Dr. DroAnd that's where Foveros 3D stacking will kick in. Manufacturing those chips is still really hard, though...

Indeed. I personally am going for either RKL or a discounted ADL chip though. I want the last and probably best of the monolithic dies.

Then I'll wait to see what comes in 3 years with Nova Lake vs Zen (probably 6). Interestingly, Zen 6 should be a big.little architecture as well, using Zen 5 performance cores and low power Zen 4 for little cores. At least, that is the current rumor. So the march towards more and more threads, which mostly nobody needs, continues.

www.hardwaretimes.com/details-regarding-amds-future-big-little-architecture-surfaces-via-patent-separate-dies-and-process-node-for-different-core-clusters/

And as high as prices seem now, this may be the last time you can get a high end CPU like a 12700K for 'cheap'.

wccftech.com/intel-price-hike-to-affect-cpus-motherboard-chipsets-in-q4-up-to-20-percent-increase/

Leshyyes .. smaller area cost less to manufacture, but also smaller area can transfer less heat :)

That smaller area is barely a few mm^2 smaller than Intel, while heat density would almost certainly be higher in Zen they're also further apart. Three chiplets for 12-16 cores, at least 2 for lower core variants, have an entire IHS to dissipate heat from. Silicon is already a poor conductor of heat with pretty much all the heat taken care of through the IHS, so no die area usually has not that much of bearing on temps ~ the bigger issue is IHS/TIM & cooler!
And we've already seen how IHS contact, appropriate pressure & TIM can especially affect cooling in the longer term.

R0H1TThat smaller area is barely a few mm^2 smaller than Intel, while heat density would almost certainly be higher in Zen they're also further apart. Three chiplets for 12-16 cores, at least 2 for lower core variants, have an entire IHS to dissipate heat from. Silicon is already a poor conductor of heat with pretty much all the heat taken care of through the IHS, so no die area usually has not that much of bearing on temps ~ the bigger issue is IHS/TIM & cooler!

This is absolutely not accurate whatsoever. Remember: AMD's cores are essentially half the size of Intel's cores. That means, if consuming the same amount of power, thermal density is doubled. That represents a massive increase in silicon temperature at the same power draw, simply because the same amount of thermal energy is released in half the area. This obviously doesn't double temperatures (temperature scales are mostly arbitrary after all, and you need complex forumulas to convert energy output in a material to temperature), but it raises them significantly. The importance of the IHS, TIM and cooler are because of this increase in die temperature. If the die didn't have higher thermal density, you wouldn't need a better IHS, TIM or cooler.

Also, saying Zen cores are further apart is similarly inaccurate. Their smaller size means they are grouped closer per CCD, meaning that per CCD, not only is thermal density higher because of smaller cores, but it's doubly higher because of smaller cores that are grouped closer together. And, as Zen works with all modern schedulers to focus workloads onto one CCD as much as possible, relatively little work will be spread across the spatially separated CCDs until this is necessitated by the workload (essentially one or more heavy workloads exceeding the non-SMT thread count of one CCD). If you run a heavy 8T workload on a 5950X, the second CCD will be used only for background tasks, or will be asleep. Zen, at least Zen3 but IIRC also Zen2, is very aggressive in shuffling low threaded work between cores to avoid them overheating, but this is AFAIK entirely limited to the same CCD, as the latency penalty of moving a thread from one CCD to the other would be far too high, and would tank performance.

What you say above about the thermal conductivity makes it seem like you don't understand the core of this issue at all. A larger die area doesn't help cooling because of the thermal conductivity of silicon, it helps because the thermal energy is converted from electrical energy across a larger area, meaning that the temperature increase acorss said area will be lower. This isn't because of thermal conductivity, but simply a product of area (or really volume) of material vs. the absolute amount of energy in play. If you put a blowtorch to a penny, that penny will melt. If you were able to spread that same blowtorch flame across a bunch of pennies - even if their circular shape makes them conduct energy between each other very poorly - each penny would be much cooler.

ValantarWhat you say above about the thermal conductivity makes it seem like you don't understand the core of this issue at all. A larger die area doesn't help cooling because of the thermal conductivity of silicon, it helps because the thermal energy is converted from electrical energy across a larger area, meaning that the temperature increase acorss said area will be lower. This isn't because of thermal conductivity, but simply a product of area (or really volume) of material vs. the absolute amount of energy in play. If you put a blowtorch to a penny, that penny will melt. If you were able to spread that same blowtorch flame across a bunch of pennies - even if their circular shape makes them conduct energy between each other very poorly - each penny would be much cooler.

Yup. I've got experience with a CPU that has an exceptionally large die size, my 4669 v3 chip at... 662 mm². It uses about 75W in an AVX2 load across all 18 cores (at ~0.8 volts), and you would never believe how easy it is to cool. It's also the most reliable processor I've ever owned... and I guess for good reason. Shame it's a bit of a snail, 2.4 GHz Haswell really doesn't cut it nowadays, even with the 45 MB of L3.

