Alder Lake-S leaks

[humbug]

Yes, not sure how accurate CPU Monkey is but they have 5900HX scoring 13,875 in Cinebench R23.
https://www.cpu-monkey.com/en/compare_cpu-amd_ryzen_9_5900hx-vs-apple_m1

they also have the 5800U:

What is more impressive for the M1 is the single core results vs the 5800U:

Far more impressive than the CPU scores, are the M1's GPU scores.
In the AT review:
https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/3

The 5800U has a configurable TDP of 10 to 25 Watts, i think we have to assume that this score is at 25 Watts, which is extremely impressive, i think it also has to be said that Zen 3 has much better performance per watt at these lower power levels than it does at higher power levels, its at its most efficient at around 20 to 30 Watts.
A 5800X scores around 50% higher but that's with 400% higher power, that should tell you just how much the efficiency drops off at high power targets.

It is what it is and at low power level its damned impressive.

The M1 Mac Mini scores just short of 8,000, the Mac Mini draws 39 Watts, the CPU alone, i would take a guess at 20 Watts. https://en.wikipedia.org/wiki/Apple_M1

So
5800U, 8 cores: 11,203 at 25 Watts.
Apple M1, 8 cores: 7,833 at 20 Watts.

20 Watts / 25 Watts = 1.25, 25% lower power to the M1
11,203 / 7,833 = 1.43, 43% higher performance to the 5800U.

In this configuration Zen 3 has about 15% better performance per Watt.

Its a bit meaningless, but i think one thing we can be sure of is the Apple M1 and Zen 3 are extremely efficient CPU's, at least when you compare them to team blue.

The thing that really does need to be said tho is that the Apple M1 is on TSMC 5nm, Zen 3 is on TSMC 7nm.

 

[Grim5]

The M1X is coming within a week or two and will settle this debate I would imagine.
No more trying to extrapolate results because Intel and AMD dont make many 15w products.

It gets 12 CPU cores and 16 GPU cores @ 35w TDP.




Here is a 5900HX in a $4k gaming laptop. https://www.notebookcheck.net/Asus-...etal-7-nm-AMD-Zen-3-Is-Stunning.540483.0.html

Puts up 13.5k in R23 while pulling 80w.

While the M1X gets 15k and while we don't yet know exactly what it pulls, I doubt its going to hit 80w and basically guarantee the macbook pro's using the m1x are going to smash 5900HX laptops in battery life.

 

[jigger]

The M1X is coming within a week or two and will settle this debate I would imagine.
No more trying to extrapolate results because Intel and AMD dont make many 15w products.

It gets 12 CPU cores and 16 GPU cores @ 35w TDP.




Here is a 5900HX in a $4k gaming laptop. https://www.notebookcheck.net/Asus-...etal-7-nm-AMD-Zen-3-Is-Stunning.540483.0.html

Puts up 13.5k in R23 while pulling 80w.

While the M1X gets 15k and while we don't yet know exactly what it pulls, I doubt its going to hit 80w and basically guarantee the macbook pro's using the m1x are going to smash 5900HX laptops in battery life.

The M1X will likely beat the 5900HS in CPU performance but the strongest point is graphics performance. Apple will likely dominate the gaming laptop market with the M1X.

 

[Zarax]

The M1X will likely beat the 5900HS in CPU performance but the strongest point is graphics performance. Apple will likely dominate the gaming laptop market with the M1X.

You're kind of forgetting about compatibility though.

 

[jigger]

You're kind of forgetting about compatibility though.

Apple have done a lot of work on Windows compatibility.

 

[Zarax]

Apple have done a lot of work on Windows compatibility.

Hate to break it to you but... https://www.thurrott.com/windows/windows-11/256250/microsoft-says-no-to-woa-support-on-m1-macs

 

[LePhuronn]

Hate to break it to you but... https://www.thurrott.com/windows/windows-11/256250/microsoft-says-no-to-woa-support-on-m1-macs

Hate to break it to you but...that's Windows on ARM. Nobody cares. Except Parallels of course. And also, Microsoft said "it's not a supported scenario". That doesn't mean it's outright blocked.

