AMD Ryzen 9000 Zen 5 "Granite Ridge" Desktop Processors Launch Late-July

[ARF]

You're not wrong there, although, I've never heard of anyone choosing a CPU based on the shape of its IHS.

Well, in the case of the AMD IHS, it has been proved that it worsens the cooling capabilities. It is not an optimal solution and needs readjustments, and like mentioned above, not good for cooler change / thermal paste reapply.

Which air cooler for a ryzen 9 5900x

It gets to much vcore stock Don't know about multicore load, which was the only time it gets hot like I describe. But at its highest, would be only with the low-thread-gaming-boost-profile. It was cool in games, even if it was at 1.5V! Multicore loads, for sure, OTOH, would be hot at only 1.3V.

www.techpowerup.com

 

[AusWolf]

Well, in the case of the AMD IHS, it has been proved that it worsens the cooling capabilities. It is not an optimal solution and needs readjustments, and like mentioned above, not good for cooler change / thermal paste reapply.

Neither my 7700X nor my 7800X3D reaches 70 °C in gaming, and both of them run just shy of Tjmax under a £50 single tower (Dark Rock 4) in Cinebench all-core. In my opinion, it's the offset compute die location combined with its huge thermal density that causes most of the heat issues. Besides, people like running their CPUs to the max for some reason, which is also unnecessary. Non-X parts do really well in thermal tests while sacrificing only a couple percent of performance compared to the X versions.

As for thermal paste cleaning, I agree, it is a bit of a pain.

 

[Tomorrow]

Hopefully there will be boards this time around with decent PCIe slot config not the current crappy offerings.

Im hoping for finally dual DIMM boards not these pointless quad DIMM boards. DDR5 standard size is 16GB so most people can get 32GB easily even with two slots (96GB with 48GB DIMM's) but manufacturers insist on putting four slots on every board. And franky let them do it for what i care. Just give same decent memory OC focused boards with two slots. Especially ATX ,not sized down ITX.

 

[Wirko]

I don't think I've ever seen another IHS with some many nooks and crannies still don't know why they designed it like that.

AMD copied Broadwell 2011-3 Xeon and i7 chips but increased the number of "feet" (actually we have a more proper terminology at TPU, look here):

Do you really think the only way to design the CPU was the one that required the IHS to have a complex shape?

As @SL2 said, the CPU would otherwise have to be larger. Larger substrate and larger socket translate to higher price. The IHS, on the other hand, is still a stamped piece of metal. It's not like the "nooks and crannies" are CNC-milled.

 

[Chaitanya]

Im hoping for finally dual DIMM boards not these pointless quad DIMM boards. DDR5 standard size is 16GB so most people can get 32GB easily even with two slots (96GB with 48GB DIMM's) but manufacturers insist on putting four slots on every board. And franky let them do it for what i care. Just give same decent memory OC focused boards with two slots. Especially ATX ,not sized down ITX.

There also are 64GB modules on the way so with dual slot boards you can get 128GB of RAM. MSI used to offer non overclocking board with 2 RAM slots but seems like they have vanished in recent times.

 

[KillSwitx]

My 7800x3D is serving me well but i'm curious to see how much a 4090 can stretch its wings with the new Zen 5 ... If it's the case of seeing a good uplift i may go for it ... but all depends if Zen 5 performance will be too dependent on new chipset and such.

 

[SL2]

In my opinion, it's the offset compute die location combined with its huge thermal density that causes most of the heat issues.

I dunno. Offset mounting lowers the temperature with < 3° C, that's nothing, and the chiplets has been there for five years now.

Take a 3950X, a 5950X, and a 7950X and set them at the same W usage with the same cooler and stress test them. I'm pretty sure the 7950X will be the hottest one.

It's nothing wrong with the irregular lid shape, it's the lid thickness that causes the high temperature, together with clock frequency changing with cooling capacity.

I still don't know why it's so thick. I mean, I know AMD wanted to keep cooler compatibility, but why isn't the socket taller instead?

