AMD Ryzen 9 7950X

[Valantar]

How does one even set the fan curves for air cooling on such a processor? If I understand correctly, it can now shoot to Tmax (95 degrees) even without the torture test like Prime95, just ordinary multicore load like unzipping, Cinebench? Even though there is still large difference in power consumption with lighter loads that push the CPU to 95 degrees, and full torture?

It's easy enough with custom water cooling, I use an Aqua Computer Quadro to run fan and pump speed relative to water temperature. But setting up a fan curve for air cooling directly from motherboard by CPU temperature will be less than ideal - either fans will overshoot often, spin at 100% when they don't need to, or you don't leave any headroom between light multicore load and a really heavy load with heavy power draw...

That's an excellent question - setting fan curves based on die temperature alone becomes insufficient with a boost system like this. Rather you would have to base it off of something more representative, like CPU power, but you also need to factor in actual temperatures at the same time. IMO, AMD should standardize a system for this and make motherboard vendors bake it into their BIOSes.

This is essentially how laptops operate: they (largely) ignore absolute temperatures and instead care about temperatures over time, power draw, etc., which allows them to both operate somewhat quietly, keep chips within their rated operating ranges, and ramp fans in a somewhat controlled manner. (The good ones, that is.)

 

[Solid State Brain]

How does one even set the fan curves for air cooling on such a processor? If I understand correctly, it can now shoot to Tmax (95 degrees) even without the torture test like Prime95, just ordinary multicore load like unzipping, Cinebench? Even though there is still large difference in power consumption with lighter loads that push the CPU to 95 degrees, and full torture?

It's easy enough with custom water cooling, I use an Aqua Computer Quadro to run fan and pump speed relative to water temperature. But setting up a fan curve for air cooling directly from motherboard by CPU temperature will be less than ideal - either fans will overshoot often, spin at 100% when they don't need to, or you don't leave any headroom between light multicore load and a really heavy load with heavy power draw...

On my MSI Z690 Intel motherboard I use socket temperature and almost completely disregard core temperature. I set the CPU to a 90 °C limit and let it regulate by itself under load while CPU fan slowly ramps up as the socket (slowly) heats up.

 

[MarsM4N]

That is true, but only if said heat is actually effectively transferred into the loop - and the problem here is that it isn't. The die is hot precisely because it fails to effectively transfer its heat into the cooling system, so despite the die being hotter, water (and general AIO) temperatures won't be.

You need also consider the factor "time". No matter how bad the heat transfer of the IHS is, if there is no water and/or air circulation, the water will heat up anyways. Just takes more time.
And the water in the CPU area will be very hot & expanding = increasing pressure in the loop.

 

[Valantar]

You need also consider the factor "time". No matter how bad the heat transfer of the IHS is, if there is no water and/or air circulation, the water will heat up anyways. Just takes more time.
And the water in the CPU area will be very hot & expanding = increasing pressure in the loop.

It absolutely will - but if you don't have flow, the CPU will be thermal throttling anyhow (ask me how I know), so that's a rather moot point as that will reduce the heat put into the system drastically. The pressure increase will also be negligible, given that water doesn't have all that much thermal expansion between 0-100°C. It definitely happens, it just isn't very significant.

And once you have flow, water temperatures will be roughly equal throughout the loop (again, thermal transfer seeks equilibrium), and in turn entirely dependent on how efficiently the IHS and cold plate are able to transfer heat into the water (as well as how efficiently the radiator and fans are able to dissipate it into ambient air, of course).

 

[zlobby]

On my MSI Z690 Intel motherboard I use socket temperature and almost completely disregard core temperature. I set the CPU to a 90 °C limit and let it regulate by itself under load while CPU fan slowly ramps up as the socket (slowly) heats up.

That's certainly one approach.

Not that anyone asked me, but here goes. Since I paid for the full potential of a CPU and I need the full potential of the CPU, I always let all fans run at 100%, both case and AIO. It could be a true nightmare, so I run Noctuas with an occasional PWM reduction of a case fan when 3000rpm (for example) is not really needed.
At the end there is just one pleasan hum, plus I got all the thermal headroom I can get.

 

[Bwaze]

That's certainly one approach.

Not that anyone asked me, but here goes. Since I paid for the full potential of a CPU and I need the full potential of the CPU, I always let all fans run at 100%, both case and AIO. It could be a true nightmare, so I run Noctuas with an occasional PWM reduction of a case fan when 3000rpm (for example) is not really needed.
At the end there is just one pleasan hum, plus I got all the thermal headroom I can get.

That's certainly one way of setting things up, but it would drive me crazy.

