'AMD vs Intel: Which CPU Cooks Better Pancakes?' - Tom's

[Fanatical Meat]

I usually spread his posts over the fields. Really helps bring the grass on.

None of the Zen CPUs "shut off" at 70 deg. They throttle.

When (like Mark), I had an Enermax failure*, the CPU just throttled clocks down to ~500 MHz. No shut off, no damage, no problem. Replace the broken cooler and away you go back to 16 Cores of 4.2 GHz goodness.


*mk2 Enermax working away very nicely bar a rattly for anyone looking at TR3 coolers

Doesn’t like every processor either shut down or throttle in the catastrophic fan/cooler failure scenario for like the last 12 years?

 

[Markfw]

I usually spread his posts over the fields. Really helps bring the grass on.

None of the Zen CPUs "shut off" at 70 deg. They throttle.

When (like Mark), I had an Enermax failure*, the CPU just throttled clocks down to ~500 MHz. No shut off, no damage, no problem. Replace the broken cooler and away you go back to 16 Cores of 4.2 GHz goodness.


*mk2 Enermax working away very nicely bar a rattly for anyone looking at TR3 coolers

Except, I don;t think they throttle until about 85 to 90. I think I had my 2990wx up to 90 before I upgraded to GOOD water.

 

[Borealis7]

(disclaimer: i am not an electrical engineer, this is just my opinion/feeling)
I feel like ever since Haswell (or even the 3770K) we have been taught by Intel to live with high load temperatures on our CPUs even with the beefiest air cooling blocks.
There is no getting around it, as Homer Simpson once said: "young lady, in this house we obey the laws of physics."

As lithography processes become smaller and denser, and dies become (sometimes) smaller and packed with more transistors (i.e transistor density rises), even though operating voltage may decrease a little, the amount of heat per cm^2 rises and CPUs become hotter.

It is not economically viable for Intel or AMD to decrease transistor density because they "get paid by the mm^2" and it could also decrease yield. So I fear we might end up in a situation where CPUs would produce so much heat that conventional air cooling just wont cut it anymore for operating under load. how much more can we milk Silicone before heat forces Intel or AMD to ship CPUs with AIO water coolers in package for mainstream models?

 

[Charlie22911]

My understanding is that Intel CPUs not only throttle at their thermal limits, but also raise interrupts that lower the duty cycle of the execution pipeline and inform the host OS of the condition. I can’t seem to find the relevant article, so take with the appropriate amount of salt. Differences in throttling mechanisms explain the difference in uptime though.

Intel’s approach is more graceful, and probably a carryover from the enterprise world where failover is important. AMD seems to just shutdown, and as far as I know it’s been this way since the P4/Athlon gen 1 days.

Edit:
Not the article I was thinking of, but directly from Intel:

Developer Software Forums

software.intel.com

Edit 2:
For those interested in reading the manual, the relevant section is 8.7.13.3 (PG 2988) and 14.7 (PG 3195):

https://software.intel.com/sites/default/files/managed/39/c5/325462-sdm-vol-1-2abcd-3abcd.pdf

 

[Markfw]

My understanding is that Intel CPUs not only throttle at their thermal limits, but also raise interrupts that lower the duty cycle of the execution pipeline and inform the host OS of the condition. I can’t seem to find the relevant article, so take with the appropriate amount of salt. Differences in throttling mechanisms explain the difference in uptime though.

Intel’s approach is more graceful, and probably a carryover from the enterprise world where failover is important. AMD seems to just shutdown, and as far as I know it’s been this way since the P4/Athlon gen 1 days.

Edit:
Not the article I was thinking of, but directly from Intel:

Developer Software Forums

software.intel.com

Edit 2:
For those interested in reading the manual, the relevant section is 8.7.13.3 (PG 2988) and 14.7 (PG 3195):

https://software.intel.com/sites/default/files/managed/39/c5/325462-sdm-vol-1-2abcd-3abcd.pdf

Nope. The 1950x you sold me as an example, just throttles down to 500 mhz when its not cooled correctly (the aio died)

 

[Zucker2k]

Even then. It doesn't matter what the upper limit is as long as you can use the rated cooler to keep it within spec easily. There was a time 70c seemed crazy, then a certain company started using tim with terrible contact and terrible properties and they had to adjust max temps to cover up the crappy thermal transfers.

