Why everyone say Zen 5 is bad ?

[Zach_01]

Just shocked AMD 9700X performance. I mean should we expect it to be faster than a 7800X3D in all retrospect?

No but normally you would expect some gains over Zen4, instead of +3% on avg.

But welcome to the new normal...
I've been saying it for a while
AMD is pushing gamers to X3D parts exclusively

 

[oxrufiioxo]

No but normally you would expect some gains over Zen4, instead of +3% on avg.

But welcome to the new normal...
I've been saying it for a while
AMD is pushing gamers to X3D parts exclusively

I mean the last 2 generations had decent gaming gains outside of X3D and sure the 5800X3D is ballpark 7600X gaming perfomance but it released late in the generation. The 2600, 3600, 5600, 7600 were all decent entry level gaming chips and different enough generation to generation.

 

[Lost_Troll]

It changes things by 1-2% because the bug also affects Zen 4 so I doubt the conclusion will change. It might also be degrading intel gaming performance.

Please watch the video in it's entirely, it clearly points out that the issue affects the 7000 series as well. Any gains from a fix are therefore moot in the performance comparison.

I know, but my point was this...

I really do think it is microcode/driver and BIOS issues that are causing allot of the performance issues. I also think that over time, these will be fixed.

depending on what the Arrow Lake i5 looks like

And I think you answered why WIN 11 is bugged now, Arrow Lake has no Hyperthreading, so now the thread manger in windows needs to be updated, and I bet that is what interduce this bug.

 

[Zach_01]

I mean the last 2 generations had decent gaming gains outside of X3D and sure the 5800X3D is ballpark 7600X gaming perfomance but it released late in the generation. The 2600, 3600, 5600, 7600 were all decent entry level gaming chips and different enough generation to generation.

I hear you, no argue there.
All Im saying is now that AMD established the X3D existence on both platforms and many taste them, its time to move on further on the path that took with the modular system architecture.
Make EPYC even stronger now and on the years to come while maintaining the universal arch for desktops too with some performance mitigations for the moment on the regular parts that are not targeting gamers anymore.

Its that simple

 

[mäggäri]

I think you should stick to the topic.

okay den Snapdragon it is since Zen4-Zen5 is not competitive thanks for confirming my thoughts

 

[SL2]

But welcome to the new normal...

No, I don't think so. This doesn't help anyone except Intel. As long as 3D and non-3D chiplets are made from the same silicon, this is bad for 3D models as well.
No matter how you dice it, this looks bad for AMD as it will affect sales, without being a catastrophe.

If games are 3 % faster with a 9700X compared to a 7700X, that's not a good starting point for an upcoming 9800X3D.

On the other hand we have of course higher possible clock speeds, if true, and maybe 3D V-cache implementation can be improved in some way.

AMD is pushing gamers to X3D parts exclusively

Unless we're seeing budget 3D models in the future, AMD would have nothing to compete with in the budget segment, even if that's not a first.

so now the thread manger in windows needs to be updated

No, running without HT isn't a new, unique thing. People have disabled it for performance reasons for decades.

Not to mention all the CPU's without HT, like the G6900.

 

[Zach_01]

No, I don't think so. This doesn't help anyone except Intel. As long as 3D and non-3D chiplets are made from the same silicon, this is bad for 3D models as well.
No matter how you dice it, this looks bad for AMD as it will affect sales, without being a catastrophe.

If games are 3 % faster with a 9700X compared to a 7700X, that's not a good starting point for an upcoming 9800X3D.

On the other hand we have of course higher possible clock speeds, if true, and maybe 3D V-cache implementation can be improved in some way.

