Rumor: Ryzen 7000 3d v-cache cpus could be limited to 6 core 8 core configurations

[P4-630]

AMD's issue isn't really performance, its pricing. Their 6 core processor is essentially competition for Intel's 13400 processor, their 8 core is the actual competitor to Intel's 13600, etc... Their core counts are too low to compete with Intel on value and performance in multithreaded apps.

The 7600x need to realistically be $200 to be worthwhile, the 7700x needs to be $300 and motherboards from the B series need to start as low as $120 in order to have any sucess. At this point Intel is just the better purchase, I don't see why anyone would go for the Ryzen 7000 series over Intel's 13000 series.

Also Intel still has DDR4 support, which may be a reason to go Intel afterall for some with the more affordable motherboards, AM5 requires DDR5, 6000MHz the sweetspot I believe they say...

 

[gffermari]

At this point Intel is just the better purchase, I don't see why anyone would go for the Ryzen 7000 series over Intel's 13000 series.

Yes, Intel's pricing is better. But there's no potential 3D or anything coming to Z690/Z790.
Purchasing an AM5, with whatever cpu now, you have Zen 4 3D, Zen 5, Zen 5 3D and probably Zen 6 on the way.

 

[Dr. Dro]

Yes, Intel's pricing is better. But there's no potential 3D or anything coming to Z690/Z790.
Purchasing an AM5, with whatever cpu now, you have Zen 4 3D, Zen 5, Zen 5 3D and probably Zen 6 on the way.

That's not really an argument in favor of a socket AM5 system in most cases. Most people won't upgrade every generation, and AMD has a history of going back on the their word on this specific forwards compatibility promise. I got burned as a X370 buyer, so fool me once shame on you, fool me twice shame on me. At least they reverted their stance and those who have more patience than me benefitted from it.

 

[gffermari]

That's not really an argument in favor of a socket AM5 system in most cases. Most people won't upgrade every generation, and AMD has a history of going back on the their word on this specific forwards compatibility promise. I got burned as a X370 buyer, so fool me once shame on you, fool me twice shame on me. At least they reverted their stance and those who have more patience than me benefitted from it.

I agree.
It's pointless to upgrade from 12th gen to 13th gen as well as from a 5800X to X3D.

AMD with the 3Ds, just created an additional upgrade step.
I moved from 3700X to the 3D. A Zen 4 owner can jump to Zen 5 3D or Zen 6.

Intel's die layout is better (P/E cores) and their tier performance is better(13600/7600, 13700/7700). AMD doesn't afford to do the same mistake again with the cpu support.

 

[ratirt]

Also Intel still has DDR4 support, which may be a reason to go Intel afterall for some with the more affordable motherboards, AM5 requires DDR5, 6000MHz the sweetspot I believe they say...

Yes Intel has DDR4 but it is also noticeable slower and that gap will be grow. If you cheap out on a Mobo and DDR5 then you will find yourself in a position of losing performance. If you buy a new platform you should move for DDR5. Maybe not necessarily with 13th gen if you are planning on upgrading when Meteor lake shows up. In any case, DDR5 is going to be more popular and Meteor lake is definitely going to use it. So you'd not need to spend money for RAM at least.

 

[Count von Schwalbe]

But maybe they can actually glue their "standard" cache chiplet to the APU.

Looking at Zen 3 (Cezanne vs Vermeer) suggests that this is not an option. The L3$ section of the die uses a 4x4 grid of L3$ complexes tied to each core, and the complexes are 512KB for Cezanne and 1MB for Vermeer, so the Vermeer L3$ die area is physically larger. Also, the Vermeer die has had TSVs for the 3D stacking designed into it from the beginning. If the APUs already have these, a "custom" L3 chiplet designed for them is not off of the table.

 

[DemonicRyzen666]

X3D is thermally limited (at least on Zen 3 design) because the cache slice is physically located on top of the CCD, insulating the IA cores themselves further from heat transfer. This is one of the reasons they are rated for less current, they are within 1-2C of the original 5800X because they also operate at a much lower voltage.

Unless you're using recent AGESA anyway which broke current control due to the introduction of the EDC amp limit for the 5800X3D that is applying to every other processor anyway - what a bother.



There is no such difficulty. Packaging technology is the same, and dual CCD chips already have double the cache (eg. 32 MB on 5800X, 64 MB on 5950X). But they can sell these perfect CCDs to the Epyc segment instead for a lot more money, or make basically twice as many 5800X3D's... it's not worth for AMD at the price they'd sell.

