AMD Says Ryzen 1700X, 1800X Have a Temperature Reporting "Offset"

[R-T-B]

AMD is now saying reports of poor thermal performance from the flagship Ryzen products can be attributed to a simple thing: Temperature Offsets. Apparently, to keep a "consistent fan policy," AMD has placed a 20C offset on the Ryzen 1700X and 1800X products, making them report temperature a good 20C above what the sensor reads. This interesting design choice may most assuredly be confusing to end users, but AMD is confident software will soon automatically adjust for this offset and report the true temperature when required.

In the same blog post detailing the changes on the 1700X and 1800X, AMD claims that temperature reporting "may be offset on certain CPU models so that all models on the AM4 Platform have the same maximum tCTL value." This could mean other future models would utilize a similar setup, so remember that moving forward with AMD's Zen-based lineup.



View at TechPowerUp Main Site

 

[VSG]

Some have already begun to adjust for it, be it knowingly or otherwise. Corsair Link for example already reports a ~19-20 °C lower value than Ryzen Master and most BIOS values.

 

[cadaveca]

Wait, so my OC'd 67C load temps @ a bit over 200W are actually 47C? I doubt it. There is more to this story than meets the eye, because based on my load temps and the cooler I am using, temps should be a bit higher, perhaps.

 

[Aenra]

Just twenty? Why not 50 degrees higher? Might as well give them a heart attack, call it a day... whatever, lol

edit: sarcasm alone is pointless so i thought i'd explain, but i see TheLaughingMan has already expressed my reasoning just fine

 

[TheLaughingMan]

Wait. When it comes to the overclocking then, were people who tested it hitting a hardware wall at 4.1ish GHz or was a software process misidentifying the CPU overheating throttling it?

 

[dorsetknob]

Another reason for @W1zzard to hold off publishing his review untill he has had time to assess his testing regarding this relevation

people are going to be upset/glad/resentfull >>>>> take your pick

 

[ZoneDymo]

Wait. When it comes to the overclocking then, were people who tested it hitting a hardware wall at 4.1ish GHz or was a software process misidentifying the CPU overheating throttling it?

well throtteling would only mean it would never go higher then 4.1ghz as it would throttle it back to 4.1 to keep the temps in check.
The problem is that a little over it wont even boot properly in a lot of cases, so I highly doubt this will do anything for more overclocking potential.

 

[OneMoar]

20c what consistent fan policy what .... in the actual frigg
this smells of amds marketing spin

why TF would you adjust the sensor +20 fracking C to tune the fan profile
what are fan curves too hard for AMD engineers to understand go figure
there is no fracking way there is that much of a difference chip2chip

 

[mcraygsx]

I never liked the idea of Offset with Temperatures. I rather have True readings. This reminds me of quote

"This a dangerous game you're playing, Damaskinos"

Damaskinos being the AMD in this case.

 

[Ferrum Master]

there is no fracking way there is that much of a difference chip2chip

Unless the sensor is a dud on half of the stones lol.

Errata? There is really some background story behind this... more like a skeleton in the closet.

 

[R-T-B]

Wait, so my OC'd 67C load temps @ a bit over 200W are actually 47C? I doubt it. There is more to this story than meets the eye, because based on my load temps and the cooler I am using, temps should be a bit higher, perhaps.

Where did you get the reading? Some bioses may precompensate, but most do not. Until recently, nearly no software did. At least that's the official line.

20c what consistent fan policy what .... in the actual frigg
this smells of amds marketing spin

Personally, and not as a journalist but just as a dude with gut instinct, I agree Onemoar.

 

[cadaveca]

Where did you get the reading? Some bioses may precompensate, but most do not. Until recently, nearly no software did. At least that's the official line.

form Ryzen master and AIDA64.

See, I know what temps I should get based on what my cooler is capable of, and what I see if temps reported a bit low. It could be the board, for sure. ASRock Taichi isn't currently rated as one of the top picks for Ryzen for no reason.

 

[R-T-B]

form Ryzen master and AIDA64.

See, I know what temps I should get based on what my cooler is capable of, and what I see if temps reported a bit low. It could be the board, for sure. ASRock Taichi isn't currently rated as one of the top picks for Ryzen for no reason.

Again, taking my "official TPU" hat off, it does smell fishy, doesn't it?

 

[cadaveca]

it does smell fishy, doesn't it?

I don't exactly have a "retail" CPU, so... maybe not. We also do not know how exactly this offset is presented... is it hard-coded into the CPU, or is it IN BIOS? Eitehr way, BIOS can deal with it for some things you would think, but I am not sure all, if it is hard-coded in. It also might make sense that not all CPUs have the same offset applied, just like not all CPUs have the same XFR rating.

If what they want to do is have a consistent level of reporting from all chips, it would make sense that each chip has its own offset, and that it is hard-coded into each chip itself. That would really make it not fishy at all.

 

[OneCool]

None of this makes any sense to me. Just want to know the damn temp of the CPU!!

This 4.1ghz wall seems like Bs too..

 

[cadaveca]

This 4.1ghz wall seems like Bs too..

It doesn't when you see the power consumption. my 1700X pulls well over 200W @ 3.8 GHz. It seems that the chips are running into a frequency wall because of power consumption from what I see.