From my perspective comparing 7000<>9000 and 7000X3D<>9000X3D is a bit like apples and oranges
If the major "problem" of Zen5 desktop is the obsolete IOD (responsible maybe also for the very high cross-CCD latency) then why doubling the cache wont help significantly, and even more than it did on Zen4?
There are hints that Zen5 desktop like a little too much tweaked timings and high speed memory. More than Zen4 did...
We know that games love low latency and high bandwidth. We know that current X3Ds dont care much for high speed/low latency memory subsystem because of the extra cache.
I think just by slapping extra cache on Zen5 will gain more than Zen4 did. And IF (big if) AMD has some more tweaks about it... then... shall see

A simple extrapolation of performance between 2 different enough chiplets... is too much simple to my understanding.

 

[AusWolf]

The admin account bug is not exclusive to Zen 5, the 7700X is affected and saw a similar performance uplift..Its all in the video of Hardware Unboxed.

Seriously, that's quite pathetic from AMD, ZEN 5 is really bad, Gamers Nexus also made a video pointing out that Zen 5 is not as efficient as it is rumored to be, it really is a case by case basis.

That's really a generation to skip, even for people buying a new machine, as Zen 4 is way cheaper and performs roughly as good, and not that far energy efficiency wise.

When it's bad, it's bad.

And AMD trying to find any excuse post release to justify the lack of performance increase is really disappointing. They lose some credit to me, and surely to larger amount of people, due to that lack of transparency and how they overhyped those new cpu's, while in fact they knew it would be disappointing, they clearly take people for idiots.

I don't think it's that dramatic. Sure, Zen 5 offers very little, if anything at all over Zen 4, but that's not a reason to lose confidence in AMD. It's not like these CPUs are gonna blow up in your PC or anything. They're just not very exciting, that's all. Zen 6 may very well be an awesome upgrade. Or not. We'll see.

 

[oxrufiioxo]

I don't think it's that dramatic. Sure, Zen 5 offers very little, if anything at all over Zen 4, but that's not a reason to lose confidence in AMD. It's not like these CPUs are gonna blow up in your PC or anything. They're just not very exciting, that's all. Zen 6 may very well be an awesome upgrade. Or not. We'll see.

This just adds more fuel to the fire that AMD sucks at launching products and even though I've never had issues even all the way back to X370 I know of many who did with memory, usb dropouts, boost clocks etc.

So while it wouldn't be a big deal if this was the first time but it isn't. Performance uplifts have allowed them to gloss over it in the past but they got under performing cpus at least by their numbers, scheduling issues with MT, and a odd windows bug that might affect all their cpus all in one shot. Will that impact the way I view them or will it stop me from buying a future AMD cpu nope. I still can understand others going really AMD again.

 

[phanbuey]

From my perspective comparing 7000<>9000 and 7000X3D<>9000X3D is a bit like apples and oranges
If the major "problem" of Zen5 desktop is the obsolete IOD (responsible maybe also for the very high cross-CCD latency) then why doubling the cache wont help significantly, and even more than it did on Zen4?
There are hints that Zen5 desktop like a little too much tweaked timings and high speed memory. More than Zen4 did...
We know that games love low latency and high bandwidth. We know that current X3Ds dont care much for high speed/low latency memory subsystem because of the extra cache.
I think just by slapping extra cache on Zen5 will gain more than Zen4 did. And IF (big if) AMD has some more tweaks about it... then... shall see

A simple extrapolation of performance between 2 different enough chiplets... is too much simple to my understanding.

I think the Zen 5 has great single core which is very promising - which was actually the weakest point of the 7800x3d. With X3D and some microcode I really think we're going to see about 10% better than 7800X3D (as a blind guess based on some of the more promising benches).

7800X3d is already a fantastic chip and is 5-10% faster than raptor lake... so if you can stack another 10% on that - then arrow lake will have to be 15-20% faster in games than Raptor lake to catch. Which - unless there is a small miracle isn't going to happen - possible, given the latency improvements on the cache and the core rearranging, but still not likely.

Zen 5 just seems like it's either bugged or bottlenecked...

 

[mäggäri]

I think the Zen 5 has great single core which is very promising - which was actually the weakest point of the 7800x3d. With X3D and some microcode I really think we're going to see about 10% better than 7800X3D (as a blind guess based on some of the more promising benches).