3D cache is not localed on any slicon layer it's located on top of the L3 cache layer for veritcal connection, before the final thin layer of silicon is added. The cores only have a structral silicon to match the height of the L3 cache, but AMD doesn't list how thick or thin this layer is of the structural silicon. It's doubtful to have much heat resistances anyways. Many old Phenom II x3/Athlon Qaud/Dual chips had dead/or disabled cores that still spread heat the a little better monolithic dual core chips. Since the added silicon worked out to spread the heat. more mass & more surface area takes away heat. The reviews here on techpowerup show 1-2C difference from the 5800x vs 5800x 3D. I can get 1-2C difference with multiple chips even with the 200mhz lower or high boost clocks. Heck, me the one guy who qouted me have cores spreads variance on CCD's far higher then difference of the x5800 X3D vs 5800 XD & we both have different clock speeds & chips. He's got a 15C difference, I have 12C difference on cores in a single CCD/ or even Dual CCD's. I got lower temps with older AGESA with higher voltages for single thread work loads. My single thread on the older AGESA was up to 1.55 volts in single thread, but multi-core Voltage hasn't really changed it still sits around 1.412-1.388. My tempertaures have gone up from On new AGESA from 75C to 82C on Prime Small FTT's & y-Cruncher. Not all cores are near that temperature.

 

[Wirko]

Looking at Zen 3 (Cezanne vs Vermeer) suggests that this is not an option. The L3$ section of the die uses a 4x4 grid of L3$ complexes tied to each core, and the complexes are 512KB for Cezanne and 1MB for Vermeer, so the Vermeer L3$ die area is physically larger.

Yeah, and I said basically the same in a later post: the cache layout on both CCD and APU dies would have to be identical, which is hardly possible.

Also, the Vermeer die has had TSVs for the 3D stacking designed into it from the beginning. If the APUs already have these, a "custom" L3 chiplet designed for them is not off of the table.

I call it highly unlikely. This variant of cache die would be a low volume part, not reusable in Epycs or anything else. Its place would be in thin notebooks and ultra small desktops where a discrete GPU chip wouldn't fit for reasons of size, heat, and cost - that's great but is that a big enough market?

 

[Dr. Dro]

3D cache is not localed on any slicon layer it's located on top of the L3 cache layer for veritcal connection, before the final thin layer of silicon is added. The cores only have a structral silicon to match the height of the L3 cache, but AMD doesn't list how thick or thin this layer is of the structural silicon. It's doubtful to have much heat resistances anyways. Many old Phenom II x3/Athlon Qaud/Dual chips had dead/or disabled cores that still spread heat the a little better monolithic dual core chips. Since the added silicon worked out to spread the heat. more mass & more surface area takes away heat. The reviews here on techpowerup show 1-2C difference from the 5800x vs 5800x 3D. I can get 1-2C difference with multiple chips even with the 200mhz lower or high boost clocks. Heck, me the one guy who qouted me have cores spreads variance on CCD's far higher then difference of the x5800 X3D vs 5800 XD & we both have different clock speeds & chips. He's got a 15C difference, I have 12C difference on cores in a single CCD/ or even Dual CCD's. I got lower temps with older AGESA with higher voltages for single thread work loads. My single thread on the older AGESA was up to 1.55 volts in single thread, but multi-core Voltage hasn't really changed it still sits around 1.412-1.388. My tempertaures have gone up from On new AGESA from 75C to 82C on Prime Small FTT's & y-Cruncher. Not all cores are near that temperature.

That's... what I said though? Also, it's false equivalence to compare the 3D cache slice to dark silicon (disabled bits) in other processors, because the 3D cache slice is active, and it does generate heat of its own.

But any resistance is resistance and can change drastically how the processor behaves. In addition to being more prone to heat buildup, the cache slice itself is known for being fragile, hence why AMD implemented such a strict amp limit for it. This limiter kicks in once rated EDC is maxed out.

AGESA newer than v2 1.2.0.3 patch C (the last that predates 5800X3D support) has a hard VID_LIMIT cutoff at that precise voltage range your processor is operating at, and this is applying to the non-3D processors as well. This is the so-called EDC bug I've been rambling about lately.... the 5800X3D simply existing basically broke the rest of the Ryzens' ability to maintain their operating frequency under heavy load. What's baffling is that almost a year has gone by and somehow no one in the media is even remotely interested in covering this problem - be it GN, Tom's, LTT, TPU, etc.

 

[wheresmycar]

That's not really an argument in favor of a socket AM5 system in most cases. Most people won't upgrade every generation, and AMD has a history of going back on the their word on this specific forwards compatibility promise. I got burned as a X370 buyer, so fool me once shame on you, fool me twice shame on me. At least they reverted their stance and those who have more patience than me benefitted from it.