 

[Daven]

AMD had to increase the z-height of the IHS to maintain cooler compatibility. Making changes on the socket side must haven been a no go. To save production costs, as much material as possible was removed while minimizing structural and thermal integrity loss. The ‘nooks and grannies’ IHS design was the result. Those with $100 plus CPU cooling solutions are grateful.

 

[SL2]

As @SL2 said, the CPU would otherwise have to be larger. Larger substrate and larger socket translate to higher price. The IHS, on the other hand, is still a stamped piece of metal. It's not like the "nooks and crannies" are CNC-milled.

It can't be stamped, just look at it, must be cast (or whatever english word I should use lol).

 

[Tomorrow]

There also are 64GB modules on the way so with dual slot boards you can get 128GB of RAM. MSI used to offer non overclocking board with 2 RAM slots but seems like they have vanished in recent times.

Unify-X if i remember correctly. Gigabyte even had B650E Tachyon but it was produced in limited quantity in Taiwan for XOC and never publicly sold despite having public product page.

 

[kondamin]

Im hoping for finally dual DIMM boards not these pointless quad DIMM boards. DDR5 standard size is 16GB so most people can get 32GB easily even with two slots (96GB with 48GB DIMM's) but manufacturers insist on putting four slots on every board. And franky let them do it for what i care. Just give same decent memory OC focused boards with two slots. Especially ATX ,not sized down ITX.

Why? having 4 slots allows you to just buy a second set of ram when the initial amount is no longer enough without you having to get rid of your first set.

I get by with 32GB right now, but I imagine that by this time next year and some windows updates I'll have to add 32 more to keep my current comfort

 

[SL2]

Making changes on the socket side must haven been a no go.

It's hard to figure out why tho. Why can't this part be ~1.7 mm taller?

To save production costs, as much material as possible was removed while minimizing structural and thermal integrity loss.

Well they added metal to the lid instead of plastic to the socket, dunno how that's cheaper, and temps have gone up, so...

It's possible that AMD wanted to avoid bendgate, simple as that. It's not a problem with PGA's (AM4).

 

[JWNoctis]

AMD had to increase the z-height of the IHS to maintain cooler compatibility. Making changes on the socket side must haven been a no go. To save production costs, as much material as possible was removed while minimizing structural and thermal integrity loss. The ‘nooks and grannies’ IHS design was the result. Those with $100 plus CPU cooling solutions are grateful.

There were stories of how there was supposed to be a vapour chamber in that thick IHS, which got dropped due to cost concerns and lack of efficiency, supposedly in low single digit celsius.

If only they had retained that, for Ryzen 9 79xx and maybe the X3D at least.

Why? having 4 slots allows you to just buy a second set of ram when the initial amount is no longer enough without you having to get rid of your first set.

I get by with 32GB right now, but I imagine that by this time next year and some windows updates I'll have to add 32 more to keep my current comfort

Getting to 6000MT/s or even 5000+MT/s is very difficult, if not impossible with 4 slots populated, it seems. The officially supported number for Zen 4 is DDR5-3600, and motherboards go up to 5000+ or thereabouts.

 

[Daven]

Why? having 4 slots allows you to just buy a second set of ram when the initial amount is no longer enough without you having to get rid of your first set.

I get by with 32GB right now, but I imagine that by this time next year and some windows updates I'll have to add 32 more to keep my current comfort

Aren’t there a bunch of limitations when using all four slots like speed, rank number, matching compatibility?

 

[dgianstefani]

There were stories of how there was supposed to be a vapour chamber in that thick IHS, which got dropped due to cost concerns and lack of efficiency, supposedly in low single digit celsius.

If only they had retained that, for Ryzen 9 79xx and maybe the X3D at least.


Getting to 6000MT/s or even 5000+MT/s is very difficult, if not impossible with 4 slots populated, it seems. The officially supported number for Zen 4 is DDR5-3600, and motherboards go up to 5000+ or thereabouts.