7800X3d is already a fantastic chip and is 5-10% faster than raptor lake... so if you can stack another 10% on that - then arrow lake will have to be 15-20% faster in games than Raptor lake to catch. Which - unless there is a small miracle isn't going to happen - possible, given the latency improvements on the cache and the core rearranging, but still not likely.

Zen 5 just seems like it's either bugged or bottlenecked...

nice nice, 0 difference - Or are you saying that TPU charts are lying

 

[phanbuey]

nice nice, 0 difference - Or are you saying that TPU charts are lying

No Im saying that with TDP boost, more aggressive clocking, and working out of the bugs with windows fixes it will get to ~10%, add X3D to that... you've got yourself a nice chip.

If you remember Zen4 had all sorts of issues, memory training, microstuttering, USB issues, fTPM issues, some of them exploded... half the motherboards didn't have subtimings figured out and there was a 15% gap between board makers with XMP profiles... HWUB did a build live that just straight up didn't post with all new in box parts.

 

[mäggäri]

View attachment 359171

No Im saying that with TDP boost, more aggressive clocking, and working out of the bugs with windows fixes it will get to ~10%, add X3D to that... you've got yourself a nice chip.

If you remember Zen4 had all sorts of issues, memory training, microstuttering, USB issues, fTPM issues, some of them exploded... HWUB did a build live that just straight up didn't post with all new in box parts.

1440p or grab your jacket, I'm waiting.

720p is nonsense, artificially created statistics which play 0 role IRL. Better to stick to meaningful graphs, than to cherry pick like there's no tomorrow.


why "amd-life" is always full of such cope, excuses, 170W on AMD is ok all good, but if competitor has 169W usage it's END OF THE WORLD Like AMD marketing hype before release, all techtubers laughing at n*x^y+z times at those full blown lies.

 

[phanbuey]

1440p or grab your jacket, I'm waiting.

720p is nonsense, artificially created statistics which play 0 role IRL. Better to stick to meaningful graphs, than to cherry pick like there's no tomorrow.

Just go to 4k ultra RT - why waste time with 1440P...

 

[oxrufiioxo]

1440p or grab your jacket, I'm waiting.

720p is nonsense, artificially created statistics which play 0 role IRL.

That's just how cpu reviews work you need to put them in scenarios where they are the limiting factor or else it's just a gpu benchmark.

Just go to 4k ultra RT - why waste time with 1440P...

His view is ok for his use case but we still need to see cpu limited testing. Not sure why people don't get that.

 

[mäggäri]

That's just how cpu reviews work you need to put them in scenarios where they are the limiting factor or else it's just a gpu benchmark.



His view is ok for his use case but we still need to see cpu limited testing. Not sure why people don't get that.

720p is like monopoly money, has 0 use when can't do anyting useful with it.

I hope soon Asetto Corsa gets removed from the game tests, like 1000players on it, such indie game... Only ruins the sum up.

 

[oxrufiioxo]

720p is like monopoly money, has 0 use when can't do anyting useful with it.

I hope soon Asetto Corsa gets removed from the game tests, like 1000players on it, such indie game... Only ruins the sum up.

With a 4090 you can be cpu limited at 4k so it's still relevant to see absolute differences. Also some gamers use competitive settings to get high fps in that scenario showing cpu limited results are beneficial.

Sorry to say it man but the world of gaming benchmarks doesn't revolve around you.

I don't care about the 720p results either doesn't make them useless.

 

[mäggäri]

With a 4090 you can be cpu limited at 4k so it's still relevant to see absolute differences. Also some gamers use competitive settings to get high fps in that scenario showing cpu limited results are beneficial.

Sorry to say it man but the world of gaming benchmarks doesn't revolve around you.

