AMD's Robert Hallock Confirms Lack of Manual CPU Overclocking for Ryzen 7 5800X3D

[Taraquin]

Thats system power consumption. No point in looking at system power, especially in gaming, cause GPU plays the biggest role


And its not in plenty of other applications as shown from gnexus review (vray , chromium compile, blender). Actually, correct me if im wrong, in every test gnexus ran besides adobe, both cpus consumed the same and performed the same.

They use the same GPU, do you mean that on the same GPU 10900K makes it use more power? MB can be a factor, but that might as well be disfavorable to the MB 5800X uses.

Please hear what Steven says in the video: "That makes it more efficient that 10900K in just about every workload including games in almost every instance." Also see what the consumption numbers were on TPU, chekc Tomshardware, it performs similar to 5800X but consumes much more in viritually everything. Ryzen 7 5800X Power Consumption, Thermals - AMD Ryzen 7 5800X Review: The Pricing Conundrum | Tom's Hardware (tomshardware.com)

Handbrake:
106W vs 206W

Y-cruncher:
112W vs 185W

 

[JustBenching]

They use the same GPU, do you mean that on the same GPU 10900K makes it use more power? MB can be a factor, but that might as well be disfavorable to the MB 5800X uses.

Please hear what Steven says in the video: "That makes it more efficient that 10900K in just about every workload including games in almost every instance." Also see what the consumption numbers were on TPU, chekc Tomshardware, it performs similar to 5800X but consumes much more in viritually everything. Ryzen 7 5800X Power Consumption, Thermals - AMD Ryzen 7 5800X Review: The Pricing Conundrum | Tom's Hardware (tomshardware.com)

I agree that it is more efficient, but the difference is negligible. Its literally less than 10% in everything except adobe where the 5800x was actually around 20% faster with the same power draw.

 

[ThrashZone]

Well then you cant compare anything really. There is no other 10 core cpu, the 5950x has 33% more threads than the 12900k so we shouldn't compare them etc. And why stop at the number of cores? Let's not compare two cpus unless they are on the same node or running the same frequency either.

Hi,
To bad nobody thought to use 11900k which is a 8 core same as 5800x.

 

[JustBenching]

Hi,
To bad nobody thought to use 11900k which is a 8 core same as 5800x.

The 11900 stinks. Nobody would argue thats it is as efficient as zen 3

 

[Taraquin]

I agree that it is more efficient, but the difference is negligible. Its literally less than 10% in everything except adobe where the 5800x was actually around 20% faster with the same power draw.

That depends. If your motherboard runs 125W limit out of the box it will be like that, slightly less efficient than 5800X, but usually slower. Many motherboards run unlimited power out of the box, Asus usually do this, then efficiency is terrible, but it performs on pair and sometimes a bit better than 5800X.

5800X always has 142W PPT no matter what unless you manually change it. On Intel it depends on MB, what bios you use etc.

 

[JustBenching]

Handbrake:
106W vs 206W

Y-cruncher:
112W vs 185W

Havent we been over this? Why are you repeating the same stuff? Those tests are done withing the pl2 240w window. Yes, nobody argues that the 10900k is more efficient when it is runninf at 240w....

I dont know, i feel like im being trolled..

 

[ThrashZone]

The 11900 stinks. Nobody would argue thats it is as efficient as zen 3

Hi,
But somehow a 10900k is ?

 

[JustBenching]

That depends. If your motherboard runs 125W limit out of the box it will be like that, slightly more effeicient, but usually slower. Many motherboards run unlimited power out of the box, Asus usually do this, then efficiency is terrible, but it performs on pair and sometimes a bit better than 5800X.

5800X always has 142W PPT no matter what unless you manually change it. On Intel it depends on MB, what bios you use etc.

Why does that matter? I really dont get you. If YOU care about efficiency then you will manually limit the cpu to 125w,it literally takes a second. So why the heck do you care what a random motherboard does? I seriously dont understand what your point is

Hi,
But somehow a 10900k is ?

If we are comparing it to a 5800x, yes, at the same watts they perform almost identical in mt workloads (blender vray chromiun corona etc.)

 

[Taraquin]

Why does that matter? I really dont get you. If YOU care about efficiency then you will manually limit the cpu to 125w,it literally takes a second. So why the heck do you care what a random motherboard does? I seriously dont understand what your point is


If we are comparing it to a 5800x, yes, at the same watts they perform almost identical in mt workloads (blender vray chromiun corona etc.)

