AMD Execution Thread [2024]

[Albuquerque]

Bit of a disaster, honestly. I knew it wasn't gonna be a great gaming uplift, but this was worse than expected. Even other workloads aren't improved all that much.

Also frustrating how many people are trying to defend it because of 'big efficiency improvements' even though all they did was lower the TDP, and too many people dont seem to know you could also run the 7700X at 65w. Actual architectural efficiency improvements seem fairly minimal.

Watched the GN video and laughed several times as he continually talks shit on the 13th and 14th gen Intel chips.

Also, I feel like Steve is probably right, in that some updated firmware and/or AGESA will probably fix the strangely low performance behavior on the older motherboards.

 

[homerdog]

"Because it's so efficient". Only real reasons I can see are
- Planning to launch 7800X soon with 105W
- Trying to make 7900X look better
If you don't use AVX512 there isn't really much point on getting 9000, we'll see if X3D changes things.

This doesn't make sense to me. Why are they releasing a 7800X at this point? Also the 7900X has been out for a long time, why would they need to make it look better?

 

[arandomguy]

This doesn't make sense to me. Why are they releasing a 7800X at this point? Also the 7900X has been out for a long time, why would they need to make it look better?

I'm guessing a typo and the poster meant 9800X and 9900X.

 

[homerdog]

I'm guessing a typo and the poster meant 9800X and 9900X.

I thought that might be the case but it's not obvious. AMD does release new old CPUs all the time.

Assuming he meant 9800X/9900X, why the hell would AMD not lead with those CPUs instead of the pathetic 9700X? Surely they could see the "Zen 5 Sucks" headlines coming with this performance and price.

 

[Kaotik]

I thought that might be the case but it's not obvious. AMD does release new old CPUs all the time.

Assuming he meant 9800X/9900X, why the hell would AMD not lead with those CPUs instead of the pathetic 9700X? Surely they could see the "Zen 5 Sucks" headlines coming with this performance and price.

Yeah, meant 9800X and 9900X, too used to writing about 7000-series. I didn't say those were good reasons, just only ones I could figure out.

 

[arandomguy]

In terms of the sentiment regarding the gaming performance for this release there's some things to keep in mind here in that gaming performance as we term it is very heavily dependent on data access latency and speeds and that is heavily dependent on everything outside of the core itself, even though we colloquially tend to lump all that into terms like IPC. With Zen 5 AMD's chosen not to update the I/O die so the memory controller and performance on that front has no uplift, unlike the case of Zen 3 -> 4 (especially DDR5). Meanwhile they've already kind of exploited some of the lower hanging fruit in terms the increased budget on cache with previous generations already.

I'll have to dig up the statements but from what I remember AMD's commented publicly before that their philosophy is that they feel at some point you need to spend a generation to clear tech debt as opposed to just trying to iteratively squeeze more out of every generation. I believe the commenting implied that Zen 5 would be the generation for that more "ground up" design as opposed to a focus on jsut iterating more performance.

Zen 5 of course unlike Zen 4 doesn't benefit from a significant node jump as the case of Zen 3 to Zen 4.

The other thing I feel expectations might need to be some what recalibrated in terms of CPU improvements gen on gen. AMD did enjoy very high gen on gen gains from Zen 1 and onwards but it's worth remembering they were so far behind Intel prior to that. This gave much more room to improve every generation and that may not (and likely not) be indicative of what the future trend is now that both AMD and Intel are roughly at parity and AMD is likely facing some of the same inherent challenges in trying to improve each gen that limited Intel. With gaming specifically (and especially high FPS gaming) that challenge is going to be latency.

 

[homerdog]

How much does the 9700X outperform the 7700X if they're clocked the same?

 

[Albuquerque]

I don't think anyone has (yet) forced the same clockspeeds across the separate CPUs to check...

 

[DegustatoR]

How much does the 9700X outperform the 7700X if they're clocked the same?

CB did some tests with a same power limit but they've only checked clocks/temps/power draw. I've yet to see anything which would suggest that Zen 5 is a sizeable step over Zen 4 in power efficiency.

 

[Newguy]

How much does the 9700X outperform the 7700X if they're clocked the same?

From Geekerwan in Spec it's +8.6% Int, +26% FP at 4GHz, although full rate AVX512 is probably putting in work there

 

[Seanspeed]

It's a fantastic architecture, for laptops and datacenter.

AMD said it wasn't for desktop gaming, fair enough, so all these youtuber reviews are whinging about something everyone was already warned about.

