AMD at CES 2023: 7950X3D, 7900X3D, 7800X3D - VCache Party In Here

Lakados said:
It’s more like because some 12 year old in a Chinese sweat shop had a faulty pressure gauge so they couldn’t correctly fill a vapour chamber.
But yes.
Click to expand...
Depends on which side,
Pull a vacuum to the right a mount (this would be pressure reading)
Most likely they use by weight and not pressure for the water, fill a tube with liquid to the right mass, open valve let the coolant fill the chamber. (weight sensor reading issue)

If there's not enough coolant, I would assume the second one, if not evaporate condensing properly, assume the first. Although honestly both could cause the same issue .

Wonder if I still have some manufacturing slides around for that stuff.
 
Balkroth said:
Depends on which side,
Pull a vacuum to the right a mount (this would be pressure reading)
Most likely they use by weight and not pressure for the water, fill a tube with liquid to the right mass, open valve let the coolant fill the chamber. (weight sensor reading issue)

If there's not enough coolant, I would assume the second one, if not evaporate condensing properly, assume the first. Although honestly both could cause the same issue .

Wonder if I still have some manufacturing slides around for that stuff.
Click to expand...
The point is all this trouble because some place paying slave wages in a developing country doesn't have redundant QC, or possibly any QC.
I have a hard time imagining that many faulty parts being produced by a reputable firm and not getting caught before they make it out the door.
 
FrgMstr said:
Really? I mean really really? How many bad memory chips did we see on RTX 2080 cards? Plenty of others worth citing as well in many industries.
Click to expand...
I'm still not convinced that wasn't industrial sabotage but yeah, you do make a point I was perhaps a little harsh in that.
I find it strange that a bunch of barrels of tainted processing liquids made their way into Taiwanese facilities at the same time China was doubling down saying that the facilities they had within the Chinese mainland were better and more reliable... Coincidence I am sure.
 
Zarathustra[H] said:
There is no better WiFi than no WiFi at all.

Be it copper or fiber, for me it is wired or bust.

I'll save WiFi for devices that need to be mobile. (phones, tablets, laptops, etc.)
Click to expand...
Additional later reply to this but: Another great thing about Intel wifi is it include very good bluetooth. Its nice to connect my Dual Sense controller to my PC, with bluetooth.

NamelessPFG said:
Much as I'd like to believe that, theory has a way of being unseated by reality.

I'll keep Process Lasso at the ready, just in case the default scheduling behavior doesn't work out.

What I really want to know is how this will affect overclocking, since now we'll have to overclock per CCD due to the V-Cached one hitting limits sooner. I'm not sure if the UEFI on AM5 boards is set up for that.

If the 7900X is 6+6, then that definitely pushes me toward either the 7800X3D (8+0, only cached CCD active) or 7950X3D (8+8) were I to do any Zen 4 builds. Eight homogeneous, performance-oriented cores is my minimum nowadays.
Click to expand...
I was overclocking per CCD with a 3900x, on a B550 itx board. One CCD was clearly better than the other.
Overall, I wouldn't worry much about Zen4x3D. The scheduling for that should be even easier than Intel's P+E cores setup-----and that doesn't really have a lot of problems.

schoolslave said:
This heterogeneous cores on a single chip trend for desktop/workstation CPUs is retarded. The scheduling complexity alone with Intel's P+E config is bad enough (talking about the amount of bullshit workload movement and heuristics crap the kernel has to do to properly "balance" across the P+E cores) but now we're talking different logical cores have different perf characteristics? WTF. Have these companies not learned anything from how hard it is to properly optimize for NUMA/multi-socket/multi-node systems? I suspect they just don't give a shit and once again (except in regards to hardware architecture/design now too): https://tonsky.me/blog/disenchantment/.

I wonder how much performance is lost due to all the overhead of managing this garbage on a general purpose OS and the general software free-for-all (so not something largely vertically integrated like iOS/Android phones).

I also wonder how many more silly-named vulnerabilities lurk in this mountain of complexity.
Click to expand...
Again, I wouldn't worry much about overhead. These CPUs make adjustments at insane speeds. Even if there were some overhead------I've been very impressed with Alder Lake. The performance is incredible and the heat output is very manageable. Raptor Lake is a little different. Feels more like an overclocked chip. Its definitely faster. But....it also runs noticeably hotter.
Still though, the performance of both Intel and AMD's new stuff is insane. I'm not sure what all the boohoo is about. I mentioned in another thread----Raptor Lake essentially doubles the single core Cinibench score of 10th gen Comet Lake. 13600k's 6 Pcores by themselves, scores about the same as a 10900k in multicore.
 