Most user will never enter the bios and do anything, so a lot of 10900K users gets a very inefficient CPU, that is my point. If you say stock you can both say 240 and 125W on the 10900K as it depends. 5800X is always the same and will always be efficient.

You first said they were equally efficient without stating that you need 125W limit for that, I disagreed, but now admit 5800X is more efficient even with 10900K running 125W limit so I`m satisfied It`s not about winning a discussion, it`s about getting the facts straight.

"not realising it is as efficient as the 5800x" that was your original claim which I disagreed on. A simple correction of saying it could be nearly as efficient if running 125W limit I would have had no reason to argue with you

 

[Chrispy_]

All-core manual overclocks died on Ryzen with the launch of PB2 and XFR2 four years ago because whilst it's easy to push a manual OC higher than Zen3's boost clocks, it's not a good idea for a long-term or daily-driver because the voltages will almost certainly kill your CPU. There are plenty of reports of Zen2 and Zen3 that have been volted too hard and are now functioning in limp mode, barely able to maintain stock clocks without boost.

At 1.35V, Zen3 is still likely going to boost to 4.6GHz based on the 5800X experiences. The cache will just have to make up for the missing 200MHz

 

[Valantar]

Of course, wasnt comparing the 10900 to the 5950x but to the 5800x.

Even for that comparison - which, again, is a wide-and-slow (at those power levels) design, at least compared to the 8c16t 5800X, which also as explained above is a worst-case scenario for Vermeer - the 5800X is ahead though. 15.7 minutes vs. 16.5 is a 5% advantage, at about 1% lower power. That's definitely not huge, but if that's where you go for your "well, actually" argument, you're really cherry-picking.

What do you mean very power limited? It's as power limited as the 5800x since they consume the same

In gaming im sure they are pretty similar as well, zen 3 consume a lot during gaming, dunno why

AMD power limits are strictly enforced, while with Intel MCE/unlocked power limits by default has been the norm (at least on moderately high end motherboards) since at least Skylake, and with ADL it is now the actual stock behaviour for K SKUs. It is thus reasonable to highlight that the 10900K is more strictly limited than what is found in most reviews, or in most users' configurations. That still doesn't make it inherently terribly inefficient, but when one part needs manual tuning in most use cases to match the stock efficiency of another, then that part is less efficient in any practical general sense of the word.

Why does that matter? I really dont get you. If YOU care about efficiency then you will manually limit the cpu to 125w,it literally takes a second. So why the heck do you care what a random motherboard does? I seriously dont understand what your point is

Given that most PC builders don't even activate XMP, this is a much higher bar than you're making it out to be.

 

[DeathtoGnomes]

I love discussion when one party intentionally changes their talking points and facts. I cant add to that discussion so I'll just make some popcorn.

 

[Valantar]

All-core manual overclocks died on Ryzen with the launch of PB2 and XFR2 four years ago because whilst it's easy to push a manual OC higher than Zen3's boost clocks, it's not a good idea for a long-term or daily-driver because the voltages will almost certainly kill your CPU. There are plenty of reports of Zen2 and Zen3 that have been volted too hard and are now functioning in limp mode, barely able to maintain stock clocks without boost.

At 1.35V, Zen3 is still likely going to boost to 4.6GHz based on the 5800X experiences. The cache will just have to make up for the missing 200MHz

My thinking exactly. And if AMD's early 15% number is trustworthy, 4.6GHz*1.15=equivalent to a 5.29GHz 5800X, so that's more than enough.

 

[Deleted member 24505]

I don't give a crap about efficiency as long as i can cool it.

 

[Taraquin]

I don't give a crap about efficiency as long as i can cool it.

Buy Rocket lake! Huge die, easier to cool

 

[JustBenching]

Even for that comparison - which, again, is a wide-and-slow (at those power levels) design, at least compared to the 8c16t 5800X, which also as explained above is a worst-case scenario for Vermeer - the 5800X is ahead though. 15.7 minutes vs. 16.5 is a 5% advantage, at about 1% lower power. That's definitely not huge, but if that's where you go for your "well, actually" argument, you're really cherry-picking.

I dont think 5% is an actual difference, and I would state the same whether its intel or amd at the top. It's basically peanuts. For me, a difference would be something like 15-20% and upwards. 5% means one CPU is consuming 125w and the other one 131w for the same workloads. It's kinda whatever. I consider them both equally efficient. You want to be precise, then sure, the 5800x is technically more efficient, by a small - irrelevant margin.

