the reviews are out and have vindicated my prior scepticism. The boosts are good, but the efficiency claims are an out and out lie. The power required to get the clocks as high as they are are greater than the performance gains (95°C is Now Normal: AMD Ryzen 9 7950X CPU Review & Benchmarks - YouTube). AMD have claimed that efficiency is greater than the 5000 series, when it is objectively, measurably not. AMD could have just focused on the level of performance or just compared strictly against Intel. They didn’t need to fabricate claims like this yet again.
1 Like
The efficiency is better than 5000 series if you run the r9 7000 at the same power level as the r9 5000 series. (because r9 5000 was gimped to the same power as 5800X)
Now they let the r9 7000 get a little room to breathe so performance is phenomenal - but of course vf curves sucks for efficiency on the top.
Best way to to get efficiency is to limit power, the more you limit the more efficient - but you’ll tank actual performance.
1 Like
Oof that’s two and a quarter hours that I don’t really have. Do you have a tl;dr?
The problem is you can make that argument with anything. Vega 64 is an extremely efficient GPU if you underclock it, undervolt it and only expect midrange performance out of it. You can get a couple of hundred watts out of a 5950X if you clock the snot out of it and have some extreme cooling. If you overclocked a 5950X, let it run pegged at 95 degrees on a triple radiator it’s going to be really inefficient – possibly moreso than the 7950X. I think the difference is that the 7950X runs like this at stock which is where AMD’s comparisons have got to be because that’s how these CPUs are being delivered to the consumer. If we’re going to call out Nvidia’s GPUs for being clocked way above the sweet spot of their efficiency curve, we can’t just let AMD say “well if we didn’t do that, it’d be really efficient trust us!”
if I had this CPU I’d probably want to set a thermal limit of 80 degrees, partly for efficiency reasons, but partly because I’m really not comfortable running a CPU that hot 24/7. Are they planning to make Ryzen Master available for Linux? For a company that people consider the darling of Linux users, there’s a lot of functionality that they just don’t make available for us. Not that Intel or Nvidia do either but they don’t really offer things like Ryzen Master on Windows either.
My previous responses in this thread should be clear that I didn’t attempt to challenge this physics. I don’t see your point repeating it to me. Or are you trying to say at 95C, the lifespan is extended to 20 years. Apparently temperature is only one dimension in affecting reliability.
Yes, individuals with EE background can tell.
Of course. Details matter here though.
A repeat of your previous point just few sentences before.
I believe I mentioned something similar before as well though not with specifics.
Look. I’m flattered to receive your response. Though I’m a bit puzzled why you picked me because I’m not the one challenging TSMC/AMD’s engineering quality in ensuring reliability of their silicon. Rather I was challenging another member who was baby-crying about TjMax 95C and implying chips to die sooner and quick.
With your hands-on industrial experience, perhaps you could shed more light what sorts of modelling is taken by TSMC/AMD to ensure TjMax 95C is fine for 24/7 operation with reasonable reliability. Also perhaps some details during silicon operations measures taken to ensure reliable and long-lasting lifespan. So that for doubters in this thread they could receive more convincing and educational information from an insider with know-how, and stop baby-crying.
Apart from tuning CO I don’t know anything that Ryzen Master can do that you cant do in bios.
Alright. I’m back folks.
Holy shit, what happened? 
Err, so… 4.5GHz on all cores it is? Or is the Base Frequency a lie from AMD? 
It seems like there was a lot of conjecture.
Ok. I just watched GNs review. And… i’m sorry, how many fucking degrees Celsius? 
This makes the RV770 GPUs from 2008 look like a mini LED. Jesus H Born-In-a-manger, read the bible christ…
Yeah, err. I don’t know. It sounds impressive, but it seems like a repeat of RV770 and Bulldozer in terms of thermal performance. Holy jumping fucking shitballs.
I remember hearing somewhere that silicon’s LIMIT is somewhere around the 100C mark? Coincidentally the point at which water boils. If they can reach these frequencies within 10 years, and we see the power jump down like when we first saw 45nm from Intel in the form of Nehalem, i’ll be impressed. No matter what kind of machine you’re talking about, I always have an intuitive sense that simply doesn’t like for a machine to struggle. In all seriousness. 
I can’t help myself.
I could NEVER own one of these if I knew it was running like this. I felt bad for my HD 4850 back in the day… temperatures in the 70s were crazy back then! Woah! And fast forward to today?
95 degrees? “Oh, yeah, yeah. That’s fine.”
Again, I’m really, really curious to see how Intel handle Nova Lake. Really curious about that Royal Cove - I don’t know the flippin names anymore 
- architecture and how it’ll perform. Both in the thermals and compute performance overall. I really hope Intel pull another Nehalem. I know it’s asking much, but we’ve really slowed down since the mid-late 2000s in CPU architecture and probably moreso in IPC increase. AMD have gained much momentum with Ryzen. But this is… a little scary.
So which company is going to come out w a machined heat plate that keys perfectly on top of the spider?
This would help lining up heatsinks vertically mounting
No need, this time it’s flat. https://twitter.com/Optimus_WC/status/1577785460335575040?s=20&t=gzk2d3CkWbhibjcw9TyEjg
1 Like
This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.