because Anand did not disclose the clockspeeds I had to use the TPU clockspeed results for an all core workload to get some all core figures. Ended up with 5.3 for Zen 4 and 4.4 for Zen 3. I then took the score / clock to get a score / Ghz, did the delta between Zen 4 and Zen 3 and then took the geomean of those deltas to get 13.7% ipc bump which is close enough given my clockspeeds were probably not 100% on the money.
Hint: clocks should be roughly the same with Zen 5. What does that tell you?Thanks for the info. I have done calculation about Zen4's improvement. SPECint did show 37% better performance compared to Zen 3. I am lazy to do delta comparison, I take your 13.7% IPC figures.
That's mean upcoming Zen5 with rumored 40% is most likely performance numbers not IPC. I know MLID is a joke today; he earned it I guess...Anyway, IPC of Zen5 should be below 20% as calculated...Someone needs to re-think what he claimed
View attachment 96419
Hint: TDP bump to more than 200W, what does that tell you?Hint: clocks should be roughly the same with Zen 5. What does that tell you?
Zen 3 had higher clocks as well.Zen4 improved ST performance respect to Zen3 a lot because of IPC+clock gains
Zen3 to Zen2 jump was almost pure architectural (IPC) gains
Zen5 will associate the biggest architecture overhaul in the Zen family to date with (probably) some clock increase but IMO likely something more akin to the Zen3 bump.
The TDP of Zen 5 is unchanged. What are you talking about?Hint: TDP bump to more than 200W, what does that tell you?
Intel accepts your challenge.You can’t achieve a 40% performance gain with a sub-20% IPC gain because the thing would need to be clocked at an obscene 6.5-6.6 GHz.
Not even a gigajuiced 14900KS with Intel's ultra-high clocking I7 node is able to achieve that.
My source told me more than Zen 4. I remembered someone mentioned 220W before, go figureZen 3 had higher clocks as well.
The TDP of Zen 5 is unchanged. What are you talking about?
Considering this could have been avoided by just... doing some basic sanity checks on the origin of his "source", I have no pity for him. He had every opportunity to do so. He backed down instantly after not getting a response on the e-mail sender's name, he was told the slide was widespread but clearly didn't check any of his other "sources" (or alternatively he just found out all of his other sources are BS-ing him, which might be even funnier). And to make things worse, he even went and lied about where he got the information from, pretending the 17% IPC claim and talk about fudging the Zen 5 IPC figures as coming from two different sources when in reality, they both came from the same source.I only now catch up to this thread. Yesterday I got to bed on 342 page, and now I am on 348, LMAO.
All I can say is: "WOW". Poor MLID XD.
Yep, but not so much, that's why I said "almost purely".Zen 3 had higher clocks as well.
I mean, why would he do that? He keeps getting quoted even in the neighboring threads, and that's supposedly a hw enthusiast forum.Considering this could have been avoided by just... doing some basic sanity checks on the origin of his "source", I have no pity for him. He had every opportunity to do so.
To be honest, MLID is just example of how Journalism in todays world works.I mean, why would he do that? He keeps getting quoted even in the neighboring threads, and that's supposedly a hw enthusiast forum.
You can only imagine how little one needs to do or understand to be considered trustworthy everywhere else on the internet.
Nah, the rumor mill presented real killer IPC gains for Zen 4 back in the days. Nobody expected a ~11% IPC gain with a much steeper frequency gain. Relevant MLID stated: "above 20% IPC increase over Zen 3".I love this, however Zen 4 does have killer IPC compared to the competition
Noice.That whole AMD China layoff thingie was a reorg after RDNA3 went oopsie.
Lots of internal changes have been made progressively based on performance and constant reviewing.
Noice.CPU ppl were sent to make clocks go big, and they did, and then they tried way harder and nearly pulled it off, just half a gen late. Oh and recent patents point to more CPU-esque scheduling mechanisms and other funny things coming to future GPUs.
It's a funny quirk at AMD, how they'll be ok with "cheapest is best" for as long as possible, but a partner jumps in (MS/Meta for AI, Sony for RT) and they'll be the best little soldier, stand up and do everything for them.Tell M$ that, AMD is only doing what Microsoft is asking of them, like putting Pluton in Rembrandt to get... nothing in return.
MS's PC division is just completely unhinged and out of touch, but getting money from them is usually worth the hassle, and Strix and X Elite are the only premium AI SoC's for some time.
Yeah indeed he has real info.
Same thing I found in the patches on the 1x ALU sched + 1x AGU sched
I am inclined to believe the ~20% IPC on his slide with few % (+3% or +4% ??) clocks
However I am not sure even partners knows exactly the specific IPC number that will be claimed at launch, pretty easy to block some prefetchers or perf enhancing chicken bits with firmware too.
he put out real slides. The slides has been corroborated with patches.
I don't see that from anyone. Do you think he has wrong slides?
Im sorry, this is my last post, on this topic.Indeed.
Kepler has to redeem himself this time. His poor RNDA3 showing still taints his reputation or at least people's perception of info he put out.
Took an L on X like a champ though.
I could not care less about it, I prefer technical discussions based on concrete information from patches patents manuals etc but this thread now is about as far from anything technical as possible.Sorry, Dis, you've been duped, and now you can see your own personal bias
Here - true, and agreed.I could not care less about it, I prefer technical discussions based on concrete information from patches patents manuals etc but this thread now is about as far from anything technical as possible.
Things were not like this before.
I come to see the forums but there are pages and pages of irrelevant stuffs
If that was me here that was a joke.My source told me more than Zen 4. I remembered someone mentioned 220W before, go figure
Oh OK. So when you say AMD's competition, you are strictly talking about Intel, and not Apple or ARM.Basically on par with Raptor Lake (maybe a few % lower) while having used less resources. Intel current Cove cores are so bloated it's not even funny.