- Mar 3, 2017
- 1,687
- 6,237
- 136
No one's ever said no to more humiliation for Intel. That is the difference. Let Intel know their great comeback was in vain. Keller's projection is only projection but he's likely to have a good idea of where AMD will stand to Intel.Zen4 already crushes Sapphire Rapids in SPEC 2017,
Could this be unified memory?View attachment 79110
View attachment 79111
Is this Strix's LLC/IFC?
Quite intriguing that they designed the LLC with CCS interface. It can plug right into a SDP instead of UMC or can plug into a GMI interface.
Really interesting as well to see if console APUs go for this.
Could be an adaptation of MCD?
20230105709 : CACHE ALLOCATION POLICY
CACHE ALLOCATION POLICY - Advanced Micro Devices, Inc.
<div p-id="p-0001">A cache includes an upstream port, a downstream port, a cache memory, and a control circuit. The control circuit temporarily stores memory access requests received from the upstreamwww.freepatentsonline.com
This is nothing more than just main memory attached to the fabric. But it is coherent being an LLC/CCS.Could this be unified memory?
Your efforts are wasted on these forums mate... a Mills & Boon book series is calling 👍The infamy surrounding Raja travels with him heavy like the scent of a moonflower in the summer evening intoxicating all around it with its simplicity through uniqueness.
I don't remember writing such flowery prose yesterday but I also found myself "sampling" a bottle of wine via inhalation through the mouth and down the gullet.Your efforts are wasted on these forums mate... a Mills & Boon book series is calling 👍
LNC is rumored to be 8 wideIt looks like all agree with me Zen5 will be 5-wide but with a proportional increase in ALU to match.
Assuming AMD solves the renamer bottleneck by expanding it from 6 to 8; and with some more tweaks it might be possible to expand the number of ALUs from 4 to 6. Unless some is diverted to AI/ML accelerator...
Either way load and store bandwidth need an expansion especially Infinity Fabric regarding the 16c CCDs. 2.5x would be perfect.
So the projection could indeed be on the low end. Considering the work, risks, pressure for early release and follow-on Zen5+ (mobile)refreshes it makes sense for vanilla Zen5 to be conservative.
I wonder if Intel will make a "surprise" move by having Nova Lake feature 8-wide decode like Apple's M1 and Tenstorrent's.
Is this based on the raichu rumor you posted a while back?LNC is rumored to be 8 wide
Going back to what I said, I'd worry about the longevity of that memory next to the CPU die/dies. Air and liquid coolers exist, but memory behaves worse the hotter it gets, and being right on the package you're looking at more frequent activity in conjunction with the heat of the compute dies. A larger package may result in more cores for mainstream consumers if there's enough room. The only real issue I can other than what I've said is people not getting the install right leading to uneven cooling.It would be nice if future chips included 4-8gb of high speed/low latency memory on the package somewhere, even if it is “just” (L)PDDR5.
MoP and PoP designs run just fine even with high temperatures. Just because e.g. an M1 consumes less power than an x86 desktop chip doesn't mean it runs cool locally. And phones literally stack the memory on top and then use thermal throttling to keep it in check.Going back to what I said, I'd worry about the longevity of that memory next to the CPU die/dies. Air and liquid coolers exist, but memory behaves worse the hotter it gets, and being right on the package you're looking at more frequent activity in conjunction with the heat of the compute dies. A larger package may result in more cores for mainstream consumers if there's enough room. The only real issue I can other than what I've said is people not getting the install right leading to uneven cooling.
I don't think LPDDR is differentiated enough by itself. The latency is more or less a wash compared to typical DDR, and the bandwidth difference isn't terribly compelling for CPU-centric use cases, particularly considering the cost.It would be nice if future chips included 4-8gb of high speed/low latency memory on the package somewhere, even if it is “just” (L)PDDR5.
It does run cooler, I never said how cool. RAM in stick form is temperature resilient until it's not. How would such a setup be performance wise in say a 13900K if it were designed to be like that unhinged at 350+ watts? Unlike you or some here involved in x86 I've never worked on x86 and am presuming the worst performance figures under the worst conditions. I work on far less important processors. Very mundane stuff. It's humbling. It sounds comical but I never had a big interest in x86 to begin with beyond consumerism. I was hours away from signing an agreement with a quaint little company that loves the color blue but decided to go with a better offer elsewhere. Zero regrets 18 years later.MoP and PoP designs run just fine even with high temperatures. Just because e.g. an M1 consumes less power than an x86 desktop chip doesn't mean it runs cool locally. And phones literally stack the memory on top and then use thermal throttling to keep it in check.
