AMD's Zen 5 is a missed opportunity in messaging
semiaccurate.com
semiaccurate.com
> There is a lot of cool stuff in the Zen 5 core but at this point we would rather chew our hands off than lose two days of sleep trying in vain to write it up and failing because of lack of time. AMD did this by design, again, so we aren’t going to ruin our lives to make up for that crap.
Hilarious.
https://www.semiaccurate.com/2024/07/24/amds-zen-5-is-a-miss...
This happens more and more to me, even now with two of my banks' websites.
Did anything change recently in the standard / browser-enforcement of the standard that'd make that more common?
Or in the way some CA emits these?
I'm really noticing that more and more.
It's also not clear why many people have no issue and are hence posting links that work for some and not for others: is a major browser fine with these while another would be stricter?
Large enough businesses tend to clam up for changes in certain quarters. This makes hot spots for renewals/expiration or really anything potentially risky.
Nobody wants to do twice the paperwork for maintenance when it lands in sales-season 'change freezes'. I actually burned half a year of lifetime once just to get a more favorable schedule
If you're put off by gaming comparisons with X3D parts, that isn't really an apples to apples comparison. The current Zen5 parts don't have the huge additional caches that are such a win on some applications, but parts with those are planned for later release.
Similarly, going to 128 cores with the full arch or 192 cores with the cut down one is a big step up from 96/128, but these parts are also not shipping yet.
Separately, the performance/joule improvement looks very good-- for plenty of applications the total cost is dominated by energy usage so zen5 looks like a big win for those. Though again, this is mostly a factor for server parts that aren't shipping yet.
I think it would be fair to say zen5 is impressive while the initial zen5 offerings are less so.
AMD's AVX512 is a big deal because unlike Intel's implementation it doesn't trigger a 50 thousand cycle delay to start using it along with a massive downclock.
That means you can do SIMD way faster on these CPUs and there's virtually no induced latency (occasionally a 50 cycle delay for the 4x512 to presumably to finish charging caps if they are still depleted) and importantly no downclock which allows you to do burst AVX512 operations wherever you see a use instead of needing to carefully apply their use to compensate for a massive perf hit.
And while games generally push expensive SIMD to the GPU, having no-cost AVX512 lets you accelerate SIMD ops that aren't big enough to justify the latency of GPU offload or the hit to GPU occupancy.
It'll require engines and devs to adopt but there is a performance opportunity there if they are willing to take advantage of it.
Zen 5 brings ~13% for integer and ~25% for Floating Point performance increase for the current code, while lowering power usage on the same node compared to Zen 4. If we were to measure power / performance the Zen 5 is easily 25% to 50% uplift.
And it is also ignoring two things. Zen 5 was largely designed for 3nm in the first place. Some of the things for 4nm could be argued as not optimal as it should be. Second being Zen 5 was a ground up design and layering the foundation for Zen 6 and Zen 7.
I dont know if it was a missed opportunity in messaging. If I were to judge tech news circle from HN, Reddit, and most popular media. Hard core nerds that used to care and studying CPU Hardware changes are largely gone. All the details about front and back end design, Core to Core latency, things that Anandtech and other media that does deep dive into it no longer gets any views. And all that consumer cares if whether they could get higher performance, or more core.
So I guess it is a mixed AMD not doing good enough in marketing and media generally dont want to talk about things that dont interest their reader.
And then finally, AMD really cares about datacenter market, more so than consumer. Which is where Zen 5 is good at.
> When taking the geometric mean of those nearly 400 raw benchmark results, it sums up the greatness of Zen 5 with the Ryzen 5 9600X and Ryzen 7 9700X processors. The Ryzen 7 9700X delivered 1.195x the performance of the Core i5 14600K competition or 1.15x the performance of the prior generation Ryzen 7 7700X. The Ryzen 5 9600X came in at 1.35x the performance of the Core i5 14500 and 1.25x the performance of the Ryzen 5 7600X. Or if still on Zen 3 for comparison, the Ryzen 5 9600X was 1.82x the performance of the Ryzen 5 5600X.
...
> The raw performance results alone were impressive for this big Linux desktop CPU comparison but it's all the more mesmerizing when accounting for the CPU power use. On average across the nearly 400 benchmarks the Ryzen 5 9600X and Ryzen 7 9700X were consuming 73 Watts on average and a peak of 101~103 Watts. The Ryzen 5 7600X meanwhile had a 92 Watt average and a 149 Watt peak while the Ryzen 7 7700X had a 99 Watt average and 140 Watt peak. The Core i5 14600K with being a power hungry Raptor Lake had a 127 Watt average and a 236 Watt peak. The power efficiency of these Zen 5 processors are phenomenal!
E.g. look at the Phoronix numbers you linked. More than "it's Linux" they benchmark a bit heavier in the workstation/server tasks which is particularly good for focusing on what this generation improves (things like AVX 512 performance). Despite this:
5600X -> 7600X = 45% increase
7600X -> 9600X = 25% increase
The last generation was nearly twice, and not just in hyper specific workloads.
If anything levels out about this generation it'll probably be when AMD gets around to releasing the 3D versions early next year. They shoot themselves in the foot from a media/review perspective on that each release because there are always CPUs from the previous generation that smash certain cache heavy workloads better than the new shiny thing at release and then the 3D version comes out with less general discussion and coverage months later.
There are also substantial gains in areas that are not AVX accelerated - looking at Python performance, database performance, etc.
82.9% in two generations/4 years is decent but few are disappointed with the performance gains of getting off Zen 3, they're disappointed with the performance gains of Zen 5 itself.
This type of workload increased 27%, nearly half the average across the board performance increase Phoronix found for these types of task for Zen 3 -> Zen 4. https://i.imgur.com/xaD3x9Z.png
Obviously less with generic code. Amdahl’s law and all.
This seems to be "Intel got bad press for their stuff and we don't want any semblance of a defect in our launch" and is a minor bit of kneecapping to avoid the occasional weak sample missing advertised speeds in less than ideal conditions.
I love AMD to death but they do have a tendency to oversell on clock speeds even if their performance is stellar otherwise.
First time I hear about Pluton and, wow, yeah, no joke. I guess it's on to other CPU architectures for us when it gets time to replace current generation... :\
OpenTitan is a great option but it'll be at least another 2 or 3 years before it truly stabilizes and has at least some amount of battletesting in ARM chips as a secure enclave.
Alternatively they may go with Caliptra as they've been involved in its development for a long time however it is a less mature option and it's not as flexible of a secure environment as OpenTitan is.
The FOSS secure enclaves have a lot of heavy investment but it looks like the AI boom happened earlier than anyone really anticipated and Microsoft jumped the gun and demanded their SE environment because the other offerings weren't mature enough.
With all the new CPUs implementing Pluton you can't opt out. I'm sure Linux users aren't interested in a backdoor created by Microsoft.
That was pretty brutal on the marketeers..
at much lower power consumption! Every time a new processor comes around, lots of people complain that they don't need faster but want less heat and less noise. Now they get it and still complain.
It's a much better release than many of the Intel refreshes, but the marketing leading up to it was quite bizarre compared to launch performance.
Moving the same performance down ~40w is great. Having better branch prediction, more registers, lower latency on many ops and double the SIMD width for no cost? Fantastic.
They sold it as a huge gaming gain that hasn't materialized, and then tried to say it was due to windows admin modes interfering with branch prediction (True, but equally seems to apply to Zen3/4)
If they'd sold this as a perf/W and backend architecture shift, they wouldn't be getting the reaction they're currently earning.