AMD power limits are strictly enforced, while with Intel MCE/unlocked power limits by default has been the norm (at least on moderately high end motherboards) since at least Skylake, and with ADL it is now the actual stock behaviour for K SKUs. It is thus reasonable to highlight that the 10900K is more strictly limited than what is found in most reviews, or in most users' configurations. That still doesn't make it inherently terribly inefficient, but when one part needs manual tuning in most use cases to match the stock efficiency of another, then that part is less efficient in any practical general sense of the word.

Sure, but I was talking more about the actual technical side of the cpu (architecture + node) rather than the actual configuration Intel or mobo manafacturers ship it with. You can change the latter, you can't change the former, that's why I'm focusing on what the CPU can do with forced power limits. If it was fundamentally inefficient, then changing the power limits wouldn't do anything, it would remain inefficient.

Other than that, I don't disagree, the way Intel decides to configure most of its' cpus, they are pretty much bad when it comes to efficiency out of the box. But as ive said, that's something that you can easily change.

You first said they were equally efficient without stating that you need 125W limit for that,

You are right, should have mentioned it. Usually when Im talking about efficiency im talking about there being a power limit.

 

[Mussels]

Having not read the thread yet, this is my comment to the poll question:


Yes this is a dealbreaker to stop me buying on launch day - i'll wait and see what people figure out before considering it.
I love my 5800x, but higher performance with some mitigation for the higher than usual temps is what I want from an upgrade - and i *like* my current low voltage all core overclock (1.2V, 4.6GHz)

It looks like PBO will still be around, there are reports of the 1.2.0.6B AGESA limiting 5800x CPU voltages - now we know why.
We should have PBO tweaking, just no manual voltage/multi control.

Edit as i go with replies:

Last time I overclocked a CPU was a AM2 CPU and havent touched OC either on my 4770k or 3700x.

The 4770K I gave to my dad was the best clocking CPU i ever experienced.

From 3.5GHz base to 4.5Ghz daily (for 10 years!) without ever a crash, running 32GB of 2400MHz DDR3 (from the measly 1333Mhz jedec standard)
Absolutely amazing how well it's aged, and it outperforms quite a few modern 4 core CPUs

Yeah we'll be missing out on that sweet, juicy GHz OC everyone is talking about..

View attachment 240110

It's the age old I-want-something-for-nothing argument, just because you're only willing to pay extra for more cores and nothing else.

If it beats the 5900X in gaming, and I'm not saying that it does (I know the demo used 12 core models), then why not charge a premium for it? Hint: You're not paying for more cores, you pay for extra cache.
It's for gaming, you can't overclock manually.. if you don't like it then don't buy it.


Just no. That's not a proof, that's a guess at best.
If the 3D cache was anywhere close to unreliable like you suggest then AMD wouldn't launch Milan-X this month.

This CPU won't cannibalize anything. Why? Well read this thread and you'll get a few hints. Nobody thinks this is a bargain so far at $450.
Besides, this CPU is supposed to have limited availability, although we'll see about that..

I get that ~2% loss you showed, but with 35W less power used and 15C lower temps. There is a benefit to it.

 

[sillyconjunkie]

Seems to be a lot of confusion about the oc lock. Pretty sure it's one main reason...Heat.

By design it was never going to be a great oc cpu. All the extra cache requires that much more package voltage which equals heat. The second cache layer + oc voltage is gonna cook the cores on the ground floor.

It's locked because it should be. (disclamer..I like amd stuff).

 

[Jism]

All-core manual overclocks died on Ryzen with the launch of PB2 and XFR2 four years ago because whilst it's easy to push a manual OC higher than Zen3's boost clocks, it's not a good idea for a long-term or daily-driver because the voltages will almost certainly kill your CPU. There are plenty of reports of Zen2 and Zen3 that have been volted too hard and are now functioning in limp mode, barely able to maintain stock clocks without boost.

At 1.35V, Zen3 is still likely going to boost to 4.6GHz based on the 5800X experiences. The cache will just have to make up for the missing 200MHz

What most review websites show, is a absolute wrong impression on which voltages to set. Like, look at the avg ryzen CPU review and notice how they all insert 1.4V up to 1.45V. It is a absolute wrong message and ive seen quite some folks blowing up their Ryzen that cant even hold stock clocks or boost anymore, because they ran daily on 1.4V or even higher.

1.33v ~ 1.37v is the absolute max. Anything above will seriously degrade your chip in months not even weeks.

Seems to be a lot of confusion about the oc lock. Pretty sure it's one main reason...Heat.

