- Mar 3, 2017
- 1,687
- 6,237
- 136
SoIC is now both bump- and bumpless-3D.but I don't know if they use SoIC or some micro-bump tech?
Yea.It is a specialized cpu chiplet
No.I have to wonder if they would make use of this somewhere else though
Yes.So are they really only making this special chiplet for MI300?
No, Zen5 brings quite literally 0 system design changes for parts not named Strix-Halo.Perhaps this will change with Zen 5
Man you're kinda-sorta-almost getting it.It would be nice to be able to get essentially 1/4 of an MI300 as a desktop or mobile part with up to 3 cpu chiplets
Well, 3D stacking is decidedly not mainstream, yes.AMD seems to not be planning on using much stacking, other than v-cache, for mainstream parts.
Are these two statements related?No, Zen5 brings quite literally 0 system design changes for parts not named Strix-Halo.
Man you're kinda-sorta-almost getting it.
Wrong focus though.
More about how Zen6 looks.Are these two statements related?
Yea, all the system design changes are loaded into Zen6.Are these two statements related?
Yea, all the system design changes are loaded into Zen6.
STX-halo is the only Zen5 part that gets the shiny new blingee (2.5D fanouts, USR links, SLC, yaddayaddayadda).
Surely Zen 6 can't be cheap with all this fancy packaging tech. Heck, @adroc_thurston , wasn't it you who said that MTL was going to be pricey for Intel to manufacture because it uses so much packaging?More about how Zen6 looks.
Fanouts are cheap.Surely Zen 6 can't be cheap with all this fancy packaging tech
Many tiles, and a real 2.5D slab.MTL was going to be pricey for Intel to manufacture because it uses so much packaging
What's the difference in latency between MALL SLC and Zen L3, if it is good enough for GPU, but not good enough for CPU?Yea, all the system design changes are loaded into Zen6.
STX-halo is the only Zen5 part that gets the shiny new blingee (2.5D fanouts, USR links, SLC, yaddayaddayadda).
That's what we have to find out.What's the difference in latency between MALL SLC and Zen L3
Ehhh no one said that.but not good enough for CPU?
Yea, RDNA3 was made to run nutso speeds and the thing hasn't been canned so nutso speed it is.Can It clock that high or even higher while also feeding 16 cores with 125W TDP?
I hope so. This would bode well also for Strix Point, but there BW could be a problem.Yea, RDNA3 was made to run nutso speeds and the thing hasn't been canned so nutso speed it is.
I hope so. This would bode well also for Strix Point, but there BW could be a problem.
BTW, you meant RDNA3.5, right? Because RDNA3 is not achieving nutso speeds.
I see R3.5 the same way I see Vega 2.RDNA3 can do nutso speeds at nutso power draw. I presume 3.5 just fixed whatever it was that prevented nutso speed from running at sane power draw.
I also presume 3.5 would not have been needed at all if 3 hit it's original targets.
Was always a thing.I also presume 3.5 would not have been needed
Vega was a bad uArch.I see R3.5 the same way I see Vega 2.
R3 was just malformed.
RDNA3 based products didn't show us a great showing compared to RDNA2.Vega was a bad uArch.
RDNA3 isn't really that.
I remember you warning me against buying RDNA3 because I was set on getting it.RDNA3 based products didn't show us a great showing compared to RDNA2.
Unless RDNA3.5 in Strix Halo(Point) shows us some significant improvement in perf/W, then RDNA3 won't be remembered in a positive light.
Duh.Unless RDNA3.5 in Strix Halo(Point) shows us some significant improvement in perf/W, then RDNA3 won't be remembered in a positive light.
Sorry, I don't remember. From AMD I would consider only 7800XT and can't say I am impressed.I remember you warning me against buying RDNA3 because I was set on getting it.
It will have the best IGP performance, but the question is how high is that performance and at what TDP.Duh.
STX-halo has no reason to exist without banger iGP perf.
The thing is designed for above-AD106 levels of perf, it's a lot-lot faster than N33 (or should be).in other words N33 performance level
It's not N4.4nm 40CU IGP
Not made to run those speeds *at all*.at 2GHz
Nobody, the 7950X is all the CPU everyone will ever need.
That sounds good, but I want to see some tests first.The thing is designed for above-AD106 levels of perf, it's a lot-lot faster than N33 (or should be).
Then N3.It's not N4.
For 4060Ti level of performance, It would need ~2.5GHz.Not made to run those speeds *at all*.
It was in the GPU section of the site. I only remember because your user name sticks out.Sorry, I don't remember. From AMD I would consider only 7800XT and can't say I am impressed.
Yea but you get more SLC bandwidth.it is comparable to what N33 has and memory is shared with the CPU.
Ahhh taping out another CCD with only 8 Zen5c is a major effort since it's a totally new floorplan and not a low-effort IFOP replacement.I am surprised they didn't save some space by using 8 Zen5 and 8 Zen5c cores
It's a CoWoS'd laptop chip and those LITTLEs aren't for nT spam at all.
Then why is AMD making 16C AM5 desktop CPUs? They should go for only 1-8C CPUs with Zen5 on AM5. Then 8C+ on Threadripper SP6 only.Because most people doing DYI PCs do it for gaming, and gaming perf doesn't require 16c parts
So all laptops are poverty segment? You do know many laptops are more expensive than desktops right? E.g. Mac Book Pro for $2000-2500. And yes, intentionally with b.L, since it's a better design.Poverty segment where costs matter.
People have to justify their purchases.Then why is it being brought up as a drawback for Intel compared to AMD, that Intel has higher power consumption?
It's a two M2 Max laptop chips glued together over a InFO-L substrate.No, the Apple M2 Ultra is not used in laptops.
Marketing.Then why is AMD making 16C AM5 desktop CPUs?
That is what sells yes.They should go for only 1-8C CPUs with Zen5 on AM5.
YES!So all laptops are poverty segment?
I do procurement, you don't have to tell me that.You do know many laptops are more expensive than desktops right?
The chip cost there is tiny relative to machine total BOM.E.g. Mac Book Pro for $2000-2500
It uses the same 4c Blizzard cluster as everything else Apple A15 gen.And yes, intentionally with b.L, since it's a better design.
You can pay for that in desktop.And have you noticed how many poverty desktops there are?
Because desktops can pay.Why is AMD not using b.L for them, if you claim b.L is only used to lower cost?