Lakados said:
Yeah but the point is AMD made the same claims with their 5000 and 6000 series mobile lineups and in the 6000 series mobile launch claimed over 20h of battery life and none of it happened.
Yeah, it's a fair comparison until Apple gets the new silicon out which they claim was pushed back because of TSMC delays but it's an old comparison they have used for a while and never actually delivered on.
So AMD can put up all the slide shows they want on this but I want benchmarks that back it up and products on the shelves that actually deliver on their theoretical numbers presented there.
Click to expand...
AMD's Rambrandt was really impressive. Not as good as Apple's battery performance but not too far off either. Not sure what claims you're referring to but it seems to perform very well.

 
The only time human don't make mistakes is when they never do anything.
 
DukenukemX said:
AMD's Rambrandt was really impressive. Not as good as Apple's battery performance but not too far off either. Not sure what claims you're referring to but it seems to perform very well.

Click to expand...

It does test well when plugged in, but when unplugged and on battery the AMD and Intel offerings fall off significantly. I’m looking forward to see if AMD delivers this time, hopefully something with Dell would be good as it makes bulk purchase easier.
Apple makes it too easy for me to bulk order but repair options for me are non existent so everything gets mailed out which pisses everyone off because it’s at best a 3 week turn around. And that of course only starts after the user is fed up with the machine so they are … perturbed at that stage. Dell just sends me parts and a YouTube live no to the repair instructions.
 
To bring us back on topic. The best chip this time will be the 7950x3d right? Full 8 cores for each of the compromise ccds. The 7900s 6+6 feels a bit of a step back on paper.
 
Since my PC is only for gaming, and I sometimes struggle to keep 240 FPS (CPU limit) in some modern FPS games with my 240 Hz monitor, I'm considering swapping out my 7950X with a 7800X3D. Anyone think I'm crazy?
 
Vega said:
Since my PC is only for gaming, and I sometimes struggle to keep 240 FPS (CPU limit) in some modern FPS games with my 240 Hz monitor, I'm considering swapping out my 7950X with a 7800X3D. Anyone think I'm crazy?
Click to expand...
No, it will be faster. The 5800x3d keeping up with the first gen AM5 stuff in most games is evidence of the benefit of the cache.
 
Vega said:
Since my PC is only for gaming, and I sometimes struggle to keep 240 FPS (CPU limit) in some modern FPS games with my 240 Hz monitor, I'm considering swapping out my 7950X with a 7800X3D. Anyone think I'm crazy?
Click to expand...

Only downside I can think of is slightly less future proofing, and future proofing is rarely a good way to spend your money, as that same performance will be cheaper in the future when it is needed.

Also, insert comment about 240hz being 100% pointless other than to cure FOMO. Yes, even for pro conpetitive gaming.
 
Last edited:
Meeho said:
There was also that one time the biggest CPU company in the world mass produced a new CPU flagship that computed wrong.
Click to expand...
You mean the P5 Pentium FDIV bug that forced many industries to recompute their results on non-bugged CPUs?

Didn't seem to set bsck the Pentium much in the grand scheme of things, though. Just think of the Quake framerates!

Vega said:
Since my PC is only for gaming, and I sometimes struggle to keep 240 FPS (CPU limit) in some modern FPS games with my 240 Hz monitor, I'm considering swapping out my 7950X with a 7800X3D. Anyone think I'm crazy?
Click to expand...
Not me. You bought a new CPU expecting a performance target to be met, it doesn't quite meet the mark, but there's new stuff that might which still fits right into your existing AM5 platform.

So you buy that 7800X3D, verify that it works, and sell off your 7950X to recoup the cost if your games of choice perform better with the cache. If not, take the 7800X3D right back for a full refund and keep the 7950X.

It may not be the most optimal use of funds, but I don't think anyone building AM5 with a 7950X right now is concerned with cost-effectiveness, only pure performance.
 
Lakados said:
It does test well when plugged in, but when unplugged and on battery the AMD and Intel offerings fall off significantly. I’m looking forward to see if AMD delivers this time, hopefully something with Dell would be good as it makes bulk purchase easier.
Click to expand...
Unlikely AMD will have the same performance plugged and unplugged. The 30% performance increase is likely plugged in.
Apple makes it too easy for me to bulk order but repair options for me are non existent so everything gets mailed out which pisses everyone off because it’s at best a 3 week turn around. And that of course only starts after the user is fed up with the machine so they are … perturbed at that stage. Dell just sends me parts and a YouTube live no to the repair instructions.
Click to expand...
Remember those SSD's I predicted would brick Apple products and make them unrepairable? Even if Apple sends you parts there's no part Apple has to solve this problem.
 
DukenukemX said:
Unlikely AMD will have the same performance plugged and unplugged. The 30% performance increase is likely plugged in.

Remember those SSD's I predicted would brick Apple products and make them unrepairable? Even if Apple sends you parts there's no part Apple has to solve this problem.
Click to expand...