Absolute power consumption doesn't really matter in this case. Mobile chips consume less power, but they have even more anemic coolers, so often they run hotter than a typical desktop chip. Pure temperature would not be a dealbreaker for memory on package.It does run cooler, I never said how cool. RAM in stick form is temperature resilient until it's not. How would such a setup be performance wise in say a 13900K if it were designed to be like that unhinged at 350+ watts? Unlike you or some here involved in x86 I've never worked on x86 and am presuming the worst performance figures under the worst conditions. I work on far less important processors. Very mundane stuff. It's humbling. It sounds comical but I never had a big interest in x86 to begin with beyond consumerism. I was hours away from signing an agreement with a quaint little company that loves the color blue but decided to go with a better offer elsewhere. Zero regrets 18 years later.
That's what intrigued me. I was in an IRC chat which I won't explain because it probably predates your generation but someone candidly mentioned that DDR5 gets hot. I looked up some figures a few days ago once I'd remembered what was said and I see an upper tolerable temperature to 80C which seemed fine because any cooler would be able to draw out the heat especially if the RAM on an package solution was placed on the perimeter whereas the most concentrated heat would be coming from the compute dies. Has any odm come up with a way to incorporate their laptop frames and metal panels into the cooler by way of pads to transfer heat or is it a futile exercise?Absolute power consumption doesn't really matter in this case. Mobile chips consume less power, but they have even more anemic coolers, so often they run hotter than a typical desktop chip. Pure temperature would not be a dealbreaker for memory on package.
Dude, IRC (the "chat" is part of the acronym...) isn't some lost art form, lol.I was in an IRC chat which I won't explain because it probably predates your generation
Yes. Fanless laptops typically feature a thin heatspreader or vapor chamber bridged to the chassis by a pyrolytic graphite thermal pad. The chassis itself is what does most of the heat dissipation.Has any odm come up with a way to incorporate their laptop frames and metal panels into the cooler by way of pads to transfer heat or is it a futile exercise?
I could see Frore's AirJet coolers being used on DIMMs and M2 SSDs in the future.It does run cooler, I never said how cool. RAM in stick form is temperature resilient until it's not
YesIs this based on the raichu rumor you posted a while back?
A 30% per core improvement should put AMD in the lead for single core IMO. I'm guessing LNC IPC is going to be more in the league of ~15% IPC, the same per core IPC improvement we see with previous 'new' Intel architectures. But clocks should be much, much better with ARL. ARL is on Intel 20A for their compute tile. On desktop it's 3nm... This is facing the 4nm of Zen 5.Going by what I remember from the Intel rumors Zen 5 should leave Intel's server, workstation, mainstream and mobile dead in the water unless they can pull off one big surprise.
Starting with Ice Lake, HP's consumer laptops seem to be using the metallic body to dissipate heat. The underside gets appreciably warm. All the Core i5 Ice Lake, Tiger Lake and Alder Lake HP laptops my company has bought, have been metallic. The only plastic HP laptop I've encountered is my own Tiger Lake one based on i3-1125G4, presumably to save on cost and also because this CPU doesn't run as hot as an i5 or i7.Has any odm come up with a way to incorporate their laptop frames and metal panels into the cooler by way of pads to transfer heat or is it a futile exercise?
You would be shocked how many people don't know what it is or ask if it's really still around. People are too used to AOL's chatrooms.Dude, IRC (the "chat" is part of the acronym...) isn't some lost art form, lol.
First "Zen 4 leaves Intel's server, workstation, mainstream and mobile dead in the water unless they can pull off one big surprise".Yes
A 30% per core improvement should put AMD in the lead for single core IMO. I'm guessing LNC IPC is going to be more in the league of ~15% IPC, the same per core IPC improvement we see with previous 'new' Intel architectures. But clocks should be much, much better with ARL. ARL is on Intel 20A for their compute tile. On desktop it's 3nm... This is facing the 4nm of Zen 5.
