- Mar 3, 2017
- 1,659
- 6,101
- 136
"a new rumor suggests that AMD is hastening the launch of next-gen Zen 4-based Ryzen 7000 series processors which might launch in April"
Back at CES 2022, AMD said that the upcoming 5nm-based Zen 4 architecture will launch in the second half of the year (2H 2022)
Ok, so there was some rumor. But how does that answer the questions in my previous post?I don't claim to know anything about Z5, because I don't know anything about it.
In light of speculation of early Z5 release in this thread, there's this quote from a Z4 article from February 2022:
It also says this though:
Irrespective of what you find, you do know that this is very weak reasoning?Ok, so there was some rumor. But how does that answer the questions in my previous post?
The rumors go that it was intended for April but delayed until September due to new platform teething issues. The same delay shouldn't occur to Zen 5 since AM5 is established and beginning to mature now.What I was getting at was that Zen4 (and Geforce 40 series as well) lined up very much at 24 months after the previous generations. Going into the summer, most assumed the next generations would arrive at the same time. AMD have been more successful lately than for example NVidia at keeping details secret until late.
I find it funny that there were the same April rumors as now though.
What is ”this” that you mean is very weak reasoning?Irrespective of what you find, you do know that this is very weak reasoning?
Trying to forecast initial sales.What is ”this” that you mean is very weak reasoning?
Even if a pattern can be found based on past track record, looking at time from announcement until actual availability? Sure it’s not certain, but do we have anything better to base the estimates on at this point?Trying to forecast initial sales.
4th January 2022: AMD showed the IHS of ZEN4 and ZEN4 running Halo with 5.5GHz clocks on All Cores.Wasn't there any teaser info published earlier either? Or dead silent until 1 month before availability in stores?
Imagine if the hype train flops and both Zen 5 and Arrow Lake(P core/Lion Cove) isn't performing up to standard lol4th January 2022: AMD showed the IHS of ZEN4 and ZEN4 running Halo with 5.5GHz clocks on All Cores.
14th April 2022: Something about AMD RAMP (what is now called EXPO), promising high DDR5 clocks.
23th May 2022: Computex, the Event where AMD broke the Hypetrain by claiming >15% singlethread leap with combined clocks and IPC. They already showed Blender Benchmarks with 7950X destroying 12900K.
9th June 2022: Financial Analyst Day, IPC of ZEN4 8-10%, most of performance gain comes from clocks.
30th August 2022: Final Presentation with the usual 1st Party Benchmarks, Specs and Prices.
27th September, Release of the CPUs, to destroy Intels Raptor Lake presentation Happening on the same day.
As you can see, there was plenty of information about ZEN4 months before Release, which is why I always doubted any April rumors. Even the whole "osborning old products" Story doesn't make sense. If ZEN5 is so good singlethread, they could have still shown the Blender run with the Multi Perfomance and let the singlethread bomb pop at real presentation.
Or show ZEN5 running games or whatever. Instead they are dead silent. Maybe they really ran into problems and so far it's not perfoming up to standard.
This is not the intel thread but is there any "hype" around ARL. Whatever I could see in social media seem to be for Lunar instead. Only interesting thing is mobile which would be 1st 20A chip and 1st chip using GAA.Imagine if the hype train flops and both Zen 5 and Arrow Lake(P core/Lion Cove) isn't performing up to standard lol
Given the extreme variance in rumors over perf it is certainly possible.Or show ZEN5 running games or whatever. Instead they are dead silent. Maybe they really ran into problems and so far it's not perfoming up to standard.
Perhaps less X Elite than what it represents from a dramatic uptick in WoA products on the market (17+?), and that's just for starters.I think X Elite could make both companies more nervous than anything else.
I guess Zen5 without Vcache is not any better than Zen4 with Vcache in gaming, that's why AMD silent.Or show ZEN5 running games or whatever. Instead they are dead silent. Maybe they really ran into problems and so far it's not perfoming up to standard.
Or they already have the gaming crown and don't expect to lose it, so why spoil the reveal hype?I guess Zen5 without Vcache is not any better than Zen4 with Vcache in gaming, that's why AMD silent.
Major changes always bring much more risk than incremental ones. It's a question of risk management. The Zen 5 development seems to be less turbulent than Bulldozer times (released 4 years after the first failed 45nm attempt).After Bulldozer I think everyone is more than a little afraid of AMD going radio silent on us before a major µArch release.
But then again Zen3 did come upon us in a much more all at once manner than Zen4, so it's possible that market factors are the cause of their reticence rather than anything else (or market factors that caused them to hype up Zen4 before release).
Bear in mind that they only just launched Zen4 based Hawk Point, and probably don't want to poison sales from that by announcing Zen5 based Strix Point which would hypothetically curbstomp it in every metric.
Arrow should come with a new core after the series of refreshes of Alder-Raptor-RaptorR-Meteor-(Bartlett?). The core is reported to be massive as both Sunny and Golden cores were compared to their competitors. However, the negative sentiment was brought by the Igor's Lab leak in summer 2023. Word on the street is the core doesn't clock as high as Raptor does.This is not the intel thread but is there any "hype" around ARL. Whatever I could see in social media seem to be for Lunar instead. Only interesting thing is mobile which would be 1st 20A chip and 1st chip using GAA.
If ARL is clocked lower, has no HT and also suffers from higher RAM latency and still manages to beat RPL by 5 to 10 percent, I guess that's not a total dud. Disappointing yes but still somewhat respectable if they manage to overcome all that to come out ahead of at least their own previous product.Word on the street is the core doesn't clock as high as Raptor does.
Lower clocks are to be expected compared to Raptor-R - the question is how much lower. The RAM latency is probably gonna suffer simply because of using a disaggregated design. Absence of HT is still a bit murky given we still have only early samples of Arrow/Lunar.If ARL is clocked lower, has no HT and also suffers from higher RAM latency and still manages to beat RPL by 5 to 10 percent, I guess that's not a total dud. Disappointing yes but still somewhat respectable if they manage to overcome all that to come out ahead of at least their own previous product.
It is highly unlikely that the move to N4P has brought about clock regressions. There might be some on the design side, but due to the node? No.And if Zen5 has experienced some clock regression due to lack of adequate fab node improvements they are even less likely to rush to PR.
The architectural changes for sure we don't know. But it seems Zen 5 was heavily revamped and they added a lot of resources alreadyAnyways, is Zen 5 such a major change like the 10h->Bulldozer trasition was?
Could you give an estimated performance improvement in typical workloads for each of these changes?
- +2 rename/dispatch
- +2 ALUs
- +1 AGU
- +1 LD/cycle
- 512b FP width
- 64B LD/ST queues
- 48K L1D
- OOO structures increased
- New BP with larger BTBs
You also want french fries to that?Could you give an estimated performance improvement in typical workloads for each of these changes?
DisEnchantment has the necessary background to give a ballpark figure. That is my expectation based on the quality of his posts here.Context is key. Its impossible to estimate such thing because you have to look at big picture to know how things affect each other.