Hot not hot, likes high temps doesn't like high temps. Doesn't matter as long as A.) It has protection for what its limits are hit B.) That it can reasonably stay within those limits with coolers rated within its TDP.

Didn't AMD introduce a -15 degree offset on certain Zen processors? That has to be one of the greatest con jobs ever.

The CON chips had a temp limit of 72C I think. Matisse can run all the way up to 95C. No idea how @VirtuaLarry got his 3600 to run @ 117C but he did it.

Never happened, unless we're counting catastrophic software readout failures. Oh, by the way, was this while he was running blender?

 

[Topweasel]

Only if you don't actually understand what the offset was and which direction. Increased reported temperature on 95w CPU's to get the fan to spool up earlier allowing for larger headroom for longer turbos during boosting with rated coolers.

It has nothing to do with running within and staying within spec. Which it could do handedly with or without the offset.

But let's ignore Intel actually including an offset on core clocks to hide the fact that their cpus got too hot when dealing with certain workloads to actually run at rated speeds. That seems like a larger con job.

 

[Zucker2k]

Only if you don't actually understand what the offset was and which direction. Increased reported temperature on 95w CPU's to get the fan to spool up earlier allowing for larger headroom for longer turbos during boosting with rated coolers.

It has nothing to do with running within and staying within spec. Which it could do handedly with or without the offset.

But let's ignore Intel actually including an offset on core clocks to hide the fact that their cpus got too hot when dealing with certain workloads to actually run at rated speeds. That seems like a larger con job.

Answer this: What's the throttling point (temp) of the Ryzen 1800x?

 

[Markfw]

Answer this: What's the throttling point (temp) of the Ryzen 1800x?

Mine did not thrreottle until about 90-95c.

 

[Topweasel]

Answer this: What's the throttling point (temp) of the Ryzen 1800x?

95c reported, 75c in actuality.

 

[Markfw]

95c reported, 75c in actuality.

Where did you see that 75c number ? Some other user ? I don't know how mine could be that different.

 

[Zucker2k]

95c reported, 75c in actuality.

LOL! I don't want to sound like a conspiracy theorist, but won't that be a serious retrogression for silicon in this day and age to be throttling at 75 degrees on the desktop? What kind of heat sensitive process was Glofo's 14nm? The 1800x can pull 180+ watts and you're telling me it throttles at 75c? Somehow the math and physics doesn't add up.

 

[Topweasel]

Where did you see that 75c number ? Some other user ? I don't know how mine could be that different.

It isn't it's the offset. It reports 20c higher to get the most fan curves to spin up the fans earlier to stretch out the XFR boost time.

 

[Topweasel]

LOL! I don't want to sound like a conspiracy theorist, but won't that be a serious retrogression for silicon in this day and age to be throttling at 75 degrees on the desktop? What kind of heat sensitive process was Glofo's 14nm? The 1800x can pull 180+ watts and you're telling me it throttles at 75c? Somehow the math and physics doesn't add up.

It's all about how well power is fed to the cooler. It A.) It never ever ever used 180w with any normal settings. Complete fud. People need to stop using system power as CPU power to stretch their point. B.) Ryzen was pretty large by consumer die sizes and used a really transfer material and was easy to cool. C.) The CPU generally ran out of silicon speed before it was an issue. D.) I7s had nearly the same limit till they changed Tim and started testing against higher temps because of the weak transfer material.

 

[Markfw]

It isn't it's the offset. It reports 20c higher to get the most fan curves to spin up the fans earlier to stretch out the XFR boost time.

The reason is that I don't use fan curves, they are all set to 100%. And Ryzen master was correctly showing temps. Before I got good cooling it would throttle at 95c.

Don't mislead others to think they really throttled at 75c.

 

[DrMrLordX]

Didn't AMD introduce a -15 degree offset on certain Zen processors? That has to be one of the greatest con jobs ever.

Con job, huh? Yeah I lost exactly $0 on that one. Anyway it was for fan curves, it didn't actually throttle or shut down the CPU, and you could disable the offset eventually. I did and I got my CPU up over 79C without throttling. I take it you're fishing for something negative about AMD here? It isn't working.

Never happened, unless we're counting catastrophic software readout failures.

Nice of you to establish yourself as grand arbiter of all that is or isn't real. Ever heard of hot spot temps? That's what Matisse reports as its main temp. 117C is entirely possible.

