- Mar 3, 2017
- 1,684
- 6,227
- 136
Some parts of SPEC can be vectorized without changing sources with recent versions of non vendor compilers.
I understand that 3D chips are better for gaming, but in that barely over 1,5 years since 7950x was released, did you truly repeatedly run into situations, where that CPU was not enough, but 3D version would be?I made the mistake of going 7950X (non-3D) last time. I'm waiting for the 3D part this time around.
From Zen3 which is my main desktop it will still be a very big jump.Who is going to upgrade for 16% gains or whatever it turns out to be? If we wait for 9950X3D it'll probably be a nice upgrade.
So it is time to be patient.
I don’t know, I have 5950x, and 9950x will be something like 33% uplift on average. Not interesting to me for ~$1200 with mobo and ram. I was hoping AMD would continue the steep core count ramp, but they took the greedy path and only upgraded their most expensive parts (and jacked up threadripper prices). They’re turning into old intel and Nvidia.From Zen3 which is my main desktop it will still be a very big jump.
im far most interested in 1T or low T performance so what that is for Zen5 is entirely unclear to me right now.
should be more like 45-48% in 1T .I don’t know, I have 5950x, and 9950x will be something like 33% uplift on average. Not interesting to me for ~$1200 with mobo and ram. I was hoping AMD would continue the steep core count ramp, but they took the greedy path and only upgraded their most expensive parts (and jacked up threadripper prices). They’re turning into old intel and Nvidia.
I don’t know, I have 5950x, and 9950x will be something like 33% uplift on average. Not interesting to me for ~$1200 with mobo and ram. I was hoping AMD would continue the steep core count ramp, but they took the greedy path and only upgraded their most expensive parts (and jacked up threadripper prices). They’re turning into old intel and Nvidia.
I'm leaning towards the second one. Full 512-bit width AVX-512 implementation will likely chug power (if it's anything like Skylake-X)This thread is still moving at hype train speeds so excuse me if this has been brought up already. What is up with AMD using considerably lower base clock speeds and TDP's?
I've read two theories. One, AMD checked out this article and saw higher TDP's were largerly useless. Second, those base clocks have to account for full AVX512, which is well known to limit clock speeds on Intel chips. Therefore those are worst case base clocks and when not running AVX512 they should stay closer to the boost clocks.
Realistically a ~15% performance increase is not worth spending money to upgrade unless you have a specific bottleneck that needs to be addressed.Who is going to upgrade for 16% gains or whatever it turns out to be? If we wait for 9950X3D it'll probably be a nice upgrade.
So it is time to be patient.
1) unless that number is lowRealistically a ~15% performance increase is not worth spending money to upgrade unless you have a specific bottleneck that needs to be addressed.
Luckily for AMD, I always have specific bottlenecks I want addressing.Realistically a ~15% performance increase is not worth spending money to upgrade unless you have a specific bottleneck that needs to be addressed.
It's a vey vocal and very, very tiny minority who wants the increase in core counts on the consumer platform. Just look at sales for the current 16c. Appreciate you say you need it, but it's not really AMD being stingy, it's more like no one would buy them. So why bother? To use it you need more memory channels, why lumber the majority with these higher costs?I don’t know, I have 5950x, and 9950x will be something like 33% uplift on average. Not interesting to me for ~$1200 with mobo and ram. I was hoping AMD would continue the steep core count ramp, but they took the greedy path and only upgraded their most expensive parts (and jacked up threadripper prices). They’re turning into old intel and Nvidia.
Luckily for AMD, I always have specific bottlenecks I want addressing.
I'm the sort of a person jumping from i7 4770k to i7 4790k or Ryzen 7 3700X to Ryzen 7 3800XT ...
Tinkering with new stuff is fun!
But I agree, no sane person not caring about desktop PCs should move for just 16% uplift, but I have a feeing this is only part of the story and we might be surprised by efficiency gains on top of IPC gains.
I want to see this new family compared to predecessors in ECO mode, should be fun!
Based on what? the non existent perf/watt data in the presentation and lower base clocks?AMD should win in efficiency easily. If they don't that's two hype trains in a row derailed (RDNA3 being the last one). I just think that a lot of us were surprised since 20% was basically considered the minimum.
It's a vey vocal and very, very tiny minority who wants the increase in core counts on the consumer platform. Just look at sales for the current 16c. Appreciate you say you need it, but it's not really AMD being stingy, it's more like no one would buy them. So why bother? To use it you need more memory channels, why lumber the majority with these higher costs?
If they can find better ways to improve performance that’s fine too. But single thread scaling isn’t great.
Also, halo products are always for a tiny minority. The point is that AMD is not pushing the envelope anymore
Based on what? the non existent perf/watt data in the presentation and lower base clocks?
How is RPL or Zen 4 relevant to next generation of competition? All major competitors are now at node parity. It’s a different competitive environment.Based on recent history. Based on RPL burning so much power Intel had to provide guidelines for BIOS's so their top CPU's don't crash. Based on the fact that Lunar Lake and Arrow Lake won't come out until after Zen 5. Based on the fact that Zen 4 is already more efficient in almost every case. Take your pick.
It's called a speculation thread not a fact thread. Am I not allowed to make a prediction?
How is RPL or Zen 4 relevant to next generation of competition? All major competitors are now at node parity. It’s a different competitive environment.
Thanks for the info. It has been unclear for a while whether the IO die actually had 12 or 16 GMI links. Perhaps with 16 they can actually make some dual link parts with 8 CCD. Someone else said somewhere that they never actually made a part using dual links to each CCD. If they only had 12 GMI links previously, then it wouldn't be that useful, since you could only do dual link (GMI-wide) with 4 CCD parts. The 4 CCD parts seem like they are too low end for that to be worthwhile.There still is merely the old rumor of a 16 core complex, no public info from AMD.
AMD subsequently gave minor clarifications to the press. Renderings from AMD:
View attachment 100647 View attachment 100648 Turin classic, up to 128 cores Turin dense, up to 192 cores
(source)
Those are more interesting. I was a bit sad that 7950X3D didn’t have cache on both dies and and stability issues (and clock speed regression). When I built a rig for my sick friend, it was with the 7950X for those reasons.Ah but they are... see the x3d models. There's no real point in pushing the core count envelope in desktop when it doesn't help gaming much.
Same here. AMD should introduce 9970x with 1 CCD Z5 X3D and 1 CCD with Z5c. However for some reason there seems to be so much resistance for increasing core count in this forum. For me 4 generations of 16 cores is not a smart decision by AMD.I don’t know, I have 5950x, and 9950x will be something like 33% uplift on average. Not interesting to me for ~$1200 with mobo and ram. I was hoping AMD would continue the steep core count ramp, but they took the greedy path and only upgraded their most expensive parts (and jacked up threadripper prices). They’re turning into old intel and Nvidia.
Price determines how many cores someone buys compared to absolute specs. If it cost as much to buy a brand new 8c cpu vs a 16c one, a lot of people would indeed prefer core count over a small amount of 1t gain.Just look at sales for the current 16c.
This is literally a myth and for 99% of people running multi-core workloads or just trying to multi-task, the already high bandwidth of DDR5 is more than enough.To use it you need more memory channels, why lumber the majority with these higher costs?
Ultra high end ram tuning is even more niche than overclocking CPU/GPU. More cores would give significant gains across many workloads meanwhile ram tuning is basically only good for gaming and very niche and uncommon workloads.It's a vey vocal and very, very tiny minority