The issue with the performance is the M1 and the M2 doesn’t really change between when it is plugged in and unplugged.

But I’ve also got well over 150 m1 devices in the field so far haven’t seen that failure yet <knock on wood> lots of double double failures though. No recovery options for data so thank you Microsoft for O365.
 
Randall Stephens said:
To bring us back on topic. The best chip this time will be the 7950x3d right? Full 8 cores for each of the compromise ccds. The 7900s 6+6 feels a bit of a step back on paper.
Click to expand...
I think it is a full 8+4. It is a must that it be on parity with the 7800 and 79503d. Also the 79003d has a very high total cache so IMO it must have the full 8x3d cores or it wouldn't get that total cache size.
 
I just did the math.
7800= 96+8x cores(1MB per core) =104 total cache
79003d=chiplet a (96+8x cores(1MB per core)+ chiplet b (32+4x cores(1MB per core) =140 total cache
79503d=chiplet a (96+8x cores(1MB per core)+ chiplet b (32+8x cores(1MB per core) =144 total cache

*edit* okay by the numbers Randall might still be right.
79003d= chiplet a (96+6x cores(1MB per core)+ chiplet b (32+6x cores(1MB per core) =140 total cache
same total.
My reason why I think they went 8+4 is for parity between 7800/79003d/79503d. 79003d can't shit the bed so I think they would mass produce 8 core 3d chiplets to be used for all. Then just castrate the cheaper non3d chiplet. Of course there could be a binning where a 6core might work for using faulty chips but I doubt that. We know AMD went chiplets because of the massive decrease is faulty chips and surely we would have seen a first gen 56003d if that was the case.
 
Last edited:
funkydmunky said:
I just did the math.
7800= 96+8x cores(1MB per core) =104 total cache
79003d=chiplet a (96+8x cores(1MB per core)+ chiplet b (32+4x cores(1MB per core) =140 total cache
79503d=chiplet a (96+8x cores(1MB per core)+ chiplet b (32+8x cores(1MB per core) =144 total cache

*edit* okay by the numbers Randall might still be right.
79003d= chiplet a (96+6x cores(1MB per core)+ chiplet b (32+6x cores(1MB per core) =140 total cache
same total.
My reason why I think they went 8+4 is for parity between 7800/79003d/79503d. 79003d can't shit the bed so I think they would mass produce 8 core 3d chiplets to be used for all. Then just castrate the cheaper non3d chiplet. Of course there could be a binning where a 6core might work for using faulty chips but I doubt that. We know AMD went chiplets because of the massive decrease is faulty chips and surely we would have seen a first gen 56003d if that was the case.
Click to expand...

Too tired to follow/TDLR the math here. Per AMD's Scott Stankard (Director of Product Management for AMD Client) who I met with last Friday stated that the 7950X 8X3D+8 CCD, 7900X is 6X3D + 6 CCD and 7800X is a single 8X3D CCD.
 
1_rick said:
The math also works out with 2x6 core CCDs: 96+6 + 32+6.
Click to expand...
My apologies for the slow edit. I realized as soon as I posted. Additionally to my above explanations for 8+4 is that I suspect that there might be issues with a 8core 3d chiplet
Schro said:
Too tired to follow/TDLR the math here. Per AMD's Scott Stankard (Director of Product Management for AMD Client) who I met with last Friday stated that the 7950X 8X3D+8 CCD, 7900X is 6X3D + 6 CCD and 7800X is a single 8X3D CCD.
Click to expand...
Interesting as it would put the 7900x3d at a disadvantage. Of course it does jump up (cache wise) in the non-3d cores.
I can't name drop like you, but feel something aint right here. I feel very suspect that AMD makes a "special" x3d core just for the 7900x.
 
funkydmunky said:
Interesting as it would put the 7900x3d at a disadvantage. Of course it does jump up (cache wise) in the non-3d cores.
I can't name drop like you, but feel something aint right here. I feel very suspect that AMD makes a "special" x3d core just for the 7900x.
Click to expand...
AMD is too cheap to develop a unique piece of silicon for a mid tier part. If they did that where would they sell their defective silicon that they have to disable a core or a memory block too. That would be wasteful.
 
funkydmunky said:
My apologies for the slow edit. I realized as soon as I posted. Additionally to my above explanations for 8+4 is that I suspect that there might be issues with a 8core 3d chiplet

Interesting as it would put the 7900x3d at a disadvantage. Of course it does jump up (cache wise) in the non-3d cores.
I can't name drop like you, but feel something aint right here. I feel very suspect that AMD makes a "special" x3d core just for the 7900x.
Click to expand...
Disadvantage is relative - X3D cores are good at gaming but not necessarily at much else, and most games don't get a significant performance bump going from 6 to 8 cores (even then, if more than 6 are needed for gaming there's still the other 6 standard cores).