DavenAm I missing something? AMD claims Zen 4 had a 9% IPC increase over Zen 3 in 1T Cinebench. But this benchmark run only shows a 14% increase with a 15% increase in clock speed.

You are. Zen 3 wasn't compared here, only the advantage from CPB.

Comparing with 12500, AMD seems not to beat Raptor. Okay, now we only have rumors, not certainties.
The i5-12500 works at 4.1GHz with all cores and can reach 4.6GHz with two. Now, it is the equivalent of the 5600 in the class. For the 7600X, the rival will be the 13600K, but it seems that it has problems even in front of the 12600K.
Regarding temperatures, the AMD processor has a smaller dissipation surface and the cooler must be more efficient. Even if it consumes less, the number of watts to be dissipated per mm2 is higher. At the same consumption as Intel, the cooler for AMD must be much more efficient.
Note: it would be perfect for us if the two companies launch processors with very similar performances. They will compete in tempting offers.

DavenAm I missing something? AMD claims Zen 4 had a 9% IPC increase over Zen 3 in 1T Cinebench. But this benchmark run only shows a 14% increase with a 15% increase in clock speed.

Where exactly did you see a ZEN3 vs ZEN4 scores in this thread?

This just the same CPU running at base clock and on full boost...

Zach_01Where exactly did you see a ZEN3 vs ZEN4 scores in this thread?

This just the same CPU running at base clock and on full boost...

You don't need Zen 3 scores in the opening article to be able to compare. A lot of people have Zen 3 CPUs at home, and even for those who don't, their performance is widely known by now.

AusWolfYou don't need Zen 3 scores in the opening article to be able to compare. A lot of people have Zen 3 CPUs at home, and even for those who don't, their performance is widely known by now.

True, but the comment being responded to was quite specifically formulated as if this was a Zen3 v Zen4 comparison.

I wouldn’t take these screens so seriously you guys. For all we know the person running these chips might not be so good at tuning. If it’s not getting enough power it will score lower even though clocks are high.

AusWolfYou don't need Zen 3 scores in the opening article to be able to compare. A lot of people have Zen 3 CPUs at home, and even for those who don't, their performance is widely known by now.

Yes I am one of them... and by my calculations according to those scores the AMD claim for IPC uplift between Zen3 and Zen4 is true.
At single thread this 7600X is doing more at 4.7GHz than my 5900X at 4.9~5.0GHz.

Read again what @Daven said... he uses percentages of the non boosting 7600X to the boosting 7600X to say something about the Zen4 IPC against Zen3 IPC...
Cant you see any flaw on that?

FeelinFroggyI'm not knocking AMD cause Intel has been building mini heaters for while now. But everyone says to just under volt the CPU to get rid of the heat. If you need to do that to keep the heat down then dont buy a $300 CPU. Why buy an expensive CPU if your just going to undervolt and slow down the chip. It's like driving a Ferrari with a governor installed to keep you from speeding.

Because some manufactures (if not all) are pushing their products to the edge and outside the efficiency curve in the name of competition.
On some chips you can cut down 20~30% of power and loose "just" 5% of performance.

Zach_01Yes I am one of them... and by my calculations according to those scores the AMD claim for IPC uplift between Zen3 and Zen4 is true.
At single thread this 7600X is doing more at 4.7GHz than my 5900X at 4.9~5.0GHz.

Read again what @Daven said... he uses percentages of the non boosting 7600X to the boosting 7600X to say something about the Zen4 IPC against Zen3 IPC...
Cant you see any flaw on that?


Because some manufactures (if not all) are pushing their products to the edge and outside the efficiency curve in the name of competition.
On some chips you can cut down 20~30% of power and loose "just" 5% of performance.

This. This is now true for almost all high-end gear - graphics cards included. Undervolting Ampere will give you silent gaming at virtually the same fps. Everyone pushing their hardware way beyond diminishing returns.

I am not finding the AMD option particularly compelling at its teased price points, the performance already exists at those price points for Alderlake.

I am honestly still ok with my old Zen3 PC :D

Contemplating 32GB 3600 14-14-14 Mushkin sticks right now :laugh:

Only because the Adatas I have in my kids rig has some issues from time to time.

freeagentI am honestly still ok with my old Zen3 PC :D

Contemplating 32GB 3600 14-14-14 Mushkin sticks right now :laugh:

Only because the Adatas I have in my kids rig has some issues from time to time.

That hurt my ego :eek: how is my 5950X rig already elderly :confused::laugh:

Dr. DroThat hurt my ego :eek: how is my 5950X rig already elderly :confused::laugh:

Its been a sweet year and a half :D