AMD Ryzen 7000 + AM5 - Livestream

[Saracen999]

To be fair, the new is very shiny here (not that the 5000 series is a bad shout). Most of the upgraded interfaces are useless to someone who doesn't want to upgrade the machine later, but that ddr5 memory is an instant win.

Yeah, it is. Very.

Aw, hell. Years ago I umm'd and ahh'd (shocking, I know) about a new car. Do I go new M3, or used Ferrari? Nowadays, BTW, I wouldn't choose either but this was a while ago. The notion of the Ferrari did appeal, and I went as far as going to Maranello's for a chat and look/see. In the end, long-story-short, the Ferrari option was very shiny, but the M3 was shiny enough, a LOT cheaper (unless the Ferrari was rather old), far more reliable and likey to be hugely less hassle.

I feel a parallel.

Do I actually need even a 5900X / X570 / DDR4? Honestly, probably not. There's at least an element of "want" not need in it. Will 5900X / X570 / DDR4 be fast enough? Undoubtedly. But my logic is the 5900X etc is not that much more expensive than coming down a couple of notches (unlike the Ferrari to M3 saving), and my value v. cost sense says, "why not"?

But the 7000-series / DDR5 route? It'll no doubt up the cost yet again, the gains (subject to third-party independent testing) do look significant. I could indulge myself but, like I said, do I really even need the 5900X / DDR4 route? No. What worries me a bit more, and probably (today) stops me indulging in 7000 etc is the questionmark over the near-inevitable reliability issues over virtually any brand new tech, certainly of this leve of complexity. Even Intel had their issues with 12th Gen, and yeah, they'll get sorted but do I want to battle though that.

If I were doing this 12 months after actual launch, and the issues were nailed down, I might be tempted enough by the new and shiny but right now, I'm more concerned about avoiding potential hassle than even the extra cost.

There's another parallel. I remember, again some years back, debating the desirability of SSD's. The concensus was "much faster, inc booting, therefore better". My perspective was the boot speed is meaningless to me, because I'm not rebooting constantly, and turn PCs on at start-of-day, go make a cuppa, and by the time I get back, they're ready. Whether it took 2 minutes or 10 seconds doesn't matter (to me). Similarly, as much of my "work" then was in Word, running on HDs meant the HD was spinning idly about 95% of the time, whereas an SSD would be idle maybe 99.9% of it but either way, unless it could speed up my typing fingers, the "performance" difference was pretty moot.

I still feel that way, BTW, though the goalposts have moved. Much more of my "work" (hobby level) is sound and video and my PC? Yup, SSD all the way. But my main data storage is spinning rust in a NAS 'cos I'm not paying for 48TB of SSDs.

Truthfully, whether I'm going 500, 7000 or not at all depends on the day, time of day, whether I enjoyed lunch what the weather is doing, if there's a Y in the day, and however many tosses of a coin it takes to get the result I want in that moment. I'm capale of going "definitely 5000" to "definitely 7000" to "aw nuts, why bother" so fast it'd make your head spin. What I actually do probably depends on my mood once those independent tests are in, and on whether, when/if I decide to press "buy" a given option is actually in stock or not. Right now, I really couldn't call it for sure, but the smart move is the devil I know, the 5900X. Am I that smart, though?

[kalniel]

And here's me umming and ahhing about a Suzuki Vitara

[Saracen999]

Well, the Ferrari thing was a rather silly urge. I got the M3 though. However, it was in the ancient past - I think Victoria was still on the throne.

[CAT-THE-FIFTH]

https://wccftech.com/amd-ryzen-7000-...at-120w-rumor/

They seem to be matching Intel now!

[kompukare]

https://wccftech.com/amd-ryzen-7000-...at-120w-rumor/

They seem to be matching Intel now!

Looks like once the power consumption race has begun, everyone has to join in

Irony being that even taking into account the worse tickets from the silicon lottery, saving 50% of the power for 1-2% performance is often possible.