But it's also not for gaming because AMD's priority of selling desktop CPUs to gamers is seconhand at best. Gamers get the X3d series months down the line, and only when inventory is free from the far more profitable laptop/datacenter market is even available. And datacenter is on 3nm while consumer is on 4nm now, meaning more inventory for gamers, but only after laptops (why Strix Point launched first). Thus gamers are third tier customers because the profit margin x revenue total just isn't there. That AMD currently isn't blowing away their tier 3 customers with the newest launch isn't really a big concern for them finance wise, especially with the lucky break of Intel screwing over the same customer segment at the same time.

What makes it so good for laptops, but not desktops? Again, the supposed efficiency improvements aren't actually real.

https://twitter.com/x/status/1821307394238116061

Shows that there's still only a 7% improvement in performance per watt over the Zen 4 7700 at an equal 65w. That's miniscule. People are being bamboozled cuz they simply lowered TDP, and lots of people thinking that means it inherently needs less power or something rather than it just being something AMD can decide. They could have made it 30w or 170w. It says nothing about the actual efficiency of the product.

I'm bewildered how many people are trying to excuse what is so clearly a disappointing release, and that everybody seems to be ignoring that AMD basically once again lied in their claims about performance. I know it's the popular thing to crap on Intel right now, but that doesn't mean people need to 'take sides' and ignore when AMD does poorly, too.

And to be clear, I'm sure Zen 5 is an important step for the future and moving to a wider architecture and all that, but as actual consumer products, they are pretty lousy outside very specific use cases.

 

[DegustatoR]

Yeah, I don't think there's anything to write home about regrading power efficiency of Zen 5 vs 4.
To me it looks more like the chips don't clock high enough for some reason (results with PBO suggest that even with some outrageously high PL performance isn't improving much) and thus AMD decided to play the "efficiency" angle here despite there not really being that much in terms of efficiency improvements.
It may of course be a result of parts binning, with worse chips going into 1CCD SKUs. We'll see once the top SKUs will launch.

 

[Wesker]

What makes it so good for laptops, but not desktops? Again, the supposed efficiency improvements aren't actually real.

https://twitter.com/x/status/1821307394238116061

Shows that there's still only a 7% improvement in performance per watt over the Zen 4 7700 at an equal 65w. That's miniscule. People are being bamboozled cuz they simply lowered TDP, and lots of people thinking that means it inherently needs less power or something rather than it just being something AMD can decide. They could have made it 30w or 170w. It says nothing about the actual efficiency of the product.

I'm bewildered how many people are trying to excuse what is so clearly a disappointing release, and that everybody seems to be ignoring that AMD basically once again lied in their claims about performance. I know it's the popular thing to crap on Intel right now, but that doesn't mean people need to 'take sides' and ignore when AMD does poorly, too.

And to be clear, I'm sure Zen 5 is an important step for the future and moving to a wider architecture and all that, but as actual consumer products, they are pretty lousy outside very specific use cases.

Maybe Zen 5 is truly able show a perf/w advantage over Zen 4 at TDPs <25w?

Because I agree: at 65w there is a negligible difference between the two uarchs.

 

[entity279]

It would be consistent with the ideea that zen5 desktop is memory bandwidth sub-system performance limited. At lower clocks and possibly with more notebook-relevant benchmarks/workloads, this issue would be reduced.

Edit: If Zen 5 turns out to be impressive both in notebook & server setting, and disapointing for desktop, that would be a strange "achievement". But also would make understanding why interesting.

 

[Erinyes]

What makes it so good for laptops, but not desktops? Again, the supposed efficiency improvements aren't actually real.

https://twitter.com/x/status/1821307394238116061

Shows that there's still only a 7% improvement in performance per watt over the Zen 4 7700 at an equal 65w. That's miniscule. People are being bamboozled cuz they simply lowered TDP, and lots of people thinking that means it inherently needs less power or something rather than it just being something AMD can decide. They could have made it 30w or 170w. It says nothing about the actual efficiency of the product.

I'm bewildered how many people are trying to excuse what is so clearly a disappointing release, and that everybody seems to be ignoring that AMD basically once again lied in their claims about performance. I know it's the popular thing to crap on Intel right now, but that doesn't mean people need to 'take sides' and ignore when AMD does poorly, too.

And to be clear, I'm sure Zen 5 is an important step for the future and moving to a wider architecture and all that, but as actual consumer products, they are pretty lousy outside very specific use cases.

