Looking forward to watching the show :)
Printable View
Looking forward to watching the show :)
leaky leaky
shows up to 20-27% improved single thread and 15% multi thread.
where's my dump truck full of salt.
I'll bet 8%-and up to 22% on averaged depending on load.
so close right now
I think the highest percentage of gains your going to see will be in games due to latency improvements and cache changes.
I think the lowest will be in multithreaded applications.
I think the highest single core gains will be highly dependent on clock speed therefore the gains can be skewed greatly and to truly find the actual IPC gain over previous you must run a fixed clock to measure the IPC gain and then after that you can measure the clock speed gain and how much that impacts the results over what previous gen could never attain.
To simplify that the actual core IPC is probably 10% IE IPC only gains. Add in a clock speed bump and that's a total of 15% gain over ryzen 3000
When going to single threaded 10% ipc + 10% from an even higher clock 20-27%% now seems far more realistic. Example CB single core avg 4.1-4.2 on my 3800x but 8m pi 4.5-4.6 Bump that 4.6 single 8m to 5.0 peaks. Bump that 4.1 CB single to 4.5 and you get your 20-27% range.
In games I was running between 4.3-4.4 ( 3800x ) avg monitoring all cores in afterburner.
Utilization is rather moderate 25-30%.
Based on previous per ryzen generational gains one could gather that a realistic and unoptimistic number will be 4.5-4.6 in games.
Food for thought.
Waiting with anticipation for the show.
I will gladly make a downgrade from 5960x to 5950x :).
As for the stuffz, something tells me teh stars are aligned for amd this time.Intel fumbles, nvidia kind of went retarded with samsung foundry choice.Seems like there is high probablity that amd cpu+amd gpu while maybe not absolute fastest, will make most sense price/features/performance wise.
I'm ready! Looking forward to some post on the new hardware as well.
hmm branch prediction changes davis ;)
Not much in advertised clock changes. Guess we will have to see what the cpu actually runs at clocks wise in apps and games vs previous.
According to the stream the 5900x ( 3900x replacement ) is the new gaming cpu which brings some hmmm factor to the table. Previously 3800x was the best performer in games.
Makes you wonder if the cache changes benefit the 8+ core variants far more than 8 core and lower variants.
I wounder if yields are bad, the 5800x is really expensive for the 8 core 1 core chiplet. 5950x seems cheap too I would have expected 2x the 5800x cost.
The problem with the 12 core is that I would assume is a 6+6 config, and that would lower gaming performance if you dont need the extra cores for that game. Maybe they are 8+4? In the EU people were selling 4700g OEM try chips for the same price as the 3800x. They must be expecting huge demand for the the 8 core.
I am a doof. The 8 core is a 4+4, they spent all of that time on the new 1 chiplet is one CCX then have it listed as 2x16MB cache
Here's what is sketchy to me the most.
3900x and 3950x were slower in gaming than 3800x. All claims are for 5900x and 5950x gains are versus 3900x and 3950x. No claims to gains were made vs 3800x/5800x.
Now I'm not saying there are not going to be gains but I imagine the gains are less and in marketing higher percentage gains matter.
So while marketing is saying we gained x% in games 5900x and 5950x.
My interpretation is we fixed a performance penalty we were having specifically with our 12 and 16 core models in games and slightly improved the IPC of all our models in all applications.
The 8 core APU had gains from their essentially unified cache. You can pick a 4750g up from some one who ordered a tray for around $400.
Did anyone confirm the 5800x is one cpu chiplet? They have removed the cache structure from the amd site and only toms has that cache chart up still.
I guess I'll have to wait till November 5 to know if Z3N has AVX 512 :/ there was no mention of it in the reveal.
Geek bench 5 results are up 5900X is showing more single core speed than 5950x.
Cache to core ratio is better for the 5900x.
interesting.
Performance increase in GPUS are much more than CPUs and that makes me feel like getting a better GPU than CPU+GPU Combo.
GPU wise it's looking pretty good, with the leaks that is lol.
Rdna2 Ray tracing is like raytracing 1.5 in-between Turing 1.0 and Ampere 2.0.
One more day.
The 3070 looks unimpressive for the price so AMD has a nice gap to jump in if they can scale for once.
My workstation use case has changed to where I actually need 128 threads and 512gb of memory on a desktop so im watching quite closely.
it looks like Z3N has an infinity link from CCD (Die or/ chiplet) to CCD. which the 3,000 series lacks I expect Threadripper 5,000 to have more than what the desktop has. the desktop only has two CCD's(two dies) with one IOD (I/O die), where Threadripper has two setups 4 CCDS and 8 CCD's with one IOD also linked all those CCD's on the substrate, it's going to need more infinity links. Hopefully they don't just do hops with single link groups of CCD's. there cases where the old 2000 series are better at other things than the 3,000 series because they have more infinity links to each (CCD) chiplet
on a side note The RX 6900 XT is nice for the price. pretty good under cut but....
at this the RTX 3090 you are paying for the ability to link two in SLI with NV for bragging rights
However it's still a feature of the card that $500 premium probably won't be moved because of that.
Big navi looks like it costs too much at $50 less than the 3080 for a similarly cut down chip. That seems like it is not cheap enough especially with the lack of ray tracing numbers in the presentation and the typically bad cooler but good pcb amd has on the reference card.