One leaker dropped the stock voltage and got the same clocks with almost half the wattage.
https://twitter.com/harukaze5719/sta...681984/photo/1



While I love the idea of undervolting, if the retail prices are set by the crazy clocks & voltage performance then that's not good.

[CAT-THE-FIFTH]

Looks like once the power consumption race has begun, everyone has to join in

Irony being that even taking into account the worse tickets from the silicon lottery, saving 50% of the power for 1-2% performance is often possible.

One leaker dropped the stock voltage and got the same clocks with almost half the wattage.
https://twitter.com/harukaze5719/sta...681984/photo/1



While I love the idea of undervolting, if the retail prices are set by the crazy clocks & voltage performance then that's not good.

AMD doing another R9 290X where they overvolted it to get a few percent extra performance at the cost of increased power consumption.

[watercooled]

I wouldn't take too much away from that screenshot. We don't know if the numbers are correct/accurate, whether this is actually stable, what else might be different, etc. And of course, whether these are actually retail CPUs or prerelease samples.

Modern CPUs are not set to a fixed voltage for all samples from the factory, because of silicon variance. Frequency and voltage profiles are far more dynamic and somewhat individual to processors and workload. Factory voltages (and therefore power) may be slightly higher than that which is minimally stable but this headroom is generally a good thing. AMD have already been doing this for years on Ryzen, and before.

But I would not expect to see such a drastic drop in power with zero performance impact on retail samples. It's not unusual to see desktop parts pushed well into diminishing returns in terms of clocks, but I kinda dismissed this image when I saw it. It just doesn't sit right with me for a number of reasons.

I don't think it is even stated which CPU this is? Not to mention we don't know where that clock speed is being measured? The cores on Ryzen CPUs are not locked to the same clock speed - we could well be seeing the measurement of one core at low boost while others are boosting much higher, vs the second where a fixed clock has all cores at the same? It could even be whacky prerelease BIOS and/or motherboard auto-OC nonsense interfering. Heck, there might even be cores disabled on the second picture! There are just too many unknowns IMO.

[watercooled]

I was checking Videocardz comments on Zen4 and this post looked interesting:
https://videocardz.com/newz/amd-ryze...ent-5964262258



Articles mentioned:
https://chipsandcheese.com/2021/07/2...s-bottlenecks/
https://www.angstronomics.com/p/ryze...esktop-preview

It does look like Zen3 was hamstrung by its front end:
https://www.pcgamer.com/amd-zen-4-pe...nce-breakdown/

So basically Zen4 is a fixed Zen3 with more cache and a higher clockspeed. This again makes me think Zen5 might be where we see more improvements to the rest of the core.

I've seen it suggested (e.g. https://www.youtube.com/watch?v=J2ohurfpfTI ) that Zen4 may move to a 6-wide decoder. Zen3 can dispatch 6 ops/clock depending on the op cache, but not sustain because of the 4 wide decoder. It would seem like a fairly big change to the front-end for AMD's version of a tick (AMD have adopted something like that sort of cadence now*), but if not in Zen4 I would expect to see this in Zen5.

However it does seem like some of the improvements to the core are in areas highlighted as limitations by the likes of chips and cheese.

Zen4 of course also introduces AVX512, though it seems like this may be achieved in a similar way to AVX2 on Zen1 i.e. instructions effectively taking two clocks through half-width FMAs. Therefore adding relatively little to die area on the decoders vs having to expand the execution back-end significantly too. Therefore you add compatibility without ballooning core size for an instruction that still has arguably limited use, especially on desktop.

*Just in case it's gone unnoticed by anyone, AMD seem to have settled into something like Intel's old Tick-Tock cadence. Zen1 > Zen2 brought improvements to a substantially similar core plan along with the move from 14/12nm to 7nm (Tick). Zen3 brought substantial core changes, plus new CCX layout, on the same node (Tock). Zen4 again appears to stick with a similar core design but new node. Though in this case it also coincides with a substantial platform change, notably including DDR5.