I don't care about the 720p results either doesn't make them useless.

higher lows are more meaningful, but ok, also FPS gets capped to avoid stuttering

it's ok it doesn't revolve around me, around 1440p it does, the mainstream resolution in modern world (none cares about steam statistics, when can't restrict poor ppl out)


Santa Claus is more real than the forcefully made up scenarios of 720p, tells enough for me.

 

[DemonicRyzen666]

nice nice, 0 difference - Or are you saying that TPU charts are lying

Zen 5 is bugged, or AGESA is bugged.

Check the youtube interview with chip & cheese.

In the interview with the zen engineer. He said zen 5 can use both, predictors, pipelines & decoders when it uses a single thread for the core.

He even mentions how if the core needed data that was retained in the either one of the predictor, pipeline, decoder, after it was done executing On the SMT thread it could be retrived &/or used on the main thread and take over thr neex for those dual parts.

On chips and cheese review his test show that is not doing that or it cannot do this. Meaning all its prefromance gains possible are lost. The gains you see right now the 0 to 5%with SMT disabled are all from load store increases and larger enrty data bases for the cpu. Thats it & nothing else.

On comment on cheese & chips pretty much nails this

Jon Brase says:
August 14, 2024 at 3:01 pm
I suspect that gaming workloads end up front-end bandwidth bound. I.e, they’re likely mostly spilling out of the uOp cache and only using one thread, at which point front-end bandwidth becomes identical to Zen 4.

This correct, but not totally i as pointed out its just not doing what its suppose to be able to.

So when it does not or cannot use those dual predictor, pipelines, & decoders it has simiar or the same ipc as zen 4

L3 cache bandwidth was increased is 1.4TBs zen 4 was 850GBs. Adding V-cache only hides latency, because less of a need to reach into ram often and more chances to find in the L3 victim cache.

The increase in L3 badwidth is good. This mean latency does not matter on zen 5 & overall bandwidth will/should matter more. Unless you hit the infinity fabric limit of bandwidth ( i don't know the limit curretly). So a test with the 2:1 ram divider are now needed, they'll probably show minimal or no differnce in games.

This acutally the opposite problem Bulldozer had. Bulldozer instruction, predictor, pipeline & cache thrashing bewteen the two cluster cores needs from the single front end caused stalls. So when disabling a core of the cluster the modules ipc went up.

This is just lack of actaully using the front end fully. When you use the SMT threads loading the core up the gains are there becuase its using both of those three things(predictor pipeline decoder) the ipc is up, & not by a little but by A LOT. Those three parts are needed for the fpu's AVX512 too.

 

[mäggäri]

Just go to 4k ultra RT - why waste time with 1440P...

Why to stop there? Just grab some 8k... Oh yes, the hardware is not even 4k capable (enough) yet. Come guys, I know you can have some realism.

Some console tier FPS is not accepted in my house.

Zen 5 is bugged, or AGESA is bugged.

Check the youtube interview with chip & cheese.

In the interview with the zen engineer. He said zen 5 can use both, predictors, pipelines & decoders when it uses a single thread for the core.

He even mentions how if the core needed data that was retained in the either one of the predictor, pipeline, decoder, after it was done executing On the SMT thread it could be retrived &/or used on the main thread and take over thr neex for those dual parts.

On chips and cheese review his test show that is not doing that or it cannot do this. Meaning all its prefromance gains possible are lost. The gains you see right now the 0 to 5%with SMT disabled are all from load store increases and larger enrty data bases for the cpu. Thats it & nothing else.

On comment on cheese & chips pretty much nails this

This correct, but not totally i as pointed out its just not doing what its suppose to be able to.

So when it does not or cannot use those dual predictor, pipelines, & decoders it has simiar or the same ipc as zen 4

L3 cache bandwidth was increased is 1.4TBs zen 4 was 850GBs. Adding V-cache only hides latency, because less of a need to reach into ram often and more chances to find in the L3 victim cache.