This is not a DIMM slot issue it's an IO die/IF issue. The chiplet design limits interconnect speed somewhat. Hence why IF speed is still around 2000 MHz on Zen 4, which was also achieved on Zen 3, but DDR speed has gone from 3800/4000 MT to around 6000 MT or 3000 MHz. So the ratio is no longer 1:1.

On Intel motherboards it's true that 1DPC boards overclock much better than 2DPC boards, but I don't think the effect is as pronounced on AMD.

I solved the poor IHS issue by delidding and replacing the socket retaining mechanism and IHS with the upgrade heatspreader from Thermal Grizzly, much better surface area and performance. This is likely too extreme for most though.

The effort and cost are mitigated by the looks and performance though, and I can reuse it for any future AM5 CPU, though I'll probably be switching to Arrow Lake anyway.

 

[AusWolf]

I dunno. Offset mounting lowers the temperature with < 3° C, that's nothing, and the chiplets has been there for five years now.

Take a 3950X, a 5950X, and a 7950X and set them at the same W usage with the same cooler and stress test them. I'm pretty sure the 7950X will be the hottest one.

It's nothing wrong with the irregular lid shape, it's the lid thickness that causes the high temperature, together with clock frequency changing with cooling capacity.

I still don't know why it's so thick. I mean, I know AMD wanted to keep cooler compatibility, but why isn't the socket taller instead?

The first chiplet CPU was built on what, 7 nm? We're on 5 now. As chips are getting denser and denser, they're also getting harder to cool. It's just the nature of things.

I'm currently running a 6500 XT (only temporarily, though) overclocked to the max at 2950 MHz, and its hotspot can reach 85-90 °C at times at 90-95 W of chip power with one of the best coolers on the market (it's a Sapphire Pulse) and no heat spreader. Why? Because it's a small and dense chip.

On the other hand, I somehow can't recall Ryzen 3700X, 5800X or X3D owners singing praises about how cool their CPUs run.

Like I said, large amounts of electricity passed through a small area will make it harder to cool. It's normal.

Edit: If the IHS really was an issue, then...
1. it probably would be at low loads as well, and
2. you wouldn't see your CPU cool back to idle temps basically immediately after a heavy workload finished. Heat would be retained in it as it is not conducted to your cooler by theory.

 

[SL2]

Edit: If the IHS really was an issue, then...
1. it probably would be at low loads as well, and
2. you wouldn't see your CPU cool back to idle temps basically immediately after a heavy workload finished. Heat would be retained in it as it is not conducted to your cooler by theory.

The thick lid acts as a bottleneck only when running the CPU at full load. I think it has to be way thicker to act like you suggest.

All I know is that when Der Bauer delidded a 7900X it dropped over 20° C in CBench, and when he did the same with the 14900K it dropped 10°.
Yes, apples to oranges, and only one CPU of each, but 20+ degrees from delidding is extreme, and AFAIK not a common thing.

 

[dgianstefani]

The thick lid acts as a bottleneck only when running the CPU at full load. I think it has to be way thicker to act like you suggest.

All I know is that when Der Bauer delidded a 7900X it dropped over 20° C in CBench, and when he did the same with the 14900K it dropped 10°.
Yes, apples to oranges, and only one CPU of each, but 20+ degrees from delidding is extreme, and AFAIK not a common thing.

Unfortunately, the bottleneck is at all levels of load. Hence why a ~50-70 W X3D chip can still touch 89 C at stock and can idle in the 60s even on good cooling, depending on silicon lottery. If you check out TPU cooler reviews, AMD chips typically have a tested cooler topping out at around 225 W before throttling, while Intel tests can push almost 300 W on the same cooler, and why the 7950X runs hotter than a 14900K, which uses more power on average.

On the other hand, my delidded chip has a core delta of 1 Celsius at low loads, with a temperature range from 23-55 Celsius, with coolant peaking at 25 Celcius but typically being around 20 Celcius. The max load temps have more of a delta because I use Process Lasso to designate certain cores for certain task categories, meaning my faster clocking cores are much more heavily loaded for foreground applications, and other cores never reach those temperatures because they're only handling background tasks.