 

[R-T-B]

I don't exactly have a "retail" CPU, so... maybe not. We also do not know how exactly this offset is presented... is it hard-coded into the CPU, or is it IN BIOS? Eitehr way, BIOS can deal with it for some things you would think, but I am not sure all, if it is hard-coded in. It also might make sense that not all CPUs have the same offset applied, just like not all CPUs have the same XFR rating.

If what they want to do is have a consistent level of reporting from all chips, it would make sense that each chip has its own offset, and that it is hard-coded into each chip itself. That would really make it not fishy at all.

I suppose that would make sense. I just don't like the vibe I get, but I could be completely off.

 

[cadaveca]

I suppose that would make sense. I just don't like the vibe I get, but I could be completely off.

I look at Ryzen this way:

It is a huge success. It got rid of iGP, and instead gave us usable cores in mainstream platform, for users that do not use onboard video (which would be most of us here).

It is low on frequency, but to expect similar clocks as Intel from AMD's fab partner simply does not make any sense, so anyone that expected Ryzen to really compete with Intel are ignoring a fair bit of reality here.

And with those in mind, AMD did pretty good. Ryzen is pretty damn close to some of Intel's offerings. I evne think, had Ryzen been made on Intel's processes, that it would beat Intel easily.

But the fact remains that AMD will NEVER have silicon as good as Intel's, and as such, will always be slightly behind when it comes to clockspeed. IPC is great as is, anyway.

We can talk about temperature things, scheduling things, memory compatibility, whatever....none of which are really big issues. Yes, there are some things there that seem a bit funny with each of those parts of Ryzen, but none of them really affect how it works... the memory thing is easily solved by buying the right sticks. Scheduling... can't be fixed, IMHO (and AMD agrees it seems). Temperatures? What about Intel's TIM? These are all non-issues.

 

[R-T-B]

I look at Ryzen this way:

It is a huge success. It got rid of iGP, and instead gave us usable cores in mainstream platform, for users that do not use onboard video (which would be most of us here).

It is low on frequency, but to expect similar clocks as Intel from AMD's fab partner simply does not make any sense, so anyone that expected Ryzen to really compete with Intel are ignoring a fair bit of reality here.

And with those in mind, AMD did pretty good. Ryzen is pretty damn close to some of Intel's offerings. I evne think, had Ryzen been made on Intel's processes, that it would beat Intel easily.

But the fact remains that AMD will NEVER have silicon as good as Intel's, and as such, will always be slightly behind when it comes to clockspeed. IPC is great as is, anyway.

We can talk about temperature things, scheduling things, memory compatibility, whatever....none of which are really big issues. Yes, there are some things there that seem a bit funny with each of those parts of Ryzen, but none of them really affect how it works... the memory thing is easily solved by buying the right sticks. Scheduling... can't be fixed, IMHO (and AMD agrees it seems). Temperatures? What about Intel's TIM? These are all non-issues.

I agree about the general success of the product. I was speaking to this claim alone... but again maybe I'm just off and don't have a clue. I'm really not basing it on anything more than gut feeling there.

 

[Ferrum Master]

Scheduling... can't be fixed, IMHO (and AMD agrees it seems). Temperatures? What about Intel's TIM? These are all non-issues.

I wonder why multisocket CPU boards didn't show this before. They should suffer from the same problems actually.

 

[OneCool]

It doesn't when you see the power consumption. my 1700X pulls well over 200W @ 3.8 GHz. It seems that the chips are running into a frequency wall because of power consumption from what I see.

That's adding voltage to run all cores at 3.8?

 

[Fouquin]

20c what consistent fan policy what .... in the actual frigg
this smells of amds marketing spin

why TF would you adjust the sensor +20 fracking C to tune the fan profile
what are fan curves too hard for AMD engineers to understand go figure
there is no fracking way there is that much of a difference chip2chip

It seems the point of the offset is to avoid fan curves. Having the fans constantly ramping up and down is probably what they were avoiding by just raising the offset. Seems useless for anything but stock coolers, which obviously anyone buying retail samples won't be receiving.

 

[cadaveca]

That's adding voltage to run all cores at 3.8?

Of course.

 

[Ferrum Master]

Digging into the NUMA stuff...

It seems Linux sees ryzen much better assigning L3 to each core group.

https://www.google.lv/amp/s/amp.red...s/5yswtr/amd_ryzen_numa_and_numactl_on_linux/


Definetly a windows derp. They will patch it.

 

[Dippyskoodlez]

But the fact remains that AMD will NEVER have silicon as good as Intel's, and as such, will always be slightly behind when it comes to clockspeed. IPC is great as is, anyway.

1.) I don't see why frequency is relevant. Frequency is an aspect of design, not performance.

2.) Why can't AMD have 'silicon as good as intel"? They walloped the shit out of them with the Athlon Xp and the base Athlon 64 run(By making really good design decisions). This is an excuse for a lackluster product that is simply compensating for poor design choices, rather than holding their feet to the fire like people hold Intel to.

I wonder why multisocket CPU boards didn't show this before. They should suffer from the same problems actually.

Multi socket boards do 'show this issue', but NUMA is important because the memory controllers are physically separated from CPUs(N.on U.niform M.emory A.ccess). Numa issues do not apply to Ryzen because they are all bound to the same pair of memory controllers. (Besides, Windows has been NUMA aware for a long, long time at this point)

ccnuma is a potential issue, but re: scheduling, AMD has cleared the air entirely of all scheduler issues. There is nothing to fix.

The real issue is just the reality of using a low speed CCX between quads for AMD.

https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update?sf62109582=1