Besides, x86 emulation on ARM is good, and Apple's implementation is stellar. So why bother with Windows on ARM at all? Just use the x86 version, especially as Windows on ARM is predominantly an x86 emulator anyway.

 

[jigger]

Hate to break it to you but... https://www.thurrott.com/windows/windows-11/256250/microsoft-says-no-to-woa-support-on-m1-macs

Maybe go and do some reading. It’s happening one way or three it seems. I hate to break it to you, but Apple is coming to drink our juice and probably rightly so. Laptops with 150watt graphics cards are ridiculous things to start with.

 

[alec]

Latency is crazy high but this is likely due to not being in gear 1 mode.

DDR5-6400 CL40 latency should be equivalent of DDR-3200 CL20 which is a terrible timing set. But that alone doesn't explain 92ns. Can't see if it is in gear2, not familiar with new intel stuff. CR2 and memory bus 3200 point to gear1 I think.

Not impressed at all. Some manufacturer better come up with B-die equivalent soon, to push timings down on enthusiast kits.

 

[Zarax]

Maybe go and do some reading. It’s happening one way or three it seems. I hate to break it to you, but Apple is coming to drink our juice and probably rightly so. Laptops with 150watt graphics cards are ridiculous things to start with.

Wake me up when Apple will start making $500 laptops, until them Chromebooks are a more serious threat to Windows than anything made by Apple.
Even Cupertino's coffers cannot afford to subsidize the porting of enough games to make a difference.

 

[LePhuronn]

Not impressed at all. Some manufacturer better come up with B-die equivalent soon, to push timings down on enthusiast kits.

The first generation of new memory is always rubbish compared the the latest generation of the outgoing tech. Always. yet things mature rapidly, so it won't be long before decent DDR5 kits show up.

 

[LePhuronn]

Even Cupertino's coffers cannot afford to subsidize the porting of enough games to make a difference.

But you don't need to port anything when you're running x86 Windows.

 

[KompuKare]

Maybe go and do some reading. It’s happening one way or three it seems. I hate to break it to you, but Apple is coming to drink our juice and probably rightly so. Laptops with 150watt graphics cards are ridiculous things to start with.

Well, if they bake themselves after a few years and people buy new ones then the manufacturers are laughing. When Nvidia sold all those solder defect parts back when the transition to lead-free solder was on (2006+, aka bumbgate), the laptops died earlier than desktop cards like the 8800GT etc.

That Anandtech chart with the Rise of the Tombraider bench is all the more impressive if you consider the M1's performance which vastly out-performs all other iGPUs is running that game via Rosseta. Although as the GFXBench 5.0 results show, native and Rosseta are almost identical:

The Rosetta team must have spend ages getting not only normal apps working but games too and it also looks like the iGPU is the bottleneck as since the calls are "simply" re-directed to the native driver. If the game was bottlenecked by the CPU the two scores should be further apart.

 

[Grim5]

ASUS Z690 lineup leaked, reveals that boards will not support both DDR4 and DDR5, DDR5 is exclusive to Ultra high end.

Alder Lake supports both DDR4 and DDR5, but DDR5 slots will be found on Ultra high end boards and everything else this generation will stick with DDR4, so be prepared to cough up lots of cash if you want DDR5.

DDR5 ASUS Z690 boards:

ROG MAXIMUS Z690 EXTREME
ROG MAXIMUS Z690 FORMULA
ROG MAXIMUS Z690 HERO
PROART ProART Z690-CREATOR 10G

DDR4 ASUS Z690 boards:

ROG STRIX Z690-A GAMING D4
ROG STRIX Z690-E GAMING D4
ROG STRIX Z690-F GAMING D4,
TUF GAMING – TUF GAMING Z690-PLUS D4
PRIME PRIME Z690-A PRIME Z690-A D4
PRIME Z690M-PLUS D4
PRIME Z690-P PRIME Z690-P D4
PRIME Z690-V PRIME Z690-V D4
PRIME Z690-V-SI-D4

 

[pcuser]

+15% for Zen 3D (games and some other tasks only?). Zen 4 rumoured about 20% IPC increase so more perf with higher clock speed.

Not difficult to believe when Intel's chips will supposedly be 100% quicker than current stuff alone just <4 years from now.

