Ok, in the specint suite it does seem that it's closer to 25%. Thanks for providing the benchmarks.I don't know, really looks like >= 1.25x is a more accurate guess than 1.20x.
AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested
www.anandtech.com
Hey man! Howz everything? Howz the wife and kids? Your cat/dog/monkey/parrot? Howz life treating ya?Cinememe is a single data point and not really how anyone measures 1t/nT perf even in projections.
Outside of 22 S-300 missile hits early in the morning, okay.Howz everything?
Wrote some nasty things on a pack of ER Grad missiles currently hitting Belgorod. Thanks.Howz life treating ya?
Yes aka the relevant metric.Ok, in the specint suite it does seem that it's closer to 25%. Thanks for providing the benchmarks
Going by this chart: https://www.guru3d.com/review/intel...ons-per-cycle-ipc-clock-for-clock-at-3500-mhzMy point just being that Zen2 to Zen3 is not 25% or more in ST on average, 20% is still mighty impressive especially when you compare to the competition.
Can't tell if serious or not.Outside of 22 S-300 missile hits early in the morning, okay.
Wrote some nasty things on a pack of ER Grad missiles currently hitting Belgorod. Thanks.
He's from Ukraine, but not sure if all the details are correct. He went dead silent on Twitter for months shortly after the war started. Thought he was dead back then. Suddenly he was back in September 2022 (I think). Right in time to hype RDNA3.Can't tell if serious or not.
Mid-August really.Suddenly he was back in September 2022 (I think).
No, I just talk like all old MMO players do.Habit he must've picked up sending messages on the military communication equipment.
nT should be 15-20% iso cc.It will be very interesting to see how Zen5 perf/W will end up like
Yes, I have high hope of Strix Point as well. STX comes with slightly better process but much bigger die area (from 178mm2 to 225mm2; most of them are from doubling of L3 cache). The delays of launching are mostly due to AMD and Qualcomm are aligned their SoC with Microsoft's WIndows 12. This round AMD should have more stock availability cause OEMs are most likely demand it.....due to lack of MTL's AI requirement, we shall see..It will be very interesting to see how Zen5 perf/W will end up like, considering It uses N4P. That's not much better than what PHX used.
All in all, I consider Strix Point the best future Zen5 product.
Why? Because It has the highest CPU gain compared to predecessor.
Strix Halo will be powerful, but along with Fire Range It won't increase core count.
At least that IGP will be capable, but considering price will be likely high, won't have a good bang for the buck.
Is it weird that I'm not? I expect nice core improvements, but with the IO system staying the same, I think most complex, interesting problems will see somewhat muted gains. I expect more from Zen6.I'm as pumped for Zen 5 as @SiliconFly is for Crater Lake. Except I'm hoping that AMD will not disappoint me as bad.
I don't think it's weird at all. Personally I see a clear line of Zen 1 and 3 mostly setting promises which then were mostly fulfilled by package/platform changes in Zen 2 and 4. As a result I'm really looking forward to the Zen 5 core and deep dives of it, but on the balance of the whole package/platform I'm seeing myself waiting for and buying into Zen 6.Is it weird that I'm not? I expect nice core improvements, but with the IO system staying the same, I think most complex, interesting problems will see somewhat muted gains. I expect more from Zen6.
225 mm²?Yes, I have high hope of Strix Point as well. STX comes with slightly better process but much bigger die area (from 178mm2 to 225mm2; most of them are from doubling of L3 cache). The delays of launching are mostly due to AMD and Qualcomm are aligned their SoC with Microsoft's WIndows 12. This round AMD should have more stock availability cause OEMs are most likely demand it.....due to lack of MTL's AI requirement, we shall see..
MLID said Zen 6 is moving to an entirely new chiplet architecture, with chiplets directly stacked on top of the IO Die.I don't think it's weird at all. Personally I see a clear line of Zen 1 and 3 mostly setting promises which then were mostly fulfilled by package/platform changes in Zen 2 and 4. As a result I'm really looking forward to the Zen 5 core and deep dives of it, but on the balance of the whole package/platform I'm seeing myself waiting for and buying into Zen 6.
Define "not as good"MLID said Zen 6 is moving to an entirely new chiplet architecture, with chiplets directly stacked on top of the IO Die.
That doesn't sound unrealistic. The current packaging technology AMD uses is not as good as Intel's Foveros in MTL and ARL. It would make sense they'd move to a superior solution.
I think he meant not as power efficient. On the other hand, It should be cheaper.Define "not as good"
Yes. Also, I believe Foveros has better latency, and hence performance implications from that?I think he meant not as power efficient. On the other hand, It should be cheaper.
Why are you quoting clueless morons.MLID said Zen 6 is moving to an entirely new chiplet architecture, with chiplets directly stacked on top of the IO Die.
Passive 2.5d slabs aren't anything interesting.The current packaging technology AMD uses is not as good as Intel's Foveros in MTL and ARL
2.5D substrates have no impact on that overall.Also, I believe Foveros has better latency
It's not rumoured, it's what it is.Strix Point is rumoured to have a setup of:
4 × Zen 5
8 × Zen 5C
Should be upper teens higher at 30W.What would be the MT performance differencs between such a 4+8 setup vs a pure 8 core Zen 5 setup?
Assuming Zen 5c has the same IPC as Zen 5, the MT uplift would be dependent on how high the Zen 5c cores are clocked. 4 more cores gives you +50% MT if clocks were equal. If Zen 5c is clocked 33% lower than the standard Zen 5 cores, then that's a 16.7% MT uplift.Strix Point is rumoured to have a setup of:
4 × Zen 5
8 × Zen 5C
What would be the MT performance differencs between such a 4+8 setup vs a pure 8 core Zen 5 setup?
This is something I've struggled to understand. How can a core with halved L3 have the same IPC? Has L3 suddenly become irrelevant?Assuming Zen 5c has the same IPC as Zen 5, the MT uplift would be dependent on how high the Zen 5c cores are clocked. 4 more cores gives you +50% MT if clocks were equal. If Zen 5c is clocked 33% lower than the standard Zen 5 cores, then that's a 16.7% MT uplift.