• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Anyone else disappointed with the AMD 3D-Vcache reveal?

I sold it because the 5800x is what everyone thinks Intel CPUs are. Stock out of the box it's pushed to the absolute max. Total stock would thermal throttle (90c) on a Noctua NH-D15 on CB23. It could be tamed by reducing PPT, EDC, and TDC, but what's the point of spending $450 on a CPU that you have to essentially tweak down to make functional? Performance was fine, and for someone who wants to take the time working with PBO2 and CO it's pretty tweakable, but I really don't have the patience or time for that.
You can run it off a U12S without throttling, unless you have one of the stupid motherboards that does an intel and unlocks PBO by default

The problem is that they'll sit around 80C regardless of what cooler you use, and that triggers people.
 
I feel pretty bad for anyone who picked up rocket lake after seeing how much better Alderlake is.... I think this is quite different though it's a singular sku and gives people with a 3800X or slower Cpu a really nice drop in upgrade without having to ditch their platform (B450/X470/B550/X570) assuming the price doesn't suck.... Nobody with a 10700k or 10600K should have been excited about Rocketlake and anyone with somthing older was SOL.
AL is only better with windows 11. That is if one has the 'early adopter' disease.
 
You can run it off a U12S without throttling, unless you have one of the stupid motherboards that does an intel and unlocks PBO by default

The problem is that they'll sit around 80C regardless of what cooler you use, and that triggers people.

Nope. Top end B550 board with PBO disabled. Same board runs the 5950x currently 100% load 24/7 at 60c on a POS Scythe Mugen 5. The 5800x is AMDs attempt to win benchmarks and maximize profits. The 5800x3d will be the same thing.
 
die-glue lol Remember the jokes at Intel using Glue
 
Nope. Top end B550 board with PBO disabled. Same board runs the 5950x currently 100% load 24/7 at 60c on a POS Scythe Mugen 5. The 5800x is AMDs attempt to win benchmarks and maximize profits. The 5800x3d will be the same thing.
You can get things with various names equivalent to intels MCE, i've just forgotten what they're called
My asus boards all have options along the lines of "performance enhancers" that smash the wattages through the roof, even with PBO disabled.

I'll never deny they run hot, cause they freakin do and it's annoying. But you should always be in the 70-80C range since the CPU targets that, even on a 120mm air cooler (or basic AIO)
 
They say 80 is ok.. and that's what I aim for :D

They can take quick blips past 100 no worries :D
 
I think it's a shame that they didn't make a 5950X version with 3D cache, as that's probably the CPU that hurts the most from the DDR4 memory bandwidth limitations. The more L3 cache it has, the less travels to the RAM it needs to make, so there could (should?) be a bigger boost in performance for the 5950X than there is for the 5800X.
 
I'm sure they're going to bring a 5900X/5950X 3D-cache version.
 
The only reason I am excited about Ryzen3D is that I loathe having to change platforms for new memory. I have five AM4 rigs in the home.
 
die-glue lol Remember the jokes at Intel using Glue
AMD use Stiction, no glue and way to go , please do add merryment.
Intel fu£#ing earned those memes, Not AMD, calling original Zen and epic glued parts before delivering, you guessed it , glued together part's.

It's replies like that ,with f all useful ,true or on point that gets me and you talking, do note.
 
I think the performance of Alder Lake has had something to do with this decision. AMD probably factored in Alder Lakes's performance but it turned out way better than they expected. The initial plan was to launch a full range of 3D vcache CPU's including the 5600X3D etc which would have kept Alder Lake at bay and allowed for no price drops but a small price increase so as to maximise profits on AM4, but as the performance increase on ADL was so large (12600K matching the 5800, 12700K matching 5900 all at a lower cost etc..), they have decided to focus on Zen 4 instead and just provide one 3d vcache CPU as a stop gap with the hope it slows ADL adoption for those who are on AM4 before AM5 launches.

It all depends on how well Intel can pull of the lower end where they should now have a serious advantage with the likes of the 12400 and B and H series motherboards. The 12400 at under $200 and the 12300 at $140 all the way down to the sub $100 12100F are potent weapons on the gaming front and AMD absolutely needs a range of sub $200 CPU's....Not sure why AMD gave that space up which really makes up the bulk of sales in the diy space..

Still I am sure Zen 4 will be a big uplift in both performance and efficiency and will go toe to toe with Raptor Lake....should be fun times ahead at least on the CPU front....
 
die-glue lol Remember the jokes at Intel using Glue

It always was weird though.

Because as Intel as making jokes about die-glue, they were advancing their EMIB / Foveros die-to-die technology. Everyone was moving towards "die-glue" so to speak, even Intel.
 
Remember AMD's jokes about using MCM?

1641579641573.png


'True Quad Core' vs the Core 2 Quad MCM...

1641579700998.png
 
I'm sure they're going to bring a 5900X/5950X 3D-cache version.
I'd like to think there's a substantial upgrade path to my 5900X for me in a handful of years down the road, when no one is making CPUs that support DDR4 anymore.

However, I'm not sure the 5950X 3D-cache is coming, considering Zen4 is coming in H2 2022 already and by then AMD will have high-end CPUs that support DDR5, PCIe 5.0, USB4, iGPU on desktop, etc.
 
I might get one if they are going to be a little rare. My first AM4 CPU was the 3600XT that I bought for the same price as the vanilla.. was a nice unit, not quite as golden as the one in the GN review, but it was close. I sold it last summer and I kind of regret it. I do have my 5600X but it’s not rare. I have always been curious about the 5800X anyways.. last I looked it was the same price as what I paid for my 5600X..
 