Those numbers could be leaning on the optimistic side at least in my understanding /up to 15% in games, I do not know whether those rumoured 20% will be compared to current 5th generation or 3D improvement and so on/.

 

[LePhuronn]

Those numbers could be leaning on the optimistic side at least in my understanding /up to 15% in games, I do not know whether those rumoured 20% will be compared to current 5th generation or 3D improvement and so on/.

This 15% boost from the cache isn't optimistic, it was actually demonstrated. It was an average across a number of games using a 12 core prototype. Other workloads may see a bigger boost, others may get a smaller one, others still may see no benefit.

The 20% uplift is the cores themselves, the cache is separate. So rough comparison would be Zen 4 cores = Zen 3 cores + 20%. If vanilla Zen 3 gains roughly 15% from the stacked cache, then Zen 4 would be Zen 3 cores + 20% + an extra boost from the stacked cache (assuming Zen 4 uses stacked cache, and why wouldn't it?).

The talk of 20% IPC with Zen 4 predates stacked cache by quite a while, so it's possible the numbers might change because cache can affect the amount of "stuff" a CPU can do per clock.

 

[pcuser]

This 15% boost from the cache isn't optimistic, it was actually demonstrated. It was an average across a number of games using a 12 core prototype. Other workloads may see a bigger boost, others may get a smaller one, others still may see no benefit.

The 20% uplift is the cores themselves, the cache is separate. So rough comparison would be Zen 4 cores = Zen 3 cores + 20%. If vanilla Zen 3 gains roughly 15% from the stacked cache, then Zen 4 would be Zen 3 cores + 20% + an extra boost from the stacked cache (assuming Zen 4 uses stacked cache, and why wouldn't it?).

The talk of 20% IPC with Zen 4 predates stacked cache by quite a while, so it's possible the numbers might change because cache can affect the amount of "stuff" a CPU can do per clock.

Up to 15% in games may be very similar but in my mind is not the same as 15%. Though games are the most used applications there is other usage as well. That is why I call it optimistic I am not questioning whether or not that number has been achieved. In my mind we have yet to see it in real life first before we can count on whatever percentage we are going to call it. Also stacking the 20% rumoured increase on top of these 15% is optimistic in my mind since we do not know what results we are going to get in real life. It may be similar but we are speculating with increased level of error margin in this case as per my understanding. I'll be happy to see even better results but am a little cautious without seeing any tests first.

 

[humbug]

Up to 15% in games may be very similar but in my mind is not the same as 15%. Though games are the most used applications there is other usage as well. That is why I call it optimistic I am not questioning whether or not that number has been achieved. In my mind we have yet to see it in real life first before we can count on whatever percentage we are going to call it. Also stacking the 20% rumoured increase on top of these 15% is optimistic in my mind since we do not know what results we are going to get in real life. It may be similar but we are speculating with increased level of error margin in this case as per my understanding. I'll be happy to see even better results but am a little cautious without seeing any tests first.

"Average" 15%, up to was 25% according to their charts.

AMD's recent history with claims.

Zen 1 over Excavator: Claim +40% Reality +52%
Zen 1+ over Zen 1: Claim +10% Reality +10%
Zen 2 over Zen 1+: Claim +20% Reality +20%
Zen 3 over Zen 2: Claim +20% Reality up to +50%

 

[humbug]

here is the chart.

5900X @ 4Ghz vs 5900X 3D Prototype @4Ghz

 

[LePhuronn]

Up to 15% in games may be very similar but in my mind is not the same as 15%.

It wasn't "up to", it was explicitly presented as "average".
https://videocardz.com/newz/amd-dem...rototype-with-3d-v-cache-stack-chiplet-design

Also stacking the 20% rumoured increase on top of these 15% is optimistic in my mind

Yes, all rumours should be taken with a grain of salt, but like I said earlier, the 20% IPC uplift was talked about before 3D stacked cache was an actual thing. I sincerely doubt AMD would even bother bringing stacked cache to desktop Zen 3 if its performance boost subsumes Zen 4's architectural improvements.

Specific numbers aside, I think it's fairly safe to say that whatever Zen 4's uplift is, it will be measured against 3D stacked Zen 3, not vanilla Zen 3.