And no, it wasn't Blender. It was a particular PrimeGrid workload, for which @VirtualLarry created an entire thread. You probably missed that one.

Answer this: What's the throttling point (temp) of the Ryzen 1800x?

Something higher than 79C, which is as hot as I ever got mine.

 

[Topweasel]

The reason is that I don't use fan curves, they are all set to 100%. And Ryzen master was correctly showing temps. Before I got good cooling it would throttle at 95c.

Don't mislead others to think they really throttled at 75c.

If you got it to read at higher then 75 without throttling you weren't seeing the correct temp.

"The maximum warrantied voltages are programmed into the CPU when it is made. The maximum temperature for Ryzen 7 1700 is 75C. "

The 1700 and 1700x and 1800x all had the upper temp limit of 75c. Anything higher was it reporting with the offset.

 

[.vodka]

Didn't AMD introduce a -15 degree offset on certain Zen processors? That has to be one of the greatest con jobs ever.

The first gen X Ryzens had a +20°C offset to the reported temperature for fan control. Proper monitoring software (hwinfo) could/can read the actual temperature.

Naturally this was dropped on 2nd and 3rd gen parts with their much better boost algorithms.

 

[VirtualLarry]

Ever heard of hot spot temps? That's what Matisse reports as its main temp. 117C is entirely possible.

And no, it wasn't Blender. It was a particular PrimeGrid workload, for which @VirtualLarry created an entire thread. You probably missed that one.

Here's what I still don't get. Ryzen Master, most current version at the time for the 3000-series CPUs, was reporting, up to 117C. That's with a fixed-clock, fixed-voltage OC, which AFAIK, disables all of the Cool 'N Quiet and most of the protection/throttling stuff.

That said, it was also on 240mm AIO WC, and if the water was truely above boiling point, I would have expected to hear something different going on with the WC kit, and I didn't.

So, because of that, I surmised that the temps weren't quite real, and that there was an offset.

Then people said, nope, no offset on 3000-series, Ryzen Master's temps are the real deal.

So that's when I said, "Oh ****", better stop this fixed-clock / fixed-voltage experiment right now.

The wierd part is, the last two days, my system has just simply shut down, Reliability Monitor didn't report "Hardware error", but just "System had unexpected shutdown".

I don't know if I've been hacked, or something in my hardware is unstable, or what.

My CPU temps (no longer overclocked, stock for like a month or more), don't exceed 82C, as measure by HWMonitor (and has always jived with Ryzen Master's temps too).

 

[VirtualLarry]

The first gen X Ryzens had a +20°C offset to the reported temperature for fan control. Proper monitoring software (hwinfo) could/can read the actual temperature.

Naturally this was dropped on 2nd and 3rd gen parts with their much better boost algorithms.

That's what I was (finally) led to believe, that there was no "fan control offset" on 3000-series Ryzen CPUs. At least, I don't know, mines a non-X, and even on 1st-gen, it supposedly only applied to the 'X' models, so my chip would have had the offset even if they continued to do that.

 

[DrMrLordX]

Here's what I still don't get. Ryzen Master, most current version at the time for the 3000-series CPUs, was reporting, up to 117C. That's with a fixed-clock, fixed-voltage OC, which AFAIK, disables all of the Cool 'N Quiet and most of the protection/throttling stuff.

That said, it was also on 240mm AIO WC, and if the water was truely above boiling point, I would have expected to hear something different going on with the WC kit, and I didn't.

It's reporting a hotspot temp. That's just the highest temp reported by one of the 1000+ sensors on the CPU. Just because there was that one sensor reading that temp, doesn't mean enough heat was getting to the AiO to boil anything. Unfortunately, heat density prevented heat from flowing out from that hotspot fast enough to let it cool down to a lower temperature. I suspect the reading was more accurate that people would like to think.

And, going back to the original topic . . . if you are going to cook something, you need high temps AND heat output. A surface with a tiny dot radiating 65W at a temp of 117C isn't going to cook a pancake. A Xeon W3175, on the other hand . . .

 

[Zucker2k]

I did and I got my CPU up over 79C without throttling.

So we have these Zen1 chips throttling at 75, 79, and 90-95c. Okay. That's a mess alright.

I take it you're fishing for something negative about AMD here? It isn't working.

