Speculation: Ryzen 3000 series

[B-Riz]

I thought of something this morning: What if the R7 1700 fire sale @ ~$135 was a test by AMD for demand? Not just an inventory clearing maneuver? Intel has never done anything like that; selling a two year old product for less than half what it launched at.

 

[coercitiv]

You should just give it up, just are just burying yourself deeper.

The deeper the hole, the bigger the meltdown, the longer the silence that follows.

 

[Abwx]

what I said was that the market that really needs the performance already bought it two years ago.

The needs of people do change,that's a big part of the problem all the people that already got 3-4000 points in CB two years ago need something better right now and not the same thing.

Lol OMG please tell me you think this will be a great gaming machine......

And? Do you consider it for your business or as a hobby?
.

It s a hobby if one use a chip overclocked to death for corporate usage...

Did it cross your mind that this i9 must be at 350W or so and wouldnt even pass something like Linpack.?..

So this perf didnt previously exist, at least not at 1400$ for Intel SKUs, on the other hand, and assuming that it s no better silicon that what was displayed at CES, the 16C Zen + was in the 125-135W range for this CB run.

 

[PotatoWithEarsOnSide]

He's missing the point anyway; AMD have had that level of performance with Threadripper for quite some time. Here we are talking about the consumer platform, on which such level of performance was beyond imagination 2 years ago when Intel were throttling the market with 4c CPUs. What's even more impressive is that AMD are bringing the consumer platform close to HEDT in terms of connectivity, and they certainly aren't doing it at Intel pricing. IMO, Zen 2 is a game-changer.

 

[IEC]

X570 with 16 cores is shaping up to meet my HEDT needs, quite possibly without the HEDT premium...

A bit of a bummer that TR/X399 doesn't appear to be getting any 7nm love (yet), but that's probably better for my upgrade budget, hehe.

 

[amrnuke]

Doesn't change the fact that everybody who actually needed a 16/32 CPU capable of getting 3 to 4 thousand points in CB had a two year head start of getting one.
Who are you going to sell one of those now after two years?Only hobbyists that don't actually need it but want it to play around with.
Also the results and clocks are rumors so the amount of juice it took (for the ryzen as well) is besides the point,let the results be confirmed first.

What, exactly, do you think the Threadripper 1950X was doing in mid 2017? Besides being the only HEDT at the time capable of getting over 3000 MT on CB and topping every rendering chart Anandtech tested (well, ok, except for single-core CB)? For reference, Intel's top HEDT chip was almost a 1000 points behind the 1950X and cost the same.

Edit: Yes, I know the 7980XE released a couple of months later and offered 3000+ CB MT scores. Then several months later the 2990WX and 2950X came out and smoked it. And then the latest -X stuff from Intel came out. Long story short, it's been a battle since mid 2017, not a one-time Intel release in 2017 followed by a mid-2019 AMD release. You need an HEDT CPU history lesson.

 

[amrnuke]

X570 with 16 cores is shaping up to meet my HEDT needs, quite possibly without the HEDT premium...

A bit of a bummer that TR/X399 doesn't appear to be getting any 7nm love (yet), but that's probably better for my upgrade budget, hehe.

Is there any impression as to what they'll do with TR? It seems to me it would make little sense to offer a 16/32 chip, if the Ryzen 39xx already will provide that at a lower cost with higher single-core performance to boot.

So perhaps the TR 3xxx would have to be at least 24/48, 32/64, possibly 64/128, though 64/128 seems untenable based on what I've heard as far as limitations on memory management etc (128 threads and only 128GB memory support, for example).

What are the expectations for TR next year, since I doubt they'll kill it given Intel's seeming commitment to this high-margin niche?

 

[lightmanek]

Is there any impression as to what they'll do with TR? It seems to me it would make little sense to offer a 16/32 chip, if the Ryzen 39xx already will provide that at a lower cost with higher single-core performance to boot.

So perhaps the TR 3xxx would have to be at least 24/48, 32/64, possibly 64/128, though 64/128 seems untenable based on what I've heard as far as limitations on memory management etc (128 threads and only 128GB memory support, for example).

What are the expectations for TR next year, since I doubt they'll kill it given Intel's seeming commitment to this high-margin niche?

Don't forget that TR has nearly double the power headroom and twice the memory channels. 16 core TR can and will have its place in market.

For some, having moar PCIe 4.0 will probably matter too 😉

 

[B-Riz]

I am guessing TR3 starts at 24c/48t, and maybe refreshed boards.