Keep in mind that 7% can easily be the difference between the same parts due to variation in silicon quality/bin alone, and coupled with the fact that they've moved from a "5nm" class process to a "4nm" class process which supposedly brings some improvements in power as well. So overall yeah Zen 5 seems pretty underwhelming and AMD's IPC increase figure, which has been fairly accurate historically, is definitely more like 10% at best this time around than 16% (and even lower in gaming of course).

There seems to be some bottleneck in the architecture as despite the increase in width and number of transistors, the performance increase is proportionately much smaller. Could it be the uncore? Supposedly that's getting completely revised for Zen 6 and perhaps might unlock more performance. And 3nm should help too. Zen 6 should be more like a Zen 4 moment for AMD, i.e. increase in IPC and also clocks due to a node switch. But until then, Zen 5 is a pretty underwhelming offering and you're probably better off buying Zen 4/X3D at a lower price. Unless Zen 5 X3D surprises, it'll be a skip generation for enthusiasts.

Maybe Zen 5 is truly able show a perf/w advantage over Zen 4 at TDPs <25w?

Because I agree: at 65w there is a negligible difference between the two uarchs.

Seemingly from the laptop tests, the Zen 5 parts have a significant increase in perf/W at sub 30W, though this is also due to having 12 cores of different types instead of 8, which can be clocked lower and be more power efficient. So it's not exactly apples to apples.

But it seems like Zen 5 was designed with a server focus, with mobile a second priority and desktop last. Kind of makes sense given the volumes and margins they are making/can make from each division.

 

[Shifty Geezer]

I'll have to dig up the statements but from what I remember AMD's commented publicly before that their philosophy is that they feel at some point you need to spend a generation to clear tech debt as opposed to just trying to iteratively squeeze more out of every generation. I believe the commenting implied that Zen 5 would be the generation for that more "ground up" design as opposed to a focus on jsut iterating more performance.

What does that get the consumer though? If it doesn't get them anything, how do you sell to them? "It works about the same as the old chip but it's all laid out so much nice in the die shots"?

 

[entity279]

I do feel that enthusiasts / gamers kinda lost the plot.

Not saying zen5 is not a dissapointment in some ways, there were suggestions and expectations set for more indeed, but:

There's little need to upgrade every 2 or even 3 generations for desktop CPUs now (since like 6 + years). Changing cpus this often doesn't really impact the gaming experience nor any other experience.
For gaming, upgrading the gpu often , as we of course know here, yields by far better value comparatively and we don't reach cpu limitations with anything mid range that's say 4 years old or less.

That zen 5 looks unimpressive next to zen 4 (which IMO was on the other hand a positive surprise at the time), will matter little for AMD or even the customers. Unless Arrow Lake will prove to be a much bigger leap, of course.

 

[Frenetic Pony]

Yeah, I don't think there's anything to write home about regrading power efficiency of Zen 5 vs 4.
To me it looks more like the chips don't clock high enough for some reason (results with PBO suggest that even with some outrageously high PL performance isn't improving much) and thus AMD decided to play the "efficiency" angle here despite there not really being that much in terms of efficiency improvements.
It may of course be a result of parts binning, with worse chips going into 1CCD SKUs. We'll see once the top SKUs will launch.

5.5ghz < 10% lower than 6ghz oxidation trouble ultra high power Intel CPUs. The clockspeeds are fine, 9950x only runs at 5.7ghz and that extra .2ghz (<5% clockspeed increase) drives TDP 30% higher (taken from doubling 8 core to 130w).

You'd get barely anything from increasing TDP. Sure maybe the 9800x3d will be binned for 6ghz, please the gamers, but watch it double the TDP to 130w just to do so.

 

[Newguy]

Zen 5 does look like the bottom of another S curve, the foundations for better things. Saner power out of the box (granted it's easy to turn on eco mode for Zen 4), much wider arch that should have much more headroom to build on, full rate avx 512. They do need Zen 6 to really push on perf and efficency again though (arch, packaging, IO die etc)

 

[homerdog]

5.5ghz < 10% lower than 6ghz oxidation trouble ultra high power Intel CPUs. The clockspeeds are fine, 9950x only runs at 5.7ghz and that extra .2ghz (<5% clockspeed increase) drives TDP 30% higher (taken from doubling 8 core to 130w).

You'd get barely anything from increasing TDP. Sure maybe the 9800x3d will be binned for 6ghz, please the gamers, but watch it double the TDP to 130w just to do so.

Very Efficient Ryzen 7 9700X Held Back by Power Limits!​

Silly to limit this part to 65W. At least give it the same 105W as the 7700X.