AMD 16 core chiplets fly under the radar

[humbug]

Bergamo 128 core Zen 4 launched, 16 core Chiplets.

Genoa with 96 cores has 12 CPU chiplets (8 cores in each) Bergamo with 128 cores has 8 chiplets, 16 cores in each.

Apparently no one knew about this, the reason being AMD didn't have to do any engineering work with motherboard vendors, so no one out side of AMD saw one, AMD have come up with a way to drop a new CPU type in to an existing platform and package that has already had all the qualification engineering done.

I'll let the tech supports tech support Windell explain....

PS: GenoaX has 1.1GB of L3 cache

128 Cores & 3D V-Cache EPYC - Launching Today!

Wendell talks Genoa and Bergamo from the new 3d V-Cache CPUs from AMDAMD Tuning Guides: https://www.amd.com/en/processors/server-tech-docs/search?f%5B0%5D=do...

youtu.be

 

[humbug]

I think a lot of that is helped by it retaining the standard IO die, and the fact that the actual cores are no different other than a Cache reduction.

Some more info here:

AMD EPYC Bergamo is a Fantastically Fresh Take on Cloud Native Compute

The AMD EPYC 9754 128 core and 256 thread part is a fantastically fresh take on cloud-native computing that beat our high expectations

www.servethehome.com

I'll read it thanks

 

[humbug]

Over 128+ cores that's not so hot - only the standard 8 MB per core. The thing is that the more cores you have the greater the chance of a cache miss - and the greater the chance of more than one - and then all the cores are slowed by accessing main memory over the same CPU-memory link. So the cache needs to be even larger.

That's Genoa, 96 cores, its 11.5MB per core.

There is no 3D cache Bergamo and never will be, they removed the through VIA's as part of compacting the architecture to make the 16 core CCD's only a little larger than the 8 core CCD's

PS: Standard in 4MB per core.

 

[humbug]

I could see a 16c chiplet+ 8c chiplet on AM5 with super fast DDR5.

Watch this...

 

[humbug]

7950X scores about 38,000 MT R23. That's at 5.0 to 5.1Ghz.

The 16 core Zen 4C cores uses about 2 watts, 32 watts per 16 core CCD at 3.1Ghz, clocked up to 4.0Ghz they might use 3 watts per core, 50 watts for the 16 core CCD, per core they would be 78% the performance of the 5.1Ghz main cores.
38,000 / 2 = 19,000, 19,000 X 0.78 = 14,820 X 2 = 29,640 + 19,000 = 48,640 score in R23 and at a reduced power from 170 watts to 135 watts.

Remember unlike E-cores AMD C-Cores are the same cores as the P-Cores re-archtectured to be more compact and more power efficient, they have the same IPC. They are more power efficient than ARM cores.

AMD just keep doing it right.

 

[humbug]

So AMD moving to 16c chiplets on AM5 next?

No, big IPC gain tho, 20% minimum, possibly more.

Performance should be up 30% per core, that counts for MT too.

Score about 49,000 in R23.

I prefer than than having more slower cores.

 

[humbug]

Intel went the E core route because they couldn't cram more P cores into their CPUs, not because of some awesome design advancement. AMD doesn't have that limitation, plus Bergamo is designed for specific use cases, so the C cores have had some things stripped out and the cache reduced.

The C-class cores offer no benefit to desktop or workstations, so there's no need to mix and match. AMD are perfectly capable of downclocking cores as required, they have no need to replicate Intel's crutch.

The argument is but but but laptops and power consumption, yes, sort of, to a limited extent.

Intel are actually being quite clever in that they want a performance lead over AMD in terns of per-core performance (games / some 2006 code Adobe apps) and MT performance.
The Ring Bus only support 8 to 10 cores, that's ok you only need 8, so you make these 8 really fat cores for games and Adobe apps and for everything else you can tag on an infinite number of these little cores.
The little cores have much less transistors and limited performance so if windows is doing something in the background it can use those cores at 6 watts instead of 12 for the big ones and it makes no difference.
Cool, ok, its not the same thing as those ARM chips that you see in mobiles, those little cores are RISC, they are striped out instruction sets, they can't do everything the big cores can and the power consumption of those is measured in tiny fractions of a watt, its really not the same thing at all, Intel little cores are full CISC cores, just crappier slower versions of the P-Cores.