I figure the 7900X3D is more a way to unload 8 core X3D chiplets that are partially defective by turning off the affected cores.

Name dropping is only to give a source for the information that's not me making up something :) (especially if reality ends up differing from what I've been told).
 
Zarathustra[H] said:
I would love to go back to that time.

Give me nothing on board, and instead give me 8 expansion slots.

Let me customize my experience with the expansion cards I want instead of picking and choosing from a series of on board motherboard bundles, none of which ever seem to align with my preferences.

This whole concept of putting everyhting on board has absolutely killed the joy in the hobby. There is no customizing my build to my preferences anymore. I just have to deal with whatever Asus, Gigabyte or MSI have chosen for me. IN many cases this results in me searching desperately for boards with more expansion so I can install at least ONE component of my choosing, and then disable a ton of stuff on board that I will never use. It is a colossal waste.

I love the good old 286 days when boards looked like this:

View attachment 539644

There was no sound on board. No networking. The most you'd ever get on board was a serial port, and even that wasn't a guarantee.
If you wanted anything, like a floppy controller, or a hard drive controller, a video card or a sound card, you added them all yourself, and you got to pick exactly which ones you wanted, and if you didn't want a certain component you just omitted it.

It resulted in diverse and interesting builds customized to the desires of their builders.

I want this back.
Click to expand...
No, what you want is every 6 months for a new CPU and GPU to come out that is twice as fast as what you just bought. I think we could all get behind this.

Just buy a good mobo, typically the higher end ASUS. The OnBoard audio rivals Creative cards, even has high-end expensive socketed DAC's. Onboard nic is good enough as CPU's are fast enough to easily handle even the software based nics (PHY types), and these more expensive boards have good nics (10G even) on them with low cpu utilization. OnBoard sata is good enough for a bluray burner. M2 can't be beat as a OS boot device. OnBoard USB 3 is fine and gets power delivery thru the mobo, addin usb3 requires supplemental power.
A mobo like this goes for $500 to $900. You get what you pay for.

I still use an add-in card for a raid 5 pcie3 controller. Not as fast as M2 but redundant. I've tried using the Creative X-Fi card, but it doesn't sound any better. Maybe if I used the pc as an HTPC, a soundcard might come in useful for Dolby Atmos support. But for gaming, no need.

Boards with the built in components are more efficient use of space and better airflow. Those pc's with 10 add in cards also used CPU's that didn't need a fan, or some didn't even need a heatsink! Those days are gone, CPU's are much too powerful for that.
 
GoodBoy said:
Boards with the built in components are more efficient use of space and better airflow. Those pc's with 10 add in cards also used CPU's that didn't need a fan, or some didn't even need a heatsink! Those days are gone, CPU's are much too powerful for that.
Click to expand...

I use many expansion cards in my systems, and I have no problems with airflow. You just have to put some thought into your build.
 
chameleoneel said:
I was overclocking per CCD with a 3900x, on a B550 itx board. One CCD was clearly better than the other.
Overall, I wouldn't worry much about Zen4x3D. The scheduling for that should be even easier than Intel's P+E cores setup-----and that doesn't really have a lot of problems.

Again, I wouldn't worry much about overhead. These CPUs make adjustments at insane speeds. Even if there were some overhead------I've been very impressed with Alder Lake. The performance is incredible and the heat output is very manageable. Raptor Lake is a little different. Feels more like an overclocked chip. Its definitely faster. But....it also runs noticeably hotter.
Still though, the performance of both Intel and AMD's new stuff is insane. I'm not sure what all the boohoo is about. I mentioned in another thread----Raptor Lake essentially doubles the single core Cinibench score of 10th gen Comet Lake. 13600k's 6 Pcores by themselves, scores about the same as a 10900k in multicore.
Click to expand...

I saw something similar on some ARM chips a while back... I believe they called it "Little Big Architecture". https://www.arm.com/technologies/big-little

This makes sense on a mobile platform... I'm not sure about it on a desktop. The market will decide.
 
Schro said:
Disadvantage is relative - X3D cores are good at gaming but not necessarily at much else, and most games don't get a significant performance bump going from 6 to 8 cores (even then, if more than 6 are needed for gaming there's still the other 6 standard cores).

I figure the 7900X3D is more a way to unload 8 core X3D chiplets that are partially defective by turning off the affected cores.

Name dropping is only to give a source for the information that's not me making up something :) (especially if reality ends up differing from what I've been told).
Click to expand...

Seems like a waste of the v-cache to put it on a 6 core CCD though. The whole thing with only putting v-cache on half of the upper tier chips doesn't interest me at all though. I grabbed a 7950x today while Microcenter still was giving away 32GB of DDR5 with them for the time being. I'll upgrade down the line when something better comes around.
 
You must log in or register to reply here.
Back
Top