*** AMD "Zen 4" thread (inc AM5/APU discussion) ***

[humbug]

Currently have 5900x with x570 motherboard and 64GB RAM

If you go current Intel you can keep the RAM, just make sure its a DDR4 motherboard, if you want something that is better at Multithreading you will have to go for the 12900K.

I do believe the new AMD boards, that's AM5, are DDR5 only, so yes it probably will cost you more as you will need to replace that 64GB of DDR4 with DDR5 and that's not cheap.

Scaling. 10%-15% clockspeeds for 60% power consumption seems likely

You're going for a 6:1 power to Mhz ratio, even tho at the silicon limits i'm only getting a 4:1 ratio?

 

[Bencher]

You're going for a 6:1 power to Mhz ratio, even tho the silicon limits i'm only getting a 4:1 ratio?

Because you are already starting from a highly inefficient clockspeed which is 4.5ghz.

The 5950x is at 3.9 ghz for cbr23. Going from 3.9 to 5.0 ghz isnt all unlikely to need 60% more wattage. Afaik pbo takes the 5950x to 4.7 ghz at over 200w,right?

 

[humbug]

Because you are already starting from a highly inefficient clockspeed which is 4.5ghz.

The 5950x is at 3.9 ghz for cbr23. Going from 3.9 to 5.0 ghz isnt all unlikely to need 60% more wattage. Afaik pbo takes the 5950x to 4.7 ghz at over 200w,right?

Right, 4.7Ghz + 25% is 5.9Ghz, no one is talking about 5.9Ghz.

I'm starting at 4.65Ghz given that's what its clocking at in R23 out of the box.

At a 6:1 ratio it would be around <35,000. so by your measure, sure. We are never going to agree on that tho.

 

[humbug]

I gave you a clue in your own maths....

 

[Dg834man]

Scaling. 10%-15% clockspeeds for 60% power consumption seems likely

Can't make stipulations for power consumption and clock speed across two different architectures on different nodes, its not intel here when they claim new architecture and what they really mean is they changed the number on the box.

 

[humbug]

He's gone off line.

 

[Curlyriff]

Right, 4.7Ghz + 25% is 5.9Ghz, no one is talking about 5.9Ghz.

I'm starting at 4.65Ghz given that's what its clocking at in R23 out of the box.

At a 6:1 ratio it would be around <35,000. so by your measure, sure. We are never going to agree on that tho.

That is the thing, My 5950x also does 4.1Ghz out the box not 3.9Ghz and it does it at 102watt. If you take 4.1Ghz then you get to 5.15Ghz with that 25% efficiency at the same wattage then.

Even the 3.9Ghz taking a 25% efficiency then you would be looking at 4.9Ghz at 25% window. Of course it wont exactly scale like this. However you have the additional power increase from 105watt to 170watt is a 62% increase. Or for my particular example since it sits at 102watt would be 66% window.

Suggesting that out the box it can pull 170watt whilst being 25% more efficient I still expect a 5Ghz all core possible compared to the previous gen.

 

[humbug]

I think its with in the realms of possibility if you add another 10% on top of that.

If the chip can do it, and it seems it can, the power envelope is there for it.

Edit: but don't quote me on it

 

[humbug]

And here comes the next bombshell rumour.

Zen 4X3D could be as high as 30% faster than standard Zen 4.

If, unlike Zen 3X3D its clocking as high as standard Zen 4, why not..... we have seen gaming performance as high as 30% higher on Zen3X3D vs Zen 3 despite being clocked a few 100Mhz lower.

AMD Zen 4 V-Cache Early Leak: Intel shouldn’t get greedy with Raptor Lake…

I have early Zen 4 V-Cache performance numbers…Intel Raptor Lake is on borrowed time… [SPONSOR: https://www.aliexpress.com/campaign/2022-sale/main-venue-US-8...

youtu.be

 

[g67575]

Why don't AMD just use their X3D chips for all their processors? The technology is there and proven, not sure what the point is releasing regular version of their chips when they can just have the X3D versions of all of them

It's expensive, still quite new and a bit niche (but apparently great for framerate stability in games). Better to have product segmentation for AMD, and produce more X3D CPUs if demand is high.

The other thing is that V-cache seems to restrict clockspeed (to a limited extent). I imagine this will be less of an issue for Zen4.

 

[g67575]

It's interesting comparing the top Intel 12th gen CPUs to the 7700X. It does seem quite likely that the 7900 and 7950 will overtake the 12900KS in single core performance, perhaps also catching up with the 13700K (essentially a higher clocked 12900K, with more L2 cache)...
Chart scores here:

Cinebench R20 (Single-Core) CPU benchmark list

Cinebench R20 (Single-Core) list with performance comparison of all CPUs in this benchmark

www.cpu-monkey.com

Another thing to remember, is that the single core 13th gen scores weren't any higher than 12th gen, at the same clocks in CPUz, so I think Intel has more competition than some people seem to realise. Link here:

https://cdn.wccftech.com/wp-content/uploads/2022/07/2022-07-13_19-52-34-740x451.png

I wonder which 13th gen CPU Intel will be selling at ~£300, and will it be clocked high enough to keep up with the 7700X in single core performance? Edit- the 13600K will probably be priced ~£300, perhaps slightly higher.