It's going to be a beast if Zen2 AM4 plays games as well as we are hoping; so, a perfect high end system, as TR2 already is a productivity heavy weight.

 

[Mockingbird]

I am guessing TR3 starts at 24c/48t, and maybe refreshed boards.

It's going to be a beast if Zen2 AM4 plays games as well as we are hoping; so, a perfect high end system, as TR2 already is a productivity heavy weight.

Well, Ryzen Threadripper used to start at 8-cores.

More likely, that gets bump to 16-cores.

 

[Kedas]

Am I correct in the assumption that there will be no motherboards that make more PCIe lances by splitting 1 PCIe4.0 in 2 PCIe 3.0 (like 2*8 PCIe 3.0 + 8 extra PCIe 3.0 lanes)
A bit of a missed opportunity if you compare it with current i9 platform they will compare it with. (instead of TR)

 

[LightningZ71]

I could easily see it offering everything from 16 cores to 64 cores in stair steps of 16 cores, or 8 cores if they have a 4 chiplet mode for the IO chip and 4 cores if they have a 2 chiplet mode for that same IO die. I suspect that the 4 chiplet mode may be the smallest that they bother with as 4 chiplets with 4 cores each would likely be the least cost configuration that has any real market relevance.

 

[Tuna-Fish]

Am I correct in the assumption that there will be no motherboards that make more PCIe lances by splitting 1 PCIe4.0 in 2 PCIe 3.0 (like 2*8 PCIe 3.0 + 8 extra PCIe 3.0 lanes)
A bit of a missed opportunity if you compare it with current i9 platform they will compare it with. (instead of TR)

I would be extremely surprised if the x570 PCH did not do just that. It takes in 4x PCIe 4.0, and it almost certainly makes available more, slower lanes.

 

[zinfamous]

@Markfw x570 board with heat pipes and fan.
Fan only activated under load

https://www.google.com/amp/s/wccfte...-gaming-plus-motherboards-amd-ryzen-3000/amp/

hmm. that massive heatsink seems to share radiating duties with the 2x M.2 slots as well as the chipset. ...I'm not sure about that. If the chipset is pumping out enough heat to need an active fan, isn't that hot air going to hurt M.2 performance which, for pro-class SSDs can still throttle quite severely?

 

[zinfamous]

instead I just wish this forum had the option to let me hide your comments.

psssst: (it does)

 

[zinfamous]

I'm thinking X570 is going to be $160+ for a board, and only *needed* for the highest perf cpu's or to get the features with a non X cpu.

I still think they are going to release an 8c/16t 65w (12c/24t???) something along the lines of $200 ish (less than that?) as two years ago the R7 1700 was $329 at launch.

That is the key here. AMD is going to keep pushing the 65w more cores cpu on the cheap. Everything else is icing on the cake.

It's probably worth the cost alone if you can get those reported memory speeds with Zen2, assuming 4x0 boards won't be able to do them--I can't imagine they will?

Otherwise, reports of Bios flashes for 470 boards that allow PCIe 4 seems like a mostly no-brainer to stick with those boards, at least another year or two if you are already invested into an expensive setup.

 

[zinfamous]

Big business is with big companies and those will only buy cutting edge.
Yes there are still potential buyers, but most of the market (big companies) already have what they need.

Great. Sounds like you're making an argument for Rome, then. Not this 16c consumer chip.

Oh, big business isn't interested in what Rome is bringing? Do tell.....

 

[DrMrLordX]

And? Do you consider it for your business or as a hobby?

If you have to ask, you've already missed my point.

16c Matisse gives me more performance for less power than anything 16c I could have purchased in the HEDT space two years ago, or even one year ago. All that, for less money. If you claim "nobody wants this chip" because they already got a 1950X, then you're nuts. A lot of 1950X owners are going to be looking hard at 16C Matisse unless they are patient enough to wait for the next gen of Threadripper!

X570 with 16 cores is shaping up to meet my HEDT needs, quite possibly without the HEDT premium...

That's exactly what I'm talking about. Only a few HEDT users actually use all those extra PCIe lanes (which, admittedly, are kind of nice). The real question will be whether DDR4-4000+ will be enough bandwidth for a 16c chip. Bring on the benchmarks.

 

[Fanatical Meat]

hmm. that massive heatsink seems to share radiating duties with the 2x M.2 slots as well as the chipset. ...I'm not sure about that. If the chipset is pumping out enough heat to need an active fan, isn't that hot air going to hurt M.2 performance which, for pro-class SSDs can still throttle quite severely?

