Yes that subtest result should be discarded to compute the speedup or that would not be fair Sorry could not resist.Yep, Background blur uses AVX-512.
I think you won't feel that way once the Linux patches are all done.But still the uplift in 2 years is mediocre.
I have all the Zen DT generations and a 2K+ cores of Zen 4 in server at the moment
A large number of 9374F for CI/CD builds, dev environments which emulates the HW my guys are developing.Fleet of machines for DevOps/docker containers or is that single server with 2k+ cores?
I'm always humbled how big some company setups are. I'm trying to justify getting even a Ryzen 7950X (or 9950X) build for a CI/CD server (it's not strictly necessary as most of our clients have their own infra) but sure would be helpful at times.A large number of 9374F for CI/CD builds, dev environments which emulates the HW my guys are developing.
Personally, as an enthusiast/hobbyist, I am eager to buy the CPUs regardless of its perf. But still the uplift in 2 years is mediocre.
Which patches are you talking about?I think you won't feel that way once the Linux patches are all done.
ASUS System Product Name vs Micro-Star International Co., Ltd. MS-7D70 - Geekbench
100MHz clock increase
+16% ST
+7% MT
Large speedup for Background Blur ST (AVX-512 improvements Zen5?).
| Memory Information | |
|---|---|
| Size | 32.00 GB |
| Transfer Rate | 5998 MT/s |
| Type | DDR5 SDRAM |
| Channels | 2 |
adroc_thurston and the order of 32% amirite?Folks who took that +32% SIR2017 koolaid having withdrawal symptoms.
It should be fairly suspect, as it was delivered by the same AMD vanguards who delivered RDNA3 chungus perf.
Ironic that they were playing pranks on YouTubers but played themselves with the Zen 5% meme
...
Comments of people trying to having a decent discussion getting drowned by advocates patrolling the threads ready to answer the skeptics on moment's notice.
Well then my conclusion is that AVX-512 disablement in the UEFI is broken and doesn't actually disable AVX-512. Or just prevents certain AVX-512 instructions from running.
I don't know about Ryzen BIOSes. But at least with my own _old_ EPYC BIOS, the BIOS switch to disable AVX-512 doesn't have any effect. As a Linux user, I can work around it though by adding a cryptic kernel command line option at boot time which masks out the respective CPU capability flags. But AFAIK even this option is not guaranteed to affect some software; in such cases you would have to have a switch in this software itself at run time or compile time.Or perhaps GB AVX-512 path is faster only on Intel machines due to the 256-bit DP of AMD Zen4. But I find it strange.
Both machines seem to share the same memory characteristics (except for the memory amount). So I'm not sure what your point is
Memory Information Size 32.00 GB Transfer Rate 5998 MT/s Type DDR5 SDRAM Channels 2
its kinda boring right now , no idea what the internal choices they have made/ why it performs the way it does. But on face value 48KB L1D , 6 ALU , dual decode , large ROB / PRF for 15 points of IPC is a failure. In my mind its worse then bulldozer , the saving grace is start point was so much higher. Once we really get the internals of the Core detailed then i might change my mind.Folks who took that +32% SIR2017 koolaid having withdrawal symptoms.
It should be fairly suspect, as it was delivered by the same AMD vanguards who delivered RDNA3 chungus perf.
Ironic that they were playing pranks on YouTubers but played themselves with the Zen 5% meme
Instead of being skeptical folks were trying to legitimize the veracity of such claims with a myriad of hallucinations when history should have shown otherwise.
Comments of people trying to having a decent discussion getting drowned by advocates patrolling the threads ready to answer the skeptics on moment's notice.
Personally, as an enthusiast/hobbyist, I am eager to buy the CPUs regardless of its perf. But still the uplift in 2 years is mediocre.
I have all the Zen DT generations and a 2K+ cores of Zen 4 in server at the moment
and grow out of it eventuallyFolks who took that +32% SIR2017 koolaid having withdrawal symptoms
i have about the same number of ice lake cores because i was forced to use a specific vendors product because of politics that they then EOL'd / killed. Atleast im only running 32 cores a proc so i can just burn power for clocks...........A large number of 9374F for CI/CD builds, dev environments which emulates the HW my guys are developing.
An example (I would have better ones if Phoronix had a better search interface with useful filtering options):Which patches are you talking about?
