Zen 3 reportedly contains a hefty single-threaded IPC boost and higher clock speeds

It is for a refresh yes.

It's not refresh, AMD has multiple times said it is a new architecture. Zen 4 will be built upon the Zen 3 architecture, so Zen 4 will more be a refresh than Zen 3 will be.

AMD has also said that the Zen 3 will be as big of a step as Zen--> Zen 2 was, NOT Zen -->Zen+. So a 10-15% IPC increase is small (A frequency bump is not counted towards the IPC increase).
 
It's not refresh, AMD has multiple times said it is a new architecture. Zen 4 will be built upon the Zen 3 architecture, so Zen 4 will more be a refresh than Zen 3 will be.

AMD has also said that the Zen 3 will be as big of a step as Zen--> Zen 2 was, NOT Zen -->Zen+. So a 10-15% IPC increase is small (A frequency bump is not counted towards the IPC increase).
Zen 2's IPC gains were 5-15% over Zen1 depending on workload. In contrast, the last Intel IPC gains on a mainstream desktop platform were in 2015 and were in the region of 2-5%, as was the one before that.

Almost all of Zen+'s gains over Zen1 were clock speed(Some v small gains from cache/uncore tweaks), any other gains Intel has made since then are clock speed, the rest of Zen2's gains over Zen+ besides that 5-15% were clock speed, IPC gains at all seem to be a rare treat nowadays.
 
Last edited:
Both AdoredTV and The Good Old Gamer were claiming 20% as a low-ball estimate in terms of IPC, and then a clock speed boost on top of that. They based this off of their interpretation of an interview with an AMD staff member. They seemed convinced. I think that's why a 10-15% IPC boost seems disappointing to them, because it's less than what they had decided in their heads Zen 3 would be.

Personally I'd say that 15% alongside a boost in clock speeds is excellent, especially when you consider the latency reduction Zen 3 is supposed to introduce. That would put AMD ahead of Intel in terms IPC as well as TDP and core count, all with higher income per unit sold. And their latency numbers would be comparable too. Zen 4 would expand on that. Even if Intel massively boosted their performance by 2021, AMD should be able to match them. And even if Intel sneaks ahead again, it'll only be in certain areas. AMD has garnered so many new fans and has developed such a cost-effective infrastructure that it may not matter if Intel are ahead in certain areas.
 
AMD don't need more IPC. It's not IPC keeping them from winning everything, it's clock speed. Getting down to the nm they are on they can fit lots of cores under a small die. Now comes time to get the clocks up to match Intel.

However, IPC is nice. It's not that keeping them from winning the gaming crown though, IMO.

Some games still get off on very high clocks.

That said, 15% as noted is way better than Intel have managed in an absolute age. So why not?
 
It's not refresh, AMD has multiple times said it is a new architecture. Zen 4 will be built upon the Zen 3 architecture, so Zen 4 will more be a refresh than Zen 3 will be.

AMD has also said that the Zen 3 will be as big of a step as Zen--> Zen 2 was, NOT Zen -->Zen+. So a 10-15% IPC increase is small (A frequency bump is not counted towards the IPC increase).

Think you need to come back to the real world and get your head straight.
 
One thing I will say, this will likely seem like a small gain in comparison to whatever Intel's IPC gains are likely to be from their next desktop architecture, which could be in the 15-25% range, BUT that has to be taken in the context of the fact the last IPC gains for Intel on desktop were with the introduction of Skylake in 2015, half a decade ago, and is likely going to come with at least a few % drop in clocks compared to these >5Ghz highs of 14nm+++.
 
Last edited:
It's not refresh, AMD has multiple times said it is a new architecture. Zen 4 will be built upon the Zen 3 architecture, so Zen 4 will more be a refresh than Zen 3 will be.

AMD has also said that the Zen 3 will be as big of a step as Zen--> Zen 2 was, NOT Zen -->Zen+. So a 10-15% IPC increase is small (A frequency bump is not counted towards the IPC increase).
Zen 4 is suppose to ship with DDR5 that alone is a huge jump in performance.
 
It's not refresh, AMD has multiple times said it is a new architecture. Zen 4 will be built upon the Zen 3 architecture, so Zen 4 will more be a refresh than Zen 3 will be.

AMD has also said that the Zen 3 will be as big of a step as Zen--> Zen 2 was, NOT Zen -->Zen+. So a 10-15% IPC increase is small (A frequency bump is not counted towards the IPC increase).

Remember that Zen 2 also has a process shrink. Process shrinks increase transistor budgets, and that is what boosts performance.

Zen 3 doesn't have that advantage, so 10-15% plus a clock boost is big. This will be much better than Zen-to-Zen+, but I won't be as big as Zen-to-Zen 2.

As always, IPC boosts are always different depending on your workload. AMD has no doubt fixed some edge cases that have so far harmed Ryzen, so some applications will get a larger boost. Just look at the difference between 32-core zen+ Threadripper compared to 32-core Zen 2 Threadripper.

Also, note that these rumours do not talk about multi-threaded performance. AMD's cache changes should have a big impact here. Having no big CCX jumps for single-die chips is a great thing for AMD. Like Zen 2, Zen 3's multi-threaded IPC boost could be larger than its single-threaded boost.

Consider that Intel is still on 14nm and can't get a big single-threaded performance boost. For AMD, 10-15% single-threaded on the same node generation is a big deal.

Zen 2's IPC gains were 5-15% over Zen1 depending on workload. In contrast, the last Intel IPC gains on a mainstream desktop platform were in 2015 and were in the region of 2-5%, as was the one before that.

Almost all of Zen+'s gains over Zen1 were clock speed(Some v small gains from cache/uncore tweaks), any other gains Intel has made since then are clock speed, the rest of Zen2's gains over Zen+ besides that 5-15% were clock speed, IPC gains at all seem to be a rare treat nowadays.