The vendor rep state the fan doesn't always run and implied that there is a lot of "fast stuff when all the fast stuff is being powered the fan runs"
This has lead people to speculate the fan is mainly for M.2 raid setups.

 

[Markfw]

Well, Ryzen Threadripper used to start at 8-cores.

More likely, that gets bump to 16-cores.

The very first threadripper was 16 core, the 1950x, I know, I bought one of the first available. the 12 core and 8 core came later. Then came 24 core, and then 32 core, and I own all of them.

 

[realibrad]

The vendor rep state the fan doesn't always run and implied that there is a lot of "fast stuff when all the fast stuff is being powered the fan runs"
This has lead people to speculate the fan is mainly for M.2 raid setups.

That, and buildzoid also said that was the reason he was given by insiders.

 

[amd6502]

45nm PDSOI Quad-core Bulldozer had a F_nom of 3.2 GHz, 32nm PDSOI Octo-core Bulldozer had a F_nom of 3.5 GHz.

=> HKMG
=> Next-gen Strain/Stress
=> etc, however.

TSMC's FinFETs are a severe downgrade from GlobalFoundries' FinFETs. So, the frequency guaranteed for 9T-HPC @ GlobalFoundries, will definitely not be achieved at 7.5T-HPC @ TSMC. 6T vs 6T edges for GlobalFoundries being able to go 1.5x power for 1.7x performance, whereas TSMC went 1.25x power for ~1.475x performance. GlobalFoundries also reduced various Resistance, Capacitance, leakage metrics to allow for really absurd speeds; IBM z, power, etc.

But they then left IBM stranded by then abandoning 7nm for FDSOI. Do you know the current status of z/power? have they moved to TSMC?

No doubt 32nm was better than 45nm. Amd then went on to utilize 28nm to the max, even though it was worse for frequencies than their older hkmg 32nm. It was a good node for APU and GPU though, and it turned out long lived (a little too long lived maybe).

 

[TheELF]

Great. Sounds like you're making an argument for Rome, then. Not this 16c consumer chip.

Oh, big business isn't interested in what Rome is bringing? Do tell.....

I'm sure they very much are interested in rome.
Right now they have to compete against servers with persistent memory that get to skip a lot of the CPUs load store cycles from and to the main storage and thus has a huge benefit over anything else on the market especially in anything that has to do with lots of data.
https://www.storagereview.com/intel_optane_dc_persistent_memory_module_pmm
In a couple of years they will have to compete against PCIe5 and DDR6 as well.
https://www.top7buy.com/intels-server-route-map-shows-ddr-5-and-pcie-5-0-memory-in-2021/

 

[TheGiant]

I'm sure they very much are interested in rome.
Right now they have to compete against servers with persistent memory that get to skip a lot of the CPUs load store cycles from and to the main storage and thus has a huge benefit over anything else on the market especially in anything that has to do with lots of data.
https://www.storagereview.com/intel_optane_dc_persistent_memory_module_pmm
In a couple of years they will have to compete against PCIe5 and DDR6 as well.
https://www.top7buy.com/intels-server-route-map-shows-ddr-5-and-pcie-5-0-memory-in-2021/

well atm noone finds intel's server cascade lake offering as bad...
just they cannot compete on power level and that is of high importance for providers...

about that 3000 desktop series....I dont get the hype around 16C at 5GHz....
bring us 8C/16T with CFL 5GHz performance and 100W, that is enough for me to buy it

 

[sgeocla123]

I'm sure they very much are interested in rome.
Right now they have to compete against servers with persistent memory that get to skip a lot of the CPUs load store cycles from and to the main storage and thus has a huge benefit over anything else on the market especially in anything that has to do with lots of data.
https://www.storagereview.com/intel_optane_dc_persistent_memory_module_pmm
In a couple of years they will have to compete against PCIe5 and DDR6 as well.
https://www.top7buy.com/intels-server-route-map-shows-ddr-5-and-pcie-5-0-memory-in-2021/

PCIe 4.0 NVMe RAID drives setups are closing in on that Optane advantage fast. Also Intel is not going to be the only one offering persistent memory for much longer https://blocksandfiles.com/2019/04/...memory-extension-tech-for-epyc-optane-battle/

As regards PCIe 5.0 and DDR5 roadmaps, Intel has been churning away at roadmaps for some while and changing them so fast that nobody knows what's really going on engineering wise. Intel has already lost the node advantage and is late to offer PCIe 4.0.
They still have a monolithic design which is less flexible and much slower to develop than AMD's IO+chiplets design.
Those Intel roadmaps are really banking on everything going right and that is overly optimistic considering Intel's recent history.