I'm not here to change anyone's opinion of their favorite brand.

Ever heard of hot spot temps? That's what Matisse reports as its main temp. 117C is entirely possible.

So the main temp is just there for show, it doesn't trigger anything? I ask because throttling has to kick in at some point? I wonder what temp that may be. The sensor is accurately reporting a temp of 117c (and has been running primegrid for a couple of hours - from @VirtualLarry's posting), according to you.

 

[Topweasel]

So we have these Zen1 chips throttling at 75, 79, and 90-95c. Okay. That's a mess alright.

I'm not here to change anyone's opinion of their favorite brand.

So the main temp is just there for show, it doesn't trigger anything? I ask because throttling has to kick in at some point? I wonder what temp that may be. The sensor is accurately reporting a temp of 117c (and has been running primegrid for a couple of hours - from @VirtualLarry's posting), according to you.

It's not or has never been 90-95c Mark is missinformed and or confused and no one said it wasn't a bad choice. But you said it was "the biggest con job ever". It was confusing but had little to do with conning people. Conning would have been to have it the other way so that people and boards thought their CPU was running cooler and not near danger zone. This was the other way around, the agesa knew when to throttle, this just spun up the fans a little earlier, to allow it to last longer on boost before it heat soaked the cooler and throttled back down to the coolers rated power level. It was poorly handled and still confusing, but its an interesting thought process to get people using min/stock coolers a way to eek out extra performance.

Tons less of a con then down-clocking the cores every-time it did AVX2 workloads.

 

[DrMrLordX]

So we have these Zen1 chips throttling at 75, 79, and 90-95c. Okay. That's a mess alright.

Mine didn't throttle. Keep fishing.

I'm not here to change anyone's opinion of their favorite brand.

Yes you are. You're arguing about whether or not AMD CPUs are suitable to cook a pancake based on you not knowing how or when they throttle. It's a mess! Oh no!

So the main temp is just there for show

There is no "main temp". The only temperature reported for Matisse is the hotspot temp. You have no idea what the average die temp is because AMD's sensor array doesn't report that, and thank goodness for that. Welcome to advanced silicon nodes.

I ask because throttling has to kick in at some point?

It's supposed to kick in at 95C, but somehow one system managed to surge past that number. Granted he may have been throttling at that point anyway.

 

[.vodka]

So we have these Zen1 chips throttling at 75, 79, and 90-95c. Okay. That's a mess alright.

Let's clear this up as you are drowning in a glass of water and this has been dragging for too long in a thread about cooking pancakes with CPUs.


AMD couldn't get PB2 working on Zen1, so it shipped with a p-state based, rigid frequency control. This included XFR, which in the 1800X let it hit 4.1GHz on one or two cores for a short period of time.

Zen1 chips were particular in the sense that parts branded as X (1600X, 1700X, 1800X) reported TWO temperature readouts. Tdie and Tctl.

Tdie is the real measurement. This is the one that throttles the CPU if it goes too far. The 1800X's page lists max temperature as 95°C. The 1700 (non X) also lists it at 95°C. Same silicon.

Tctl is Tdie + 20°C, this is only used for fan control. Decent monitoring software like hwinfo could always see both. Some other crappy software probably reported Tctl as the real value and people with X parts were going crazy at launch time... until they noticed the exact 20°C difference with non X parts. AMD clarified it later.

This is from a friend's 1600X in hwinfo, I had to clear this up for him a little while ago

Zen1's boost was primitive and its XFR mechanism even more so, AMD thought fooling the fans to run faster to quickly get rid of the initial heat from boosting would help X parts sustain that little extra clockspeed with better cooling, which is true. This was solved with Zen+ and its Precision Boost 2 which in comparison seems alien tech. It works even better with Zen2's refined implementation.

Non X parts have Tctl = Tdie.

As a result it's just better to do an all core OC on Zen1, X or non X edition. This is not the case with Zen+ or Zen2. You can't beat the boost algorithm.


-----------------------------------------------------------------------------------------------

Now, on to the pancake cooking. Intel chips can run constantly at 100°C while throttling for a while already (even running without a heatsink), whereas AMD chips tend to throttle then shut down if the situation isn't solved. der8auer's video on Zen2 without heatsink:

I think I prefer AMD's approach, better to stop the anomaly right there instead of cooking the CPU (even if it's within spec)