The increase in L3 badwidth is good. This mean latency does not matter on zen 5 & overall bandwidth will/should matter more. Unless you hit the infinity fabric limit of bandwidth ( i don't know the limit curretly). So a test with the 2:1 ram divider are now needed, they'll probably show minimal or no differnce in games.

This acutally the opposite problem Bulldozer had. Bulldozer instruction, predictor, pipeline & cache thrashing bewteen the two cluster cores needs from the single front end caused stalls. So when disabling a core of the cluster the modules ipc went up.

This is just lack of actaully using the front end fully. When you use the SMT threads loading the core up the gains are there becuase its using both of those three things(predictor pipeline decoder) the ipc is up, & not by a little but by A LOT. Those three parts are needed for the fpu's AVX512 too.

Now we wait for the Zen5 fix, hopefully happens some day... IDK why AMD tests SOC max voltage on consumers, then this case. Previously 5700XT black screen which never got fixed.

ATM playing on AMD cpu PC, having AM5 mobo. Laptop has AMD+nvidia. 0 Intel here. Mby next to get ASrock Steel Legend B650 (white one) when it's 180e again. Or just B650 Taichi lite.

 

[oxrufiioxo]

ATM playing on AMD cpu PC, having AM5 mobo. Laptop has AMD+nvidia. 0 Intel here. Mby next to get ASrock Steel Legend B650 (white one) when it's 180e again. Or just B650 Taichi lite.

The Taichi lite looks awesome I wish I had one, I got 2 Gigabyte X670E PRO X boards for free they are kinda meh but free is free I guess.

For sure hopefully they iron out these issues.

 

[mäggäri]

The Taichi lite looks awesome I wish I had one, I got 2 Gigabyte X670E PRO X boards for free they are kinda meh but free is free I guess.

For sure hopefully they iron out these issues.

B650 (non-E) steel legend has pcie 5.0 for GPU too, if matters for some. Weird mobo, that's why it's interesting at same time and nice extra for Lian Li PCMR (space grey, that mirror case)

typical Asus TUF B650-Plus WIFI here, was 147e including shipping and 24% taxes, couldn't skip... also 3year warranty which is nice nowadays.

 

[oxrufiioxo]

B650 (non-E) steel legend has pcie 5.0 for GPU too, if matters for some.

Yeah, Asrock has a decent stack from bottom to top this round hopefully they continue that with the 800 series boards. I like the post code and usb4 on the tachi.

 

[phanbuey]

Why to stop there? Just grab some 8k... Oh yes, the hardware is not even 4k capable (enough) yet. Come guys, I know you can have some realism.

I game at 4K - so why would I care about your 1440p? - realism bro. The point of my statement was exactly that - all 1440p does is introduce a gpu bind… so it’s not good for a cpu test.

it’s a CPU benchmark and we’re discussing cpu performance - the reality is anything 13600k and above performs the same at 4K - does that mean there isn’t a processor faster at gaming than a 13600k?

When u look to future proof a gpu - u look at 4K, when u look to future proof a cpu - u look to 720p/1080p. Why? Those are the resolutions that bottleneck each respective part the most.

 

[AusWolf]

This just adds more fuel to the fire that AMD sucks at launching products and even though I've never had issues even all the way back to X370 I know of many who did with memory, usb dropouts, boost clocks etc.

So while it wouldn't be a big deal if this was the first time but it isn't. Performance uplifts have allowed them to gloss over it in the past but they got under performing cpus at least by their numbers, scheduling issues with MT, and a odd windows bug that might affect all their cpus all in one shot. Will that impact the way I view them or will it stop me from buying a future AMD cpu nope. I still can understand others going really AMD again.

It's not a big deal because there's nothing wrong with the CPUs, they're just not interesting compared to Zen 4. Similarly, Intel's 4-core stagnation era wasn't a big deal, either. At least you could keep your Sandy Bridge i7 for almost a decade. If Zen 4 turns out to be the new Sandy Bridge, I won't complain.