Ugh... the first test was done on Ubuntu 22.04; kernel 6.0; Blender 3.2. The second one with Ubuntu 23.04; kernel 6.2; Blender 3.4.An example (I would have better ones if Phoronix had a better search interface with useful filtering options):
AMD Ryzen 9 7950X3D Linux Performance Review - Phoronix
www.phoronix.com
View attachment 102650
8% improvement in 150 days. Who knows what the current score of 7950X is in this benchmark.
This has to be related to Linux kernel patches or compiler improvements because I can't see any mention of Zen 4 specific optimizations in the Blender 3.4 release notes.
Not only that, but Zen 4 added multiple AVX-512 instructions, that the compilers / newer kernel could theoretically take advantage of. Zen 5 widens the cores, but very few instructions are added, so most of the net-benefit from the wider AVX-512 units should be there from the get-go.Ugh... the first test was done on Ubuntu 22.04; kernel 6.0; Blender 3.2. The second one with Ubuntu 23.04; kernel 6.2; Blender 3.4.
Completely different version of software producing different results using the same hardware. What's the point?
Old 5950X scores 800 vs 754 seconds, so yeah, that newer setup is faster.
An example (I would have better ones if Phoronix had a better search interface with useful filtering options):
AMD Ryzen 9 7950X3D Linux Performance Review - Phoronix
www.phoronix.com
View attachment 102650
8% improvement in 150 days. Who knows what the current score of 7950X is in this benchmark.
This has to be related to Linux kernel patches or compiler improvements because I can't see any mention of Zen 4 specific optimizations in the Blender 3.4 release notes.
its kinda boring right now , no idea what the internal choices they have made/ why it performs the way it does. But on face value 48KB L1D , 6 ALU , dual decode , large ROB / PRF for 15 points of IPC is a failure. In my mind its worse then bulldozer , the saving grace is start point was so much higher. Once we really get the internals of the Core detailed then i might change my mind.
I think that AMD just haven't really been that impressive with consumer products lately and people are disappointed. RDNA2 and Zen3 were really great, however while Zen 4 was relatively solid, it was also at the cost of increasing TDP by a lot and was sullied by an expensive AM5 platform. RDNA3 was a flop, RDNA4 looks like another flop or maybe an OKish midrange thing, and people were hoping for Zen5 to be totally amazing and it just isn't.That seems extreme.
It's a solid performance bump with no full shrink. Big-picture uarch changes do not necessarily buy a ton of performance on their own for free - just look at Golden Cove, which was arguably a bigger big-picture change than Zen 5 and had a new process, but still only produced medium-sized iso-clock perf changes. 15% with an apparent efficiency improvement is entirely respectable.
I'm not sure Zen5 is at 15% in GB if we exclude tests that benefit from AVX-512 (which I'll say again is the feature I'm the most interested in) and take into account a 1-2% clock increase. And clang is only ~10%. But that's only a dot, and we'll have to wait for way more benchmarks to be released to get a more accurate pictureThat seems extreme.
It's a solid performance bump with no full shrink. Big-picture uarch changes do not necessarily buy a ton of performance on their own for free - just look at Golden Cove, which was arguably a bigger big-picture change than Zen 5 and had a new process, but still only produced medium-sized iso-clock perf changes. 15% with an apparent efficiency improvement is entirely respectable.
Zen4 has AVX-512 so it is fair game.I'm not sure Zen5 is at 15% in GB if we exclude tests that benefit from AVX-512 (which I'll say again is the feature I'm the most interested in) and take into account a 1-2% clock increase. And clang is only ~10%. But that's only a dot, and we'll have to wait for way more benchmarks to be released to get a more accurate picture
I'm not sure Zen5 is at 15% in GB if we exclude tests that benefit from AVX-512 (which I'll say again is the feature I'm the most interested in) and take into account a 1-2% clock increase. And clang is only ~10%. But that's only a dot, and we'll have to wait for way more benchmarks to be released to get a more accurate picture
Intel did 18% from SNC to GLC by blasting the core area (with L2) from 4.36mm^2 for SNC to 7.53mm^2 for GLC, that's a seventy-fricking-two percent increase in area. And don't get me started on power.Why you so bitter ? Intel did ~20% of IPC twice going from ~ Zen3/4 resources to ~Zen5 resources. A14 onwards exists , AMD has been delivering good Gen on Gen Performance. There is plenty of justification to expect more then what we are getting.