I use my PC for working, gaming, media consumption, light browsing... And it's idling a lot. These are uses with very different power consumption, and even 5900X and RTX 3080 can run very efficiently or dump large amounts of heat. And both at full throttle can in no way be quiet enough to place that PC by the TV and turntable, but by setting fans and pump by water temperature it's one if the quietest systems I ever built (4x140mm + 3x120mm radiator area also helps).

 

[Zubasa]

That's certainly one way of setting things up, but it would drive me crazy.

I use my PC for working, gaming, media consumption, light browsing... And it's idling a lot. These are uses with very different power consumption, and even 5900X and RTX 3080 can run very efficiently or dump large amounts of heat. And both at full throttle can in no way be quiet enough to place that PC by the TV and turntable, but by setting fans and pump by water temperature it's one if the quietest systems I ever built (4x140mm + 3x120mm radiator area also helps).

That is one of the reasons why I went Gigabyte, many of their boards have a temp sensor header.

 

[Legacy-ZA]

Does anyone know if AMD is planning to release a BIOS/UEFI update for AM4 platforms to auto-check the best memory timings without me having to do anything? I have to say, I kinda suck at memory overclocking, but wouldn't mind having tighter timings.

 

[Totally]

How can it be less efficient if its running at 25% higher performance? Your take is nonsensical!

Because it consumes double the power, it would need to have 100% higher performance to have the same efficiency.

 

[zlobby]

Does anyone know if AMD is planning to release a BIOS/UEFI update for AM4 platforms to auto-check the best memory timings without me having to do anything? I have to say, I kinda suck at memory overclocking, but wouldn't mind having tighter timings.

Hardly. AM4 work with XMP and anything extra is done by some mobo manufacturers as a bonus.

That's certainly one way of setting things up, but it would drive me crazy.

I use my PC for working, gaming, media consumption, light browsing... And it's idling a lot. These are uses with very different power consumption, and even 5900X and RTX 3080 can run very efficiently or dump large amounts of heat. And both at full throttle can in no way be quiet enough to place that PC by the TV and turntable, but by setting fans and pump by water temperature it's one if the quietest systems I ever built (4x140mm + 3x120mm radiator area also helps).

True. As it is common sense, eveyone shoul aim to satisfy their own needs and requirements. We must admit that vendors give us a lot of control an options to achieve that.

 

[Quake]

This is interesting - 20°C lower temps and higher clocks when delidded

 

[onemanhitsquad]

As I kept saying and people like Ratirl kept laughing, the 7950x is less efficient than the 5950x. Probably the same will apply to the whole zen 4 range. Yikes

Steve said in GN review, that AMD design Zen4 to boost like GPU, with a temp limit. So it goes to 95C in seconds and stays there, but does not throttle.

JaysTwoCents test showed the same...also claims slightly more efficient than 12900

 

[Bwaze]

This is interesting - 20°C lower temps and higher clocks when delidded

View attachment 263200

Yeah, but it's not something you can safely do yourself, Ryzen 4 processors have soldered heatspreader. Warranty on a processor costlier than average monthly salary in some European countries isn't something you throw away.

And that's one processor, one data point. And that's Der8auer, I wouldn't count on other mere mortals to achieve 20 degrees cooler CPU while overclocking.

 

[Valantar]

Yeah, but it's not something you can safely do yourself, Ryzen 4 processors have soldered heatspreader. Warranty on a processor costlier than average monthly salary in some European countries isn't something you throw away.

And that's one processor, one data point. And that's Der8auer, I wouldn't count on other mere mortals to achieve 20 degrees cooler CPU while overclocking.

Did you watch the video? And have you ever used any of Der8auer's delid tools? They are surprisingly easy to use as long as you follow the instructions, and are designed to minimize the chance of damage to the CPU, while breaking the solder bond safely.. Make sure everything is inserted and aligned properly and be careful when applying force, and anyone should be able to do this. The 20°C-ish thermal drop will be pretty much the same for anyone delidding and doing direct die cooling with liquid metal - there just isn't much variance to be found as long as you do all those things. That is obviously not to say that any of this is trivial, and the point about sacrificing your warranty is very valid (though CPUs essentially never fail unless they're DOA), but once those tools become available this should be reasonably accessible to any enthusiast with a moderate tolerance for risk.

 

[vmarv]

Those +90°C at full load from just the cpu will transform the room in an oven during summer months.
Imagine this thing paired with one or two 4090 and using both gpu and cpu for rendering.

 

[MarsM4N]

This is interesting - 20°C lower temps and higher clocks when delidded

Did he test the performance improvement after the delid? If there is none or just minor it's pretty pointless.

