So you think those dies are legit, or are just basically empty spacers/supports?
I don’t see why it would be an issue for them to just put something in there to act as a spacer. I can’t see why they would put good dies in there and deactivate them or something. Seems wasteful as all hell.
When the processor is released, reviewers will be able to use that simple core to core time test that PcPer made to ensure that only 2 ccx’s are being used. I’m sure AMD is being true to their word and only using two, since they know they’d be caught red handed if they were trying to use ccxs (even if it really isn’t that big of a deal anyways). It seems like if there was an internal miscommunication between AMD engineering and AMD marketing, and it had been falsely communicated that 2 dies instead of 4 were going to be used, that AMD would have came out and clarified by now. But considering that no such correction has come forth, and that AMD has reaffirmed that only 2 ccxs are being used, I’ll just assume that those two ccxs were maybe bad or just had issues hitting particular performance per watt metrics and were thus just used as supports. I don’t know of any other logical reason to include them, but who cares anyways?
Stupid question, what is CCX short for?
CPU-Complex?
It’s the groupings of 4 or 8 cores that AMD is combining in pairs or sets of 4 to create their entire lineup of processors.
Interesting.
I hope they do well, and I really hope AMD doesn’t get shafted by poor quality motherboards from the manufacturers. The X299 debacle was hilarious, but Intel can afford to take the hit.
Soldering will help transfer heat and thus keep the socket temperature cooler, the socket is larger and will spread the heat out more effectively, and the processors are more power efficient. Hopefully the manufactures think about their heatsink designs more after dealing with the X299 screwups. Fingers crossed they do the VRM justice, but the socket being so large does limit the space available to fit the VRM components in.
Kind of in a weird reverse way, the manufacturing process being employed and its inherent inability to consistently well above 4Ghz on all cores also means that the amount of power the vrm will ever need to provide will probably be lesser than the more power hungry and higher clocking Skylake-X competition.
Honestly, I cannot wait to see the custom waterblocks. Gonna be awesome looking with Acrylic and whatnot.
Fullcover blocks on thread ripper will indeed be very cool. The prohibitive pricing also means that probably a higher percentage of users will be custom waterloop guys. The boards are going to be expensive considering the socket size, power needs, and amount of pcie slots that are possible due to the 64 lanes.
But the heat spreader all ready has the bar in the middle to stop the flex or sag, so having dummy chips for that reason make no sense. it is glued all around the edges and in the middle where the support is for that very reason.
Throwing in an @anon43920604 as well as you were part of this line of talk too.
EDIT: As for the why they are there. They could be failed cores on the dies but the interposer might still need them to be there to function, as an electrical passthrough or shunt. Non core/non critical parts may remain functioning even if the logic is dead and act as just place holder to make the thing function but other wise are not actual chips. That is the only thing that would make sense to me right now.
Support for the heat spreader does not sound right. Functional dies sounds like a waste of money like you pointed out, non functional dies would not do anything any way so you could leave them off the interposer all together, so there must be a function to them even is it not compute based and the only thing left in my mind is simply electrical connection to pass through power.
They may well have started life intended to be 4 die EPYC CPUs but could not meet spec or died along the way so they laser cut the logic and left them solely as wires effectively. It would be even cheaper to just put 2 dies on there instead of 4 even if 2 are deal. less manufacturing work, less materials like solder and so on.
Though part of me still really wants all 4 to be working with just 4 cores on eact CCX for the thermal balancing. they may not clock great but with that config you could pin them at 4.1 or 2 all day long and not hit any worrying temps on all 16 cores.
Didn’t know this. Maybe they want more support?
Its supported in the middle yeah but force on the other side of the bar could easily cause flex…also glue around the edges? Wut. Yeah of course it is glued there. Glue doesn’t support anything lmao
No but it could in some way resist force in the middle. Although I’d wager that amount is negligible.
Harder to push the middle in if the edges don’t move along with it?
Ehh I think they will be cheaper than you think.
Yeah lots of PCIE lanes but compared to X299 the chipset is significantly simpler. Plus no dual support for different family CPUs.
X299 boards have significantly more power delivery components on them then is necessary to run Skylake X because they are required for Kabylake X.
Tbh I think it will all balance out. Prob fairly similar in cost to X299 maybe a little bit more.
As for VRM heatsinks photos from computex and current product photos show that the VRM cooling on pretty much every X399 board has been changed and made better. Not to mention I think under most loads TR will use less power.
I guess. But the middle isn’t the issue. The middle is supported but there is a shit ton of space between that bar and the edge. You’d need something there.
I am firmly betting on spacers to prevent flex or plug holes in the package.
What is AMD charging for a threadripper chipset? And what is intel charging for a X299 chipset? If I remember right (which who knows anymore), I thought intel was pretty steep on the enthusiast platform chipset costs. Since Zen is only a couple steps away from SoC status, I bet the chipset has to be pretty cheap.
Edit, X299 is $74 per chipset. Can’t think AMD is charging that much.
I was going to try and make an analogy to medieval buildings having larger floors on top to prevent the floor joists from sagging (like a counter weight) but yeah I get what you mean.
(push down on the outer edges, or in this case resist pull, to prevent flexing in the middle)
Also IIRC the socket puts force on the outer edged of the head spreader anyways so… yeah.
Yeah exactly. I couldn’t give you specific numbers but if imagine X299 is very expensive.
The chipset needs to do so much work because they nerfed the PCIE connectivity and the double nature of having to support Kabylake also drives up price.
Plus there is the cost not just of the chipset itself but actually the design work and manufacturing cost to implement it.
On iBuyPower website you can configure and preorder TR. They have multiple options for motherboards. The top end one adds about $250.
I dont think that the issue was with the AIB regarding the Motherboards X299, i think that Intel previous to the Ryzen/TR Release had sent the specs for the New HEDT i7 that possibly had a max of 10Cores, but after realizing that Ryzen was good, they needed to step up the game, and starting Binning E5 Xeons and unlocking the Multiplier with MicroCode .
The Result?, well Smoking hot beast with high TDP and high Power Consumption, almost out of the spec of the Already engineered Motherboards, thats almost the same thing that happened to the Fury/FuryX, they thougt they were Competing with the 980, but then the 980Ti got out of no where, so AMD had to Delay the Fury Launch to bin better and try to clock balls to the wall to atleast compete with the 980ti.
That will explain why we didnt have any of the Top End HEDT CPU and why it took this much time to atleast get more information about it.
IIRC changing the heatsink fixed the majority of the issue so, I’d say a bit of both? Intel panicing and the AIB’s being crappy (ASUS in particular seems to have been having QC issues as of late)