For server, RWC in GNR should be considerably less powerful than Zen 5, on a per core basis. However, if RWC+ is a decent gain. or Lion Cove is used in GNR, then the difference shouldn't be too bad. And this isn't just fanfic either, Gelsinger himself said GNR uses a 'new core' and GNR as a product changed so much they wanted to change the codename of the product. Plus, server also has some reprieve, it should be on Intel 3, which at the very least should be better than the TSMC 4nm node Zen 5 uses even if it's not a true 3nm class node.
Nothing should be left 'dead in the water', nor does Intel really need a 'big surprise'. As long as LNC is a run of the mill architectural uplift, like GLC was or SNC was before it, Intel should be fine, if not in the lead but competitive. At the very least, it should be in a much better state they are right now, esp in server lol.
Lastly, this doesn't even hinge on Intel's own internal nodes, since rumors claim they are contracting TSMC for 3nm. All the better nodes should allow for, both in desktop and server especially, is better clocks in clock limited workloads (Server and desktop MT) and potentially better ST as well, though as we have seen with Intel 14nm, and now Intel 7, that's not a guarantee lol.
I've been a Dell XPS customer for a long time and it's the only laptop refresh I buy into because of its looks and size. I've never opened it before which is why I asked. The last laptop I opened up was an Aspire 7 which is simple but takes a lot of screws to open up. If I need work done on my XPS laptop I image the computer and wipe it before sending it in.Starting with Ice Lake, HP's consumer laptops seem to be using the metallic body to dissipate heat. The underside gets appreciably warm. All the Core i5 Ice Lake, Tiger Lake and Alder Lake HP laptops my company has bought, have been metallic. The only plastic HP laptop I've encountered is my own Tiger Lake one based on i3-1125G4, presumably to save on cost and also because this CPU doesn't run as hot as an i5 or i7.
That's a great idea, except I don't see OEMs wanting to pay royalties to Frore.I could see Frore's AirJet coolers being used on DIMMs and M2 SSDs in the future.
Unrelated but do we know if any company bought up the 2 months of tsmc 5nm apple halted or did they retain 2 months of production?As far as TSMC 3mm for Intel, lets talk about that once we at least have ES leaks.
I recall this. Was it in the news this January or on Linus Tech Tips?I have to look into it and see how it works but I'm guessing it sucks in air from outside the chassis and blows it onto the chip?I could see Frore's AirJet coolers being used on DIMMs and M2 SSDs in the future.
The 'Pro' model of their 1st gen MEMS device can move about 10W worth which should be enough to cover both applications.
Intel being competitive is what matters here if they can't gain over AMD. Intel's outstanding contracts will carry until they earn out. At which point the Intel hardware needs to be both competitive and use power in line or better than AMD.Yes
A 30% per core improvement should put AMD in the lead for single core IMO. I'm guessing LNC IPC is going to be more in the league of ~15% IPC, the same per core IPC improvement we see with previous 'new' Intel architectures. But clocks should be much, much better with ARL. ARL is on Intel 20A for their compute tile. On desktop it's 3nm... This is facing the 4nm of Zen 5.
For server, RWC in GNR should be considerably less powerful than Zen 5, on a per core basis. However, if RWC+ is a decent gain. or Lion Cove is used in GNR, then the difference shouldn't be too bad. And this isn't just fanfic either, Gelsinger himself said GNR uses a 'new core' and GNR as a product changed so much they wanted to change the codename of the product. Plus, server also has some reprieve, it should be on Intel 3, which at the very least should be better than the TSMC 4nm node Zen 5 uses even if it's not a true 3nm class node.
Nothing should be left 'dead in the water', nor does Intel really need a 'big surprise'. As long as LNC is a run of the mill architectural uplift, like GLC was or SNC was before it, Intel should be fine, if not in the lead but competitive. At the very least, it should be in a much better state they are right now, esp in server lol.
Lastly, this doesn't even hinge on Intel's own internal nodes, since rumors claim they are contracting TSMC for 3nm. All the better nodes should allow for, both in desktop and server especially, is better clocks in clock limited workloads (Server and desktop MT) and potentially better ST as well, though as we have seen with Intel 14nm, and now Intel 7, that's not a guarantee lol.