By design it was never going to be a great oc cpu. All the extra cache requires that much more package voltage which equals heat. The second cache layer + oc voltage is gonna cook the cores on the ground floor.

It's locked because it should be. (disclamer..I like amd stuff).

Incorrect. They said it themself the Cache is dependent of the CPU core voltage. They coud'nt design a seperate voltage rail among it due to compatible pin layouts. So you go with what you have, really. Since it's a EPYC gimmick, epyc's where never tested against OC's since the clocks of those CPU avg on 2 to 3.4Ghz.

 

[Cutechri]

1.33v ~ 1.37v is the absolute max. Anything above will seriously degrade your chip in months not even weeks.

I'd say 1.28v is the absolute max, seeing as that's the SVI2 TFN every single one of my Zen 2 and above CPUs has been sitting at during all core loads. Someone had their 3600 die within 6 months at 1.325v manual OC.

 

[Jism]

I'd say 1.28v is the absolute max, seeing as that's the SVI2 TFN every single one of my Zen 2 and above CPUs has been sitting at during all core loads. Someone had their 3600 die within 6 months at 1.325v manual OC.

I have quite some experience with bulldozer's / FX, they ate 1.65V for breakfast on water from the stock 1.35V. But the Ryzen is a complete different animal; it's so small that any higher voltage in combination with a high current will cause electromigration.

Degradation is real; i woud'nt want to mess with the 5800X's Cache at all. If that degrades your toast. However i still believe BCLK oc'ing should be possible within the realms of acceptable voltages.

 

[trieste15]

The 4770K I gave to my dad was the best clocking CPU i ever experienced.

From 3.5GHz base to 4.5Ghz daily (for 10 years!) without ever a crash, running 32GB of 2400MHz DDR3 (from the measly 1333Mhz jedec standard)
Absolutely amazing how well it's aged, and it outperforms quite a few modern 4 core CPUs

Actually, this kinda shows just how much reserve CPU capability that Intel was holding back from all of us..

It makes me thankful that AMD surged back and basically the situation now is that overclocks are no longer easy... and this is a good thing for the majority (not so good for the overclocking enthusiast).

 

[Chrispy_]

What most review websites show, is a absolute wrong impression on which voltages to set. Like, look at the avg ryzen CPU review and notice how they all insert 1.4V up to 1.45V. It is a absolute wrong message and ive seen quite some folks blowing up their Ryzen that cant even hold stock clocks or boost anymore, because they ran daily on 1.4V or even higher.

1.33v ~ 1.37v is the absolute max. Anything above will seriously degrade your chip in months not even weeks.

Yeah. When websites set voltages above 1.4V it's usually just to see what the absolute limit is of the silicon for a one-off result. It's definitely not a daily-driver overclock.

AMD themselves clarified that fixed voltages above 1.4V are seriously detrimental to the health of the silicon, and just because your CPU shows boost voltages of up to 1.55V in monitoring software doesn't mean that it's safe to set that as a manual voltage. The boost algorithm allows single-core peak voltages up to 1.55V only for a few hundred milliseconds at a time and software will only report the peak reading at much slower update intervals.

In reality, when you are running a single-threaded benchmark using a non-overclocked Ryzen with regular XFR boost, you're getting voltages of ~1.5V only as the final voltage before the algorithm swaps the load to another core, allowing the original core to both cool and discharge. Ryzen Master shows the core-juggling better than most software but even that doesn't paint the true picture, and was explained in more depth by Rob @amd in a tweet to der8auer. When boosting on a core, the voltage isn't fixed for the roughly half-second of load on that core; It starts off lower, at ~1.3V and ramps up as the charges and thermals in that area of silicon increase over the duration of that brief boost on the core. Over that single half-second cycle on one core, the average voltage to that core will likely still be under 1.4V and even if it isn't, the percentage of time that the core spends at voltages that could promote electromigration is so negligible that the effective time-to-death of the CPU vastly exceeds the warranty period, and likely the relevance of the CPU before obsolescence.

 

[Octopuss]

I had to stop reading on page 4.
So much idiocy in one thread. I can't take anymore of that.

 

[Melvis]

Who gives a flying F if its doesnt OC? If the performance is as good as what they say then its a win win for us who are on a AM4 4yr+ old System to make it a beast in 2022. This would probably the CPU I would upgrade to, depending on the price though, if its not crazy priced above the 5800X then why not! Just so awesome to have a PC or many of them and to have SO many CPU upgrade options is just amazing.