 

[kapone32]

This is not a DIMM slot issue it's an IO die/IF issue. The chiplet design limits interconnect speed somewhat. Hence why IF speed is still around 2000 MHz on Zen 4, which was also achieved on Zen 3, but DDR speed has gone from 3800/4000 MT to around 6000 MT or 3000 MHz. So the ratio is no longer 1:1.

On Intel motherboards it's true that 1DPC boards overclock much better than 2DPC boards, but I don't think the effect is as pronounced on AMD.

I solved the poor IHS issue by delidding and replacing the socket retaining mechanism and IHS with the upgrade heatspreader from Thermal Grizzly, much better surface area and performance. This is likely too extreme for most though.

The effort and cost mitigated by the looks and performance though.
View attachment 349036

Idle in the 60s? That sounds very hyperbolic. If you have a 7X3D chip or just 7000 chip sitting in the 60s at idle you have done something wrong.

 

[swirl09]

Sadly as cool as this new stuff will be, i just got a 7800X3D, so im probably good until 2027-8. Oh well, End of july is my birthday ;p

Why sadly? You just got a new PC, enjoy it! Great chip, Id gladly have it ;p

 

[AusWolf]

Unfortunately, the bottleneck is at all levels of load. Hence why a ~50-70 W X3D chip can still touch 89 C at stock

Mine runs at 82-83 °C max in Cinebench at 80-85 W under a Dark Rock 4.

... and can idle in the 60s even on good cooling, depending on silicon lottery.

What? Mine idles in the low 40s with the aforementioned relatively affordable single-fan single tower.

 

[dgianstefani]

Mine runs at 82-83 °C max in Cinebench at 80-85 W under a Dark Rock 4.


What? Mine idles in the low 40s with the aforementioned relatively affordable single-fan single tower.

Depends on if you have certain power saving features enabled. I lock IF clocks to prevent the sleep bug and clock variation instability. Other low power states/efficiency features are fine though.
Still, low 40s vs low 20s indicates a problem with the IHS.

80 C on a 80 W part under a good air cooler isn't good. It's not terrible, but it's not good.

 

[AusWolf]

Depends on if you have certain power saving features enabled. I lock IF clocks to prevent the sleep bug and clock variation instability. Other low power states/efficiency features are fine though.
Still, low 40s vs low 20s indicates a problem with the IHS.

Sleep bug? Clock variation instability? I probably skipped a few articles recently, but I've never heard of these.

Low 40s vs 20s mostly indicates a 20-25 W idle vs a 2-5 W one. This is what AMD needs to get right in future generations first and foremost, imo.

80 C on a 80 W part under a good air cooler isn't good. It's not terrible, but it's not good.

Yep, agreed. As we keep shrinking our dies, it's only gonna get worse, I'm afraid.

 

[SL2]

Unfortunately, the bottleneck is at all levels of load. Hence why a ~50-70 W X3D chip can still touch 89 C at stock and can idle in the 60s even on good cooling, depending on silicon lottery.

I was talking about Zen 4 in general, not 3D models.

If you check out TPU cooler reviews, AMD chips typically have a tested cooler topping out at around 225 W before throttling, while Intel tests can push almost 300 W on the same cooler, and why the 7950X runs hotter than a 14900K, which uses more power on average.

That's what I meant with the bottleneck.

 

[dgianstefani]

Sleep bug? Clock variation instability? I probably skipped a few articles recently, but I've never heard of these.

Low 40s vs 20s mostly indicates a 20-25 W idle vs a 2-5 W one. This is what AMD needs to get right in future generations first and foremost, imo.


Yep, agreed. As we keep shrinking our dies, it's only gonna get worse, I'm afraid.

My chip doesn't idle at 2-5 W. It idles at similar power as every other single CCD Zen 4 part, actually slightly more because several power saving features are disabled. So around 30 W. The temperature difference is due to the stock IHS having been replaced.