AMD at CES

[pax]

AMD just did its presentation and showed some cool laptop cpus... They showed also the desktop 5800x3d thats 15% faster than the 5800. Pretty much what was expected but only that one sku as seems AM5 is closer than we thought because they said AM5 is '2nd half 2022' so anytime after July. Cool thing tho is AM5 Ryzen 7000 can run 5 ghz all core already in the lab.

No navi3 info tho. I think 5nm is booked to the hilt.

[acroig]

Originally Posted by pax

Cool thing tho is AM5 Ryzen 7000 can run 5 ghz all core already in the lab.

Nice!!

[Sasquach]

-Radeon 6500XT for $200 US. Not sure what the performance level of this is, Seems like the specs are half of a 6600XT
-Ryzen 6000 series mobile APU, Zen3+ CPU + 12 RDNA2 CU's
-Ryzen 5800X3d (with vcache), seems like the last AM4 CPU (+15% increase over 5900X) spring 2022
-Ryzen 7000 CPUs, Zen4 and new socket AM5 (LGA1718) second half of 2022

[acroig]

Originally Posted by Sasquach

-Ryzen 5800X3d (with vcache), seems like the last AM4 CPU (+15% increase over 5900X) spring 2022

Is that confirmed for socket AM4?

[pax]

Originally Posted by acroig

Is that confirmed for socket AM4?

Yes but there is a speed degradation at 3.7 to 4.5 ghz max vs 3.8-4.7 ghz on the 5800x. Still they get that 15% after shaving 200 mhz so cache matters and Im sure they will have 3d cache on the ryzen 7000.

Probably one of the reasons why they didnt launch the 5900x3d tho.

[pax]

Nvidias presentation was meh, 3050 low end and 3090ti... but Intel's 12900KS at 5.5GHZ single core and 5.2 allcore was pretty impressive.

[acroig]

Originally Posted by pax

Intel's 12900KS at 5.5GHZ single core and 5.2 allcore was pretty impressive.

Agreed!

[Trunks0]

Yeah I'm kind of guessing here, but maybe the reason they didn't announce more like the 5800X 3D is lack of production capacity. I'm kinda hoping other 3D variants trickle out later as capacity frees up, but we may be well into AM5 by then so it may not happen.

[bill dennison]

Originally Posted by pax

Nvidias presentation was meh, 3050 low end and 3090ti... but Intel's 12900KS at 5.5GHZ single core and 5.2 allcore was pretty impressive.

Originally Posted by acroig

Agreed!

you know the 12900KS is going to draw more then the 12900K
https://www.techpowerup.com/review/i...th-gen/20.html

with a 4090 & 12900KS your going to need two PSU s

[acroig]

Originally Posted by bill dennison

you know the 12900KS is going to draw more then the 12900K
https://www.techpowerup.com/review/i...th-gen/20.html

with a 4090 & 12900KS your going to need two PSU s

[pax]

Originally Posted by Trunks0

Yeah I'm kind of guessing here, but maybe the reason they didn't announce more like the 5800X 3D is lack of production capacity. I'm kinda hoping other 3D variants trickle out later as capacity frees up, but we may be well into AM5 by then so it may not happen.

Feels like the swan song for am4 for gamers only. That tells me AM5 is 3 to 6 months after so it has me wanting to wait. Hope DDR5 situation is better by then.

[bill dennison]

Originally Posted by pax

Feels like the swan song for am4 for gamers only. That tells me AM5 is 3 to 6 months after so it has me wanting to wait. Hope DDR5 situation is better by then.

i need to upgrade my 1900x soon but wanted a 5900X3D 12 core

then upgrade my 3800x with AM5

[pax]

Originally Posted by bill dennison

i need to upgrade my 1900x soon but wanted a 5900X3D 12 core

then upgrade my 3800x with AM5

Ya my 1700x is way long in the tooth... Its going to htpc soon. But I also wanted to upgrade that to a 5800x or the 3d one. I just cant justify a mobo for the last gen cpu.

[KAC]

Spring is too late. May wait till AM5 and buy it end of 2022. I am in no hurry.

[Napoleonic]

5800x3d? Are going back to the 90s and early 2000s silly naming scheme?

[Nunz]

Barely wins 3 and ties 3.. and the 3 that "tied" were the CPU limited benches. It's just crap marketing slides and don't mean anything, but showing 3 ties isn't exactly what I'd call high faith in your product.

Originally Posted by bill dennison

you know the 12900KS is going to draw more then the 12900K
https://www.techpowerup.com/review/i...th-gen/20.html

with a 4090 & 12900KS your going to need two PSU s

The 12900K is a power-hog in cinebench and P95, but in actual gaming it's actually really efficient - more efficient than my 10900K.