Now here is the thing about all of this, the P-Cores are 4X the size of the E-Cores, impressive right? Well those P-Cores are also 3X the size of AMD's P-Cores, AMD big cores are far closer in size to Intel's little cores, and for those massive 3X the size P-cores Intel gain......................... 10% IPC over AMD's cores, what's more AMD's P-Cores use no more power during these low stress you don't even know its happening tasks, so the Intel battery life in pure idle is no better than AMD's while working AMD's battery life is better.

What Intel are doing is clever, exactly the sort of innovation you would expect from Pat Gelsinger, but completely ineffective. I have no doubt Pat thought it would be his second Sandy Bridge moment, no. Not this time buddy.

 

[humbug]

You cant just go flying chips in under the radar... its rude. Looks like it goes brum brum! I was was in the market for an Epyc chip right now you could be damn sure it would need to go brum brum... I still find what they are doing here and how they approached it pretty spot on.

The rear mounted brum approach.

 

[humbug]

Resurrecting this thread.
I was having a look at the potential for having the equivalent of 16 P cores in the nearish future. Intel don't seem to have anything planned.
Some AMD CPUs are "in theory" 16 core, but in use are x2 CCDs of 8 cores, with the CCDs seemingly not talking to each other very effectively, meaning that gaming is apparently worse on a 7950 than a 7800.
I did look at info on the AMD Zen4c, which does look like having 16 cores on a single CCD. However, what I then read was that AMD are doing this by cramming x2 CCX of 8 cores onto a single CCD, with lower cache per core.

Has anyone read of a genuine 16 core (on a single board) in the pipeline?
For that matter, does anyone really understand why many of the synthetic benchmarks can make full use of the 16 cores in current AMD 7950s, whilst most games seem to fail miserably? If so, is there a potential fix?

The problem with 16 P-cores is it would have to be at least 400mm^2, the 13900K is already 260mm^2, that's already huge for a CPU, for context the RTX 4080 is about 380mm^2, the price of something like that would have to be astronomical, its why they have these E-Cores, they are 1/4 the size of the P-Cores, they can cram lots in to keep up with AMD's 16 P-Core CPU's in MT.

AMD's 8 core Zen 5 CCD's are 75mm^2 each, but they are on TSMC 5nm and frankly just far more efficient, even on Intel 10nm they would still be smaller. AMD also use an MCM chiplet design because it further improves the cost effectiveness of the CPU and they can make higher core count CPU's for Data centre than you can monolithic.
Intel have gone down the same MCM rout for its Datacentre CPU's, the Xen 8480+ has 56 core, that's with 4X 14 core chiplets, each one is 480mm^2 (go back to what i said at the beginning) they are very large cores. AMD are at 96 cores, 12X 8 core 75mm^2 chilets

There is no "16 P-Core" chiplet that i know of on AMD's roadmap.

On gaming, what AMD have is an algorithm that keeps the game entirely in one 8 cores CCD, but will allow less latency sensitive workloads on to the seconds CCD if needed, there is a slight penalty in performance when that happens, but it is only slight, i bought a Ryzen 5800X because the gaming performance was more consistent than the 5950X and certainly more consistent than the 5900X, but the difference overall is only a few %.

Intel do the same thing with the E-Core CPU's
Star Citizen is a very good example of a game that uses a lot of threads, all 16 of my threads to be loaded at 80%+ is not unusual, that's with a 2070 Super at 1440P, Intel E-Core CPU's are a problem with this game when running a fast GPU like a 3090Ti or 4080 and up.. the CPU works so hard it is quite often spilling over to the E-Core and because there is a huge performance difference between the E and P-Cores frames are being rendered at very difference performance levels, or speed, with that they stutter in this game, quite badly, Intel have been to CIG to try to sort it out, and its better than it was but its still happening...

Personally i wouldn't buy one, i tend to keep my CPU's for years and games only get more demanding, not less.