 

[Bencher]

That is the thing, My 5950x also does 4.1Ghz out the box not 3.9Ghz and it does it at 102watt. If you take 4.1Ghz then you get to 5.15Ghz with that 25% efficiency at the same wattage then.

Even the 3.9Ghz taking a 25% efficiency then you would be looking at 4.9Ghz at 25% window. Of course it wont exactly scale like this. However you have the additional power increase from 105watt to 170watt is a 62% increase. Or for my particular example since it sits at 102watt would be 66% window.

Suggesting that out the box it can pull 170watt whilst being 25% more efficient I still expect a 5Ghz all core possible compared to the previous gen.

Thats not how it works. 25% more efficient means it will score 25% better at same wattage. The 7950x at 170w tdp will not be 25% more efficient than the 5950x at 105w tdp.

Actually stock for stock the 5950x will probably be more efficient

 

[g67575]

Another thing you can see from the Cinebench R20 (single-core) results, is that the 7700X is probably slightly faster than Goldencove CPUs at 5.0-5.1ghz We know that, because the max clockspeed of the 12900f is 5.1ghz...

I think the 13900K is probably overclocked a lot in the r20 Cinebench result, there is too much of a gap between the 12900KS and the 13900K.

 

[humbug]

It's interesting comparing the top Intel 12th gen CPUs to the 7700X. It does seem quite likely that the 7900 and 7950 will overtake the 12900KS in single core performance, perhaps also catching up with the 13700K (essentially a higher clocked 12900K, with more L2 cache)...
Chart scores here:

Cinebench R20 (Single-Core) CPU benchmark list

Cinebench R20 (Single-Core) list with performance comparison of all CPUs in this benchmark

www.cpu-monkey.com

Another thing to remember, is that the single core 13th gen scores weren't any higher than 12th gen, at the same clocks in CPUz, so I think Intel has more competition than some people seem to realise. Link here:

https://cdn.wccftech.com/wp-content/uploads/2022/07/2022-07-13_19-52-34-740x451.png

I wonder which 13th gen CPU Intel will be selling at ~£300, and will it be clocked high enough to keep up with the 7700X in single core performance? Edit- the 13600K will probably be priced ~£300, perhaps slightly higher.

It'll be close, the cited boost of the 7950X is 5.5% higher than the 7700X, 5.4Ghz vs 5.7Ghz

7700X: 773
12900KS: 795

773 + 5.5% = 815

Thing is tho they have the 13900K listed there, at 888, which is 14% higher than the 12900K, if we assume the P-Cores are the same for 12 and 13 gen then the 13900K is running 6.1Ghz to achieve 888 points.

 

[g67575]

Yes, it makes a direct comparison very difficult when they don't always show the results at stock clocks.

Intel ofc has an advantage in early benchmarks, as Ryzen CPUs aren't really designed to clock much higher than the speeds you get 'out of the box'.

 

[Bencher]

It'll be close, the cited boost of the 7950X is 5.5% higher than the 7700X, 5.4Ghz vs 5.7Ghz

7700X: 773
12900KS: 795

773 + 5.5% = 815

Thing is tho they have the 13900K listed there, at 888, which is 14% higher than the 12900K, if we assume the P-Cores are the same for 12 and 13 gen then the 13900K is running 6.1Ghz to achieve 888 points.

From my testing the 13900k has no IPC increase over alderlake.

 

[humbug]

From my testing the 13900k has no IPC increase over alderlake.

From your testing?

It has more per core L2 cache, that should amount to something.
----------

At this point, with all this now said and done, i still think Raptorlake will have an edge in performance, all be it a small one, both MT and ST.

AMD should still have an edge in efficiency, please don't bencher.

Gaming, no idea, Zen 4 also gets a doubling of L2, which should help, the clocks are very much higher than Zen 3, and there should be a boost from fast DDR5, put all that together and it looks interesting.

Once Zen 4X3d land tho, those will be 'thie' gaming chips.

 

[Joxeon]

That is the thing, My 5950x also does 4.1Ghz out the box not 3.9Ghz and it does it at 102watt. If you take 4.1Ghz then you get to 5.15Ghz with that 25% efficiency at the same wattage then.

Even the 3.9Ghz taking a 25% efficiency then you would be looking at 4.9Ghz at 25% window. Of course it wont exactly scale like this. However you have the additional power increase from 105watt to 170watt is a 62% increase. Or for my particular example since it sits at 102watt would be 66% window.

Suggesting that out the box it can pull 170watt whilst being 25% more efficient I still expect a 5Ghz all core possible compared to the previous gen.

The thing with efficiency though is it doesn't scale linearly with clock speed.

 

[g67575]

Thats a good point, it's the reason why base clocks matter (in terms of efficiency).

 

[humbug]

Thats a good point, it's the reason why base clocks matter (in terms of efficiency).

Unless the performance per watt is cited.