I'd be lying if I said that I wasn't bummed. I wanted something with at more than 8-cores as I already planned on doing a lot of CAD, rendering, and also to skip AM5 altogether no matter what. Currently building an ADL secondary, guess it's going to be my primary instead. I'll wait and see what a 5800X vs 5800X3D vs 5900X comparison looks like.

Remember AMD's jokes about using MCM?

View attachment 231597

'True Quad Core' vs the Core 2 Quad MCM...

View attachment 231598

Core 2 wasn't really MCM, iirc the two dies still had to communicate through the chipset like a 2p setup.
 
Last edited:
I'd be lying if I said that I wasn't bummed. I wanted something with at more than 8-cores as I already planned on doing a lot of CAD, rendering, and also to skip AM5 altogether no matter what. Currently building an ADL secondary, guess it's going to be my primary instead. I'll wait and see what a 5800X vs 5800X3D vs 5900X comparison looks like.



Core 2 wasn't really MCM, iirc the two dies still had to communicate through the chipset like a 2p setup.
1641615022109.png


Still technically an MCM - even without the direct chiplet communication.
 
Last edited:
Honestly I was fully prepared to flip my 5950X and get a 5950X3D, but well, AMD made the decision there for me.

I'm sure they're going to bring a 5900X/5950X 3D-cache version.

Honestly? If not now, probably not worth ever doing it. With Zen 4 shipping this year and being free of the limitations of the aging AM4 platform, might as well not bother. Intel prices are excellent, and that is bound to improve further with B660 and locked processors launching soon.
 
96MB is lot of cache and probly worth whatever price tag they stick on it

its a hail marry product for a dead platform AM4 is Dead amd said so months ago there will be no new chips for it

also VCache only really works on single CCD chips like the 5600x/5800x for obvious reasons it doesn't make sense to stick vcache on multi CCD designs
 
96MB is lot of cache and probly worth whatever price tag they stick on it

its a hail marry product for a dead platform AM4 is Dead amd said so months ago there will be no new chips for it

also VCache only really works on single CCD chips like the 5600x/5800x for obvious reasons it doesn't make sense to stick vcache on multi CCD designs

I don't see why wouldn't it work, at most it would behave exactly like the 5950X does - which is really just two 5800Xs in a single processor package. Zen 3 is not as picky as Zen 2 was for optimal core allocation because CCX was increased to 8 cores, so 1 CCD = 1 CCX, so instead of limiting inter-core latency sensitive applications to four cores, you can get that safely across eight and since it's just 2 CCXs to manage, the inter-CCD latency is also lower than it was on Zen 2.

It should be better regardless, remember when we first learned of the 3D cache technology, it was demoed on a 12-core/5900X-type prototype sample. It seems things changed (such as 6 nm class lithography on CCDs). That chip already showed some improvements back then, I suppose the only reason not to release a dual-CCD 3D part is costs... it would not decisively outclass Alder Lake and it would definitely be pricier, and saddled by AM4's aging features (lacks PCIe 5.0, DDR5, etc.). Not even the latest generation boards with tons of makeup like the C8DH and C8E really shake this off.
 
I'm underwhelmed by the low clock speeds. Let's see if PBO can help make up the lost ground.
 
I'm underwhelmed by the low clock speeds. Let's see if PBO can help make up the lost ground.

clock speeds don't matter nearly as much as the cache the increased cache alone is going to more then make up for any minor clock differences

I don't see why wouldn't it work, at most it would behave exactly like the 5950X does - which is really just two 5800Xs in a single processor package. Zen 3 is not as picky as Zen 2 was for optimal core allocation because CCX was increased to 8 cores, so 1 CCD = 1 CCX, so instead of limiting inter-core latency sensitive applications to four cores, you can get that safely across eight and since it's just 2 CCXs to manage, the inter-CCD latency is also lower than it was on Zen 2.

It should be better regardless, remember when we first learned of the 3D cache technology, it was demoed on a 12-core/5900X-type prototype sample. It seems things changed (such as 6 nm class lithography on CCDs). That chip already showed some improvements back then, I suppose the only reason not to release a dual-CCD 3D part is costs... it would not decisively outclass Alder Lake and it would definitely be pricier, and saddled by AM4's aging features (lacks PCIe 5.0, DDR5, etc.). Not even the latest generation boards with tons of makeup like the C8DH and C8E really shake this off.
remember that you need for the cores to be-able to communicate and share data across the caches I suspect having a extra 64MB of cache on each CCD might be to much for the IF to handle
 
96MB is lot of cache and probly worth whatever price tag they stick on it

its a hail marry product for a dead platform AM4 is Dead amd said so months ago there will be no new chips for it

also VCache only really works on single CCD chips like the 5600x/5800x for obvious reasons it doesn't make sense to stick vcache on multi CCD designs
Umm, you know like Milan-X o_O
 
More disappointed in Ryzen's availability and following from that pricing lol.

Current situation in my country is so bad that you can't justify going Team Red w/o being rabid fanboy.
 
clock speeds don't matter nearly as much as the cache the increased cache alone is going to more then make up for any minor clock differences


remember that you need for the cores to be-able to communicate and share data across the caches I suspect having a extra 64MB of cache on each CCD might be to much for the IF to handle

I'm not sure, man. It may be an uneducated guess of mine but I suppose it would be that big of a deal. That is more or less what Milan-X is, is it not? My understanding is that each CCX's cache slices should be independent from each other, kind of like each CCX on Zen 2 had four cores and 16 MB of L3, and two CCXs per CCD, on Zen 3 this is doubled to 32 MB per CCX and the core count is also doubled, basically what Zen 3D is taking this and adding the layer that contains the extra 64 MB seamlessly for a 96 MB L3 (the original existing 32+64 3D slice), but I may be wrong.
 
Back
Top