Exactly. It's also worth noting that after Zen 2, a lot of the low lying fruit for Zen architectural improvements is gone. AMD now has to work a lot harder for its IPC gains.

One thing I will say, this will likely seem like a small gain in comparison to whatever Intel's IPC gains are likely to be from their next desktop architecture, which could be in the 15-25% range, BUT that has to be taken in the context of the fact the last IPC gains for Intel on desktop were with the introduction of Skylake in 2015, half a decade ago, and is likely going to come with at least a few % drop in clocks compared to these >5Ghz highs of 14nm+++.

Well, Intel has had literal years to prepare for this. They need a big leap after Skylake, and their engineers haven't been sitting on their hands this past five years.

Manufacturing screwed Intel over, but there's a reason why Ice Lake and Tiger lake promise big IPC gains. The problem right now is that 10nm is broken, and you need both a good process and a good architecture to make a CPU sing.

Rumour has it that the Tiger Lake "Willow Cove" core architecture is getting backported to 14nm. Seems like the increased transistor budget per core will reduce Intel from 10 cores with Comet Lake desktop to 8 cores with Rocket Lake. Not sure if these reports are true, but I've heard rumours of this for a while.

TBH, I heard that Intel was working on 10-core LGA 1151 processors since Intel released 6-core Coffee Lake. These things take years to sort out.
 
AMD don't need more IPC. It's not IPC keeping them from winning everything, it's clock speed. Getting down to the nm they are on they can fit lots of cores under a small die. Now comes time to get the clocks up to match Intel.

However, IPC is nice. It's not that keeping them from winning the gaming crown though, IMO.

Some games still get off on very high clocks.

That said, 15% as noted is way better than Intel have managed in an absolute age. So why not?

Not clock speed. Real-time tasks are their biggest problem. It doesn't matter if they are single or multi-threaded. Games, as the most common ones, are a good example of a real-time task. Music production is even better examle of a heavy CPU and memory real-time task. Multiple CPU dies have many advantages, but the biggest negative is a very large latency between various components. They ironed it out a bit, but that is the reason pretty much everyone was noticing that editing on the Intel system feels snappier compared to earlier Ryzens.

Tiled rendering tasks are easier to handle. Throw more cores at them and they'll love it. More cores = more performance (if the cores are at the same ballpark). That is what AMD has been doing.

If they can solve real-time performance Intel would be in a pickle.

See this graph. It is not even a contest.

Dawbench-VI-Chart-2019Q3-2.jpg
 
AdoredTV just makes stuff up. I wouldn't take anything they say seriously unless it's also reported elsewhere. In which case, just get your rumors elsewhere.
 
See this graph. It is not even a contest.


yeah well i only trust the statistics i faked myself. :)


a more recent test:




hmsZpxp.jpg






The 10980XE holds its crown in the VI test but not by a huge margin, whilst falling behind in the DSP test, meaning that if the chips were more equally priced then this would be far too tough to call. However, even with Intels price adjustments over recent months the 10980XE is sat around the £1250 price point at the time of writting and looking at the chart and specifically that 3950X in comparison which is sat at less than £700, it becomes clear that whilst the 10980XE is a perfectly fine chip in isolation it really should be £300 cheaper than it currently is in order to make sense in a value comparison with the rest of the market.
 
Last edited:
he was spot on many times.

To be fair from what I've seen they seemed to get quite a lot wrong about Navi and Zen2, sure they got some things right but the hit rate isn't too impressive given how much they pump out. While they do have some value I'd always caution on applying a critical eye over their predictions (Which Mark does well).
 
Last edited:
AMD don't need more IPC. It's not IPC keeping them from winning everything, it's clock speed. Getting down to the nm they are on they can fit lots of cores under a small die. Now comes time to get the clocks up to match Intel.

However, IPC is nice. It's not that keeping them from winning the gaming crown though, IMO.

Some games still get off on very high clocks.

That said, 15% as noted is way better than Intel have managed in an absolute age. So why not?

I have a feeling that AMD aren't chasing big clock speeds with Zen 3. Even if they are, it'll still trail behind Intel. But that doesn't bother me personally, especially when Intel actually release a new architecture on a new node and have to reduce clocks. Anyway, I don't see a huge distinction between IPC and clock speed. They both serve similar purposes, and in some cases virtually the exact same. If AMD could compete in the music production space and in gaming, or even beat Intel, folks would have no reason not to go with AMD in almost any market. An improvement in latency and IPC can help with music production and gaming as much as or more than clock speed.

To be fair from what I've seen they seemed to get quite a lot wrong about Navi and Zen2, sure they got some things right but the hit rate isn't too impressive given how much they pump out. While they do have some value I'd always caution on applying a critical eye over their predictions (Which Mark does well).

Funnily enough, it makes sense that as his channel and notoriety grows he might feel like his leaks and sources woudl become more accurate, but I'd wager they'd become less accurate. I do like his content, and the same goes for The Good Old Gamer and a few of the other guys, but their leaks are often inaccurate (yet are claimed to be reliable), and their interruptions of them I don't always agree with. Sometimes I think they're a little AMD-favoured. Which is fine and dandy as long as it's recognised. I wanna see AMD succeed more than Intel and Nvidia. And if I were in their shoes making predictions and interruptions, and earning money for it in the process, I'd be one-sided at times too.
 
If they can solve real-time performance Intel would be in a pickle.

See this graph. It is not even a contest.

So you refer to the only thing Intel has going for it now? and you don't think they are already in a gallon of Branston? :D

You mean if AMD catch them on that too then they are totally screwed. Beyond Piccalilli.
 
Back
Top