AMD Desktop 8000 series

There is no rumour with 40% IPC gain from my understanding. 40+% performance gain in one specific benchmark (SPECint Rate 2017 1T) is the thing that is going around. But it seems that some commentary in the media is just running with 40% IPC... ie https://www.tweaktown.com/news/9734...uplift-will-not-be-close-40-rumors/index.html

 

yeah, but it's still not 100% confirmed, and you can be sure people are treating it as 40% across the board, end of story

 

I agree, it'll get trolled by the zealots regardless of any disclaimers placed on it.

Even 20% seems more than respectable which would be lovely

 

The 40% uplift in AVX FPU workloads is totally believable. They have doubled the size of the AVX unit to 512bits. So its more than 100% faster. In CPU 3d rendering, media encoding, AI workloads, 40-100% performance uplift is entirely possible.
However, games, office programs, window boot times, will likely not see any benefit from the more powerful FPU. Maybe a 10-20% IPC improvement from other things they have done like the 6 ALU, better caching improvements in decoding etc.

https://www.hwcooling.net/en/amd-confirms-zen-5-details-6-alus-full-performance-avx-512en/

THat is confirmed.

So for content creators, this is huge. Yeh, a 40% uplift in your actual application speed is massive. AVX2 has been around for a long while now, so the software support is now pretty good, its just CPU manufacturers didn't fully commit to it, because it was not clear if it was going to be a big thing, 5 years ago when they started taping this stuff out. AI was extra icing, as well, as that is now the biggest thing driving the tech industry.

But for gamers, no. Most games avoid using the FPU at all times, because it slows things down, and since 3D graphic card do all the 3d caculations they haven't since Unreal/Quake.

Except perhaps for games like factorio, some other strategy games. But mostly they still use ALU because its much faster, and mostly they are memory dependant as well.

 

Seems caveats are coming out now. The SPECInt performance rumour comes from serverside cpu's where they have been sampling with partners. Seems the results are +50% on this particular benchmark but at +25% greater power usage.

 

Realistically does that mean that current gen might not be that efficient but further revisions should see that reigned back in?

 

Not sure mate. I guess there is a question mark whether or not the bench was run on an engineering sample or not. 50% performance gain (in this one benchmark) at 25% power increase is pretty tidy anyways. AM5 socket supports 170W TDP and the 7950X is rated at this TDP, so AMD can't increase power on the desktop like they can on the serverside which is up to 500W apparently.

There is also a question mark whether the latest rumour update is 1T or nT for this bench. nT could be power limited, 1T would be clock limited - but I'm not familiar with the power characteristics of this benchmark.

Edit: its nT (96 core Turin @500W vs 96 core Genoa @400W) ... so slightly different performance rumour over the previous one which was 1T.

 

9000 seems to show a 22% increase in single thread cb24, and 46% increase in multithread.

https://www.tweaktown.com/news/9749...faster-in-cinebench-r24-than-7950x/index.html

Totally believable. Similar gain from zen3 to Zen 4 as Zen 4 to Zen 5. Particularly in fpu heavy workloads.

Makes my 5950x rather pathetic. A 8 core Zen 5 will probably be faster than my 16 core CPU.

Which makes a 8 core 3d cache very tempting. Also with am5, I will go with 4 dimm slots and 196 to 256gb ram making it a 400% improvement in dram capacity.

I am also finding now that pretty much all my software supports gpu acceleration. So my need for mega cores is reduced.

 

Extremely impressive performance uplift if those numbers are true can carry through to other apps.

 

I'm not sure AM5 likes 4 dimms full, for ram speed. I know AM4 doesnt.
Hell even my old socket 939 system didnt

 

If true it could be the first upgrade I do with fewer cores, 12 core to 8 core. Maybe with a bump to the 3Dcache chip next year. Like others GPU accel is reducing my need for moar cores.

 

It won't run at crazy fast speeds. something like 5800 Mhz with 196Gb is achievable on AMD systems currently with 7000 series CPU's.

The whole reason I avoided first gen DDR5 is because of the issues with multiple dimms. Which are now mostly solved. 196GB+ Is a daily event for people on both Intel and AMD platforms, and runs reliabily.

40%+ It is very much likely to be only for multithreaded FPU apps. Encoding, rendering, scientific calculation, AI. But I do use cad, my wife is a materials scientist, I teach and encode a lot of videos. So for people like us a 40% improvement is real, and is huge. Some specific media codecs or scientific work it may be greater than even 40%, perhaps even greater than 100%.

Regular IPC is more like 10-15%. Which is still very good. Particularly if you get 3D V-Cache with it.

I did go from a x6 1055t to a 4790k. So 6 to 4 before. It was still a pretty good upgrade, the the 4 cores were a lot faster than the 6, even its multithreaded performance was lifted up.

So yeh, I think for me, 8 cores will be enough if they are fast enough. 16 cores is nice, but unless they sort out the vcache (one for each die), its not attractive to me. 8 Cores is still a heck of a lot of power, even for professional apps that often don't scale well over more than 16 threads.
I am also ditching mITX> too small, particularly for a 170w+ CPU, even water cooled. Matx is just that little bit bigger, and more sensible if you want to run something like 5090 GPU and 8 or 16 core 170w CPU. Plus 4 DIMMS.

The money I don't spend on CPU I can spend on RAM, Disk or GPU. For those, capacity is more important than speed. Some software recommend 24Gb of VRAM, minimum, and hundreds of GB of system ram.

 

I just wanted to say how impressed I am with the 8700g graphics . I am always using a 1080p high refresh monitor and the apu goes well with it . for games you can also use the AMD super res and software will auto scale it to 1080p but keep the fps 30~40% higher!

 

Hi all, I was recently obtained (i.e. free) an X670E MB. Currently, I am well set with the rig in my sig (hah!) and loath to endure the process of reinstalling software with a new build. I am however in the process of revisiting my home-server Kaby lake build.

Q: Where can I find the 8300G? I would like a super-duper cheapie AM5 CPU to play with. Is it worth the platform cost to finish this build or will I be better off selling the MB?

 

Rumour: Granite Ridge and Strix Point to be announced on June 3. Granite Ridge to be available in July with Strix Point in August. Strix Halo is probably Q1 2025.

 

mmm that 8 series is looking appealing.
ive been totally alseep wrt cpu's. i didnt even notice the 7000's . but im thoroughly satisfied with my current AMD baby. really should upgrade the sig....

AMD ontop. wow. it's been a long wait.
now what would really rock my world is an AMD sys with an intel grfx card i wonder if they'd play together?

 

Why?

 

boredom

variety in the product space.

 

that mobo and 8300G combo is for the home server ? depends what your needs are really. if you've got the mobo for free then thats already a substantial saving for the build.
spend up on decent ram and a 8700G maybe

 

Is this the one with decent onboard graphics?
I'm kind of wanting to upgrade my HTPC so I can play games etc on it as well.... not sure I want to wait that long though