 

[A Computer Guy]

Man, stop. Just stop. Nothing pumps out temperature. Stuff pump out HEAT. Heat is not temperature. It doesn't make any freaking difference for the AIO whether your cpu is at 50c or 9000c.

Yes of course it pumps out heat. A poorly phrased question perhaps but the idea behind it is maybe you can't use your older cooling solution because you had a CPU that didn't pump out so much heat consistently. If the cooling was inadequate what would happen to an AIO in that scenario if you pushed it too far?

 

[Palindrome]

My takeaway from this wasn't even the performance of the 7950X but that the 5800X3D seems to deliver as much as twice the performance of my 2700X might have to consider an upgrade to that instead, especially since it seems to be able to hold its own against the 7950X in certain circumstances.

 

[JustBenching]

Yes of course it pumps out heat. A poorly phrased question perhaps but the idea behind it is maybe you can't use your older cooling solution because you had a CPU that didn't pump out so much heat consistently. If the cooling was inadequate what would happen to an AIO in that scenario if you pushed it too far?

Nothing will happen to your aio. The cpu will thermal throttle and drop power draw and clocks.

But who's going to have inadequate cooling for an 800 msrp cpu? I mean come on...

 

[Quake]

Yeah, but it's not something you can safely do yourself, Ryzen 4 processors have soldered heatspreader. Warranty on a processor costlier than average monthly salary in some European countries isn't something you throw away.

And that's one processor, one data point. And that's Der8auer, I wouldn't count on other mere mortals to achieve 20 degrees cooler CPU while overclocking.

I agree, it's not for every user. But my point is more towards possible issues with IHS/solder/pressure. 20°C for direct die on AIO cooler seems a lot to me.

 

[BorisDG]

Those +90°C at full load from just the cpu will transform the room in an oven during summer months.
Imagine this thing paired with one or two 4090 and using both gpu and cpu for rendering.

There is NO NVLink on 4090. nVidia killed it.

 

[Quake]

Did he test the performance improvement after the delid? If there is none or just minor it's pretty pointless.

I think he mentioned 50 MHz gain.

 

[vmarv]

There is NO NVLink on 4090. nVidia killed it.

With rendering programs you don't need to bridge them with a connector. The software sees that you have two gpus and use both, and some programs allow also to use the cpu with them (like Blender).
By the way, NVLink was removed from the 4090 and the RTX6000, but looks like they are keeping the technology for the servers/scientific lab solutions.

 

[oxrufiioxo]

Kinda surprised more outlets didn't test this cpu at 65w it still seems to beat a stock 12900k in a lot of workloads which is kinda crazy.

Seems like mobile Zen4 is going to be insane.

Either way this seems like the great milking generation even though AMD dropped 100 off the price of this cpu vs last generation the platform cost for a decent motherboard and some nice expo ddr5 is very high not that a 13900k/Z790 combo will be any better.

New builders are going to have a hard time deciding between a dead platform with lga 1700 and trying to save money with Z690 or going with a much more expensive but more future proof AM5. #ripdiypcbuilding

 

[Aranarth]

Yeah, but the problem is most people won't receive the advertised performance with a standard prebuilt PC like HP, Dell, or Lenovo. So they are paying for 7900x or 7950x performance they see in the reviews, but then when they buy the PC they receive far less performance (due to the cooler etc). It's essentially a scam: Advertising performance with a high-end cooler, but then delivering far lower performance (while not disclosing it).

I'll use an analogy to cars. Assume you bought a sports car for $80k based on reviews that said the car does the 1/4 mile in 9.7 seconds. But once you purchased it, you found the car was only capable of 12.7 seconds because all the reviews had equipped custom upgrades like drag slicks for tires, custom suspension etc. Would you be happy? You paid for 9.7 seconds of performance, but received 12.7 seconds.

AMD is doing essentially the same thing.

Yes I agree sort of.

The chip runs and the max speed the cooler is capable of, ANY cooler.
Want max performance? Build your own.

To go with a car analogy, BMW makes cars for driving or for racing.
The biggest difference after the increase in HP is the radiator.

A typical bmw is designed to do 125mph all day long on the autobahn with the odd vrroom from a stop light it is NOT designed for racing. If you try to race it you get heat soak.

But you can take any bmw put some upgrades such as better radiator, brakes, etc and race it. Heat is no problem.

Porsche does the same thing, they have 911's for cruising or they have 911's for racing. the racing verision is NOT fun drive every single day on city streets in (IE Detroit), Your back will hate you. The regular driving version is fine for regular driving.

Want a max performance car for racing? Build your own.

Need a machine to really crunch data build your own.
Otherwise go at whatever speed your radiator can give you and this chip will find it.