I dunno if i agree with forward support not being one of the factors which makes the AM5 proposition an attractive one. Personally i fancy the idea of being tied into a platform which facilitates a cost-effective upgrade path. I didn't/don't have the same priviledges with my 7700K setup or my current 9700K build and the upgrade itch has got my balls in a twist. This doesn't mean upgrading every generation but throughout the AM5 support plan at some point upgrading when the performance uplift is appealing enough . I totally get it, you've had a bad deal with the X370 initial rejection - yep that sucks and would make any buyer distraught enough not to trust or buy into future promises. But on the outset AM4 did pull it off and went a long mile beyond with 5000-series and X3D (IMO, a massive gamer's bonus). So a little AM5 confidence for me. At the moment I have zero interest in buying into AM5 at the current inflated asking price + not entirely keen on higher power consumption/thermals. Should AMD come to their senses with a more affordable package + a couple of well-performing non-X SKUs or a real balance shifter 7000-X3D... i'd be interested!

So in short, yep people buying into platforms today with zero attention to 3 year+ upgrade possibilities shouldn't be bothered with the support plan. Others who fancy oiling their rigs for a meaningful performance uplift without forking out for a platform swap would certainly find the same platform fwd-Gen-support proposition attractive. If AMD pulls a dirty on its promise... and stays dirty, I'm gonna join U!!

 

[freeagent]

Honestly.. if you can’t tune it, I’m not interested. 3D performs well in game, but it’s slow for everything else.

 

[Dr. Dro]

Honestly.. if you can’t tune it, I’m not interested. 3D performs well in game, but it’s slow for everything else.

Well you'll be heartbroken to find that PBO is effectively broken and now only reduces score with new AGESA due to aforementioned EDC bug. Expect up to 15% lower MT scores on 5950X processors and up to 10% on 5900X. 5600X should have it rough too as it can no longer exceed 90A EDC before the problem manifests itself.

I went from 29.3K to 26.8K with the 5950X on R23, same settings, by updating AGESA from 1.2.0.3 C to 1.2.0.7. It's a good thing my motherboard has USB flashback functionality that makes BIOS downgrading possible.

 

[freeagent]

Well you'll be heartbroken to find that PBO is effectively broken and now only reduces score with new AGESA due to aforementioned EDC bug. Expect up to 15% lower MT scores on 5950X processors and up to 10% on 5900X. 5600X should have it rough too as it can no longer exceed 90A EDC before the problem manifests itself.

I went from 29.3K to 26.8K with the 5950X on R23, same settings, by updating AGESA from 1.2.0.3 C to 1.2.0.7. It's a good thing my motherboard has USB flashback functionality that makes BIOS downgrading possible.

1206 and 1207 are for X3D, I like 1203 for my 5600X and 5900X. I get my 240PPT 160TDC and 190EDC. 1207 on 5900X is pitiful.

 

[Dr. Dro]

1206 and 1207 are for X3D, I like 1203 for my 5600X and 5900X. I get my 240PPT 160TDC and 190EDC. 1207 on 5900X is pitiful.

Unfortunately 1203 is pretty buggy in regards to memory training and it has the fTPM stutter issue which I'm affected by. With this AGESA my motherboard never trains my memory correctly from a cold start (like Biff's car in Back to the Future I need to start it twice before it will POST), and since the B550-E doesn't have a dedicated TPM header I can't workaround these issues.

1207 fixed both at the blood price of performance and it sucks to the point I just want to get rid of this platform sometimes.

 

[tfdsaf]

I dunno if i agree with forward support not being one of the factors which makes the AM5 proposition an attractive one. Personally i fancy the idea of being tied into a platform which facilitates a cost-effective upgrade path. I didn't/don't have the same priviledges with my 7700K setup or my current 9700K build and the upgrade itch has got my balls in a twist. This doesn't mean upgrading every generation but throughout the AM5 support plan at some point upgrading when the performance uplift is appealing enough . I totally get it, you've had a bad deal with the X370 initial rejection - yep that sucks and would make any buyer distraught enough not to trust or buy into future promises. But on the outset AM4 did pull it off and went a long mile beyond with 5000-series and X3D (IMO, a massive gamer's bonus). So a little AM5 confidence for me. At the moment I have zero interest in buying into AM5 at the current inflated asking price + not entirely keen on higher power consumption/thermals. Should AMD come to their senses with a more affordable package + a couple of well-performing non-X SKUs or a real balance shifter 7000-X3D... i'd be interested!

So in short, yep people buying into platforms today with zero attention to 3 year+ upgrade possibilities shouldn't be bothered with the support plan. Others who fancy oiling their rigs for a meaningful performance uplift without forking out for a platform swap would certainly find the same platform fwd-Gen-support proposition attractive. If AMD pulls a dirty on its promise... and stays dirty, I'm gonna join U!!

Most people don't upgrade each new generation and usually use their processor for 4-6 years, which could potentially be outside of the supported period.

Even if that was way more important, current performance and value trumps everything by a mile.

 

[Tech_fanatic]

It must be a fab or manufacturing limitation or even if it was possible it would be insanely expensive for a gaming cpu where today's ryzen 5s and core i5s do a very decent job by themselves.