They are idling at 86watts because they made the chip do that - it means they turned off dynamic ratio and turned off the C-States for the chip. My 12700K idles at 10-12watts with C-States limited to C3 (no core goes to sleep). At 5.1 P cores with E cores at 4.1, I only pull 230watts in CB23. You can make any chip pull stupid power numbers if you smash the voltage up lol

As for this 3d cache .. I'm curious how the cache is going to affect cooling. They had to build a strong supporting die and although it didn't increase the Z-height, I don't see how they could have done this without affecting thermal dissipation. Very curious to see a tear down of the chip compared to non 3d cache

[bill dennison]

we should know in a month or two how the 3d cache will do

but i'll have it under water so it should be fine

i will need a new block as i'm sure the solid copper bar it got on the threadripper won't work

and may not any of the current am4 waterblocks without new mounting as it maybe a taller chip

[Megaman]

I'd hate to clean thermal paste off the new AM5 processors.

[Megaman]

Originally Posted by Nunz

My 12700K idles at 10-12watts

I have had my cpu idle using only 9.005 watts. There....I can finally say I have better scores in something in this damn forum.

[KAC]

Originally Posted by Megaman

I'd hate to clean thermal paste off the new AM5 processors.

Still a better story than cleaning a GPU

[Nunz]

Originally Posted by bill dennison

we should know in a month or two how the 3d cache will do

but i'll have it under water so it should be fine

i will need a new block as i'm sure the solid copper bar it got on the threadripper won't work

and may not any of the current am4 waterblocks without new mounting as it maybe a taller chip

The issue with the cache+die changes is that your cooling doesn't matter - the limitation becomes the ability of the IHS to transfer the heat to your cold-plate. That's why the 5000 series will still hit 80c even with a custom loop. The thermal density is high and thermal transfer is the limiting factor.

The stacked cache and shaved die will only make that worse, unless they came up with some new method, but everything I've read says they know that the thermal transfer capabilities will be worse - hence the reduced core speeds.

They've said the chip height is not changing, so all AM4 blocks will work. That's the "Z-Height" that was referred to.

[demo]

Originally Posted by Nunz

Barely wins 3 and ties 3.. and the 3 that "tied" were the CPU limited benches. It's just crap marketing slides and don't mean anything, but showing 3 ties isn't exactly what I'd call high faith in your product.

Bit weird. Never seen a marketing slide with so many 'ties'



That said, there will be plenty of games that do benefit from the extra cache and those that don't.

[bill dennison]

Originally Posted by demo

Bit weird. Never seen a marketing slide with so many 'ties'



That said, there will be plenty of games that do benefit from the extra cache and those that don't.

games will likely need to be coded alittle different for 3d cache or a patch

new games after release maybe better

[bill dennison]

Originally Posted by Megaman

I'd hate to clean thermal paste off the new AM5 processors.

would be interesting if they can socket a water block over that tall odd shape IHS

[Nunz]

Originally Posted by bill dennison

games will likely need to be coded alittle different for 3d cache or a patch

new games after release maybe better

The cache is no different from the cache already existing on the chip - the only thing new is the way they placed the cache. Software doesn't see anything different.

Adding cache is not a 1:1 performance increase. I said it a while ago, that 15% is generous and likely will not be that number in gaming applications. You'll see more boost in workstation loads and synthetic benchmarks, though.

[pax]

They were talking about putting in cooling layers in 3d stacked chips. One model was using empty silicon thats used to help dissipate the heat and another fancier model had actual liquid cycling between the layers... I wonder if they skipped it this time for this one off 7nm cpu and achieved their temp goals with the slight clock drops.

All those gaps in the IHS of the am5 model might be for some cooling system...

[Nunz]

The more layers mean the worse the thermal transfer. The empty silicon would just make already poor thermal transfer a larger issue - direct-die is king for that very reason. I'm no engineer, but it makes sense to me. You want to be as close to the heat source as possible, and extra layers are pulling you further away from the heat source. I would not want a liquid layer in the silicon itself unless it was part of a connected system that interacted directly with that liquid.

[andino]

There was some talk about in die liquid cooling for the last few years that was gaining traction. The chips would have channels in place for a liquid to cool the hottest parts of the CPU in the die and then that liquid would be sent to the IHS for cooling. It cycled from heat alone so there was no need for a mechanical pump in die. Pretty sure that they were contemplating using that tech in the near future.

https://arstechnica.com/science/2020...water-cooling/

https://www.extremetech.com/computin...rmance-silicon

With AMD going to multiple layers it would be easier for them to put that kind of cooling system in place.