Speculation: Ryzen 3000 series

[Shivansps]

Good problem solved stop whining about the A320

Who whinning? I said A520 news settle the matter of A320, as many belived it would not be replaced because it was worthless.

 

[ubern00b]

My Asrock AB350m which I bought in 2017 has one of the best VRM's for a b350 chipset and still better than a lot of X370/B450 budget motherboards, £60 2 years ago, bios update to support Zen2 also, waiting to jump on a 3600x and OC it 4.5Ghz all core

 

[DrMrLordX]

Yes it does, the A320 is now certified poop.

At least it's certified. Can you say that? I can't!

Who whinning? I said A520 news settle the matter of A320, as many belived it would not be replaced because it was worthless.

I figured A320 would just be replaced. OEMs would rather sell you something new than let you keep trying to update the same platform. What's funny is that AMD launched A320 - after a fashion - months before Summit Ridge, almost as a platform pipecleaner (if such a thing exists). This time around, AMD is going full-bore with X570 first.

 

[Fanatical Meat]

At least it's certified. Can you say that? I can't!

I work in real estate, everything can be certified. A Proctologist or a Doctor who performs colonoscopies or even a sewerage worker could certify a poop.

 

[DrMrLordX]

I work in real estate, everything can be certified. A Proctologist or a Doctor who performs colonoscopies or even a sewerage worker could certify a poop.

At least someone can figure out this A320 crap.

 

[prtskg]

The cost of 550 and 520 boards won't be too different from their earlier counterparts as they are still on PCIe.

 

[Kedas]

The rumor of the X590 is a bit strange because if you are in that price bracket wouldn't it make more sense to buy a threadripper board.

 

[Gideon]

The rumor of the X590 is a bit strange because if you are in that price bracket wouldn't it make more sense to buy a threadripper board.

That's why i thought it might be the name of the new threadripper board (in order to not confuse it with intel x599). But some rumour sites are adamant that the 99 suffix is kept, so who knows

 

[Ajay]

Dammit, could we please stop having endless debates about motherboards in this CPU thread!!!

 

[DrMrLordX]

Dammit, could we please stop having endless debates about motherboards in this CPU thread!!!

It would seem not.

 

[jpiniero]

https://twitter.com/x/status/1141027878634573824

Possible Rome spec list. Oddly, they are doing all the way down to 8 core (at 155 and 120 W no less). Also no 250 W parts, only 225 and 200 for 64 cores.

 

[Markfw]

https://twitter.com/x/status/1141027878634573824

Possible Rome spec list. Oddly, they are doing all the way down to 8 core (at 155 and 120 W no less). Also no 250 W parts, only 225 and 200 for 64 cores.

Well, for some IO applications, 8 core is enough. The way Rome is made, this makes sense to cover all parts of the market. The above may cover all server applications today, and only requires a IO chiplet, and an 8 core chiplet, and just put them together like legos to get the desired CPU.

 

[Tuna-Fish]

https://twitter.com/x/status/1141027878634573824

Possible Rome spec list. Oddly, they are doing all the way down to 8 core (at 155 and 120 W no less). Also no 250 W parts, only 225 and 200 for 64 cores.

The main reason for having low-core parts for the platform is that plenty of really expensive software is licensed per core. (And those license costs can be steep, often exceeding the cost of the hardware they run on by 50x) This leads to companies optimizing for having the beefiest possible system, while still having a low core count. Intel makes cpus specialized for this too, the very highest clocking low-core-count Xeons are actually auctioned, with no public price. The actual sale prices on those are much higher than the list price of the highest core count xeon.

 

[ubern00b]

So yes, ARM is a much more refined core than x86 which is a big stupid core and can't keep up with ARM in IPC or efficency but the whole wide world doesnt seem to see this goliath vs david

 

[jpiniero]

Intel makes cpus specialized for this too, the very highest clocking low-core-count Xeons are actually auctioned, with no public price. The actual sale prices on those are much higher than the list price of the highest core count xeon.

Those are more for HFT than anything else. The high licensing cost per core crowd would still want more than 8 cores I would think, if only for density reasons. There are some others who want lots of L3 cache per core, I don't know if AMD would ship 8 core products with 128 MB L3.

 

[beginner99]

Oddly, they are doing all the way down to 8 core (at 155 and 120 W no less)

I suspect the high wattage is due to the bigger IO die compared to desktop IO die.

 

[Tuna-Fish]

Those are more for HFT than anything else. The high licensing cost per core crowd would still want more than 8 cores I would think, if only for density reasons.

No, they really wouldn't. If the software costs you $200,000 a year if you run it on an 8-core, or $800,000 if you run it on a 32-core, you don't give a damn about density. (No, you usually cannot just decide to only use some of the cores for a given software. For instance, Oracle bills you for every core that your code could plausibly run on in your setup.)

 

[moinmoin]

Oddly, they are doing all the way down to 8 core (at 155 and 120 W no less).

I guess those will contain 4 chiplets at most (1 activated core per CCX).

I suspect the high wattage is due to the bigger IO die compared to desktop IO die.

Definitely.

 

[Topweasel]

No, they really wouldn't. If the software costs you $200,000 a year if you run it on an 8-core, or $800,000 if you run it on a 32-core, you don't give a damn about density. (No, you usually cannot just decide to only use some of the cores for a given software. For instance, Oracle bills you for every core that your code could plausibly run on in your setup.)

This is one of the things AMD is really set well with. Cascade Lake AP might give you 50% more memory modules but requires one of their super high but not quite Rome sized packages. Because it is really 2 Skylakes "glued" together. With Rome they can go all the way down to 8 cores (4 Chiplets 2 cores per), without losing any memory channels. So all of a sudden you can 2 super clocked 8 core cpu's and still have 32 slots for memory. Or just 8c and 16 slots. With Rome's seemingly infinite PCIE (128 Lanes of PCIe 4 whether 1S or 2S).

 

[maddie]

I guess those will contain 4 chiplets at most (1 activated core per CCX).


Definitely.

And the highest clocking core in each CCX enabled. Should be the fastest clocked CPUs on sale.

 

[tsamolotoff]

EPYC ROME 8-CORE 7252P 3.2GHZ/SKT SP3 32MB CACHE 120W - P/N: 100-000000081 (1 chiplet)​

EPYC ROME 8-CORE 7252 3.2GHZ/SKT SP3 64MB CACHE 120W - P/N: 100-000000080 (2 chiplet)​

EPYC ROME 8-CORE 7262 3.4GHZ/SKT SP3 64MB CACHE 155W - P/N: 100-000000041 (2 chiplet)​

EPYC ROME 12-CORE 7272 3.2GHZ/SKT SP3 64MB CACHE 120W - P/N: 100-000000079 (2 chiplet)​

EPYC ROME 16-CORE 7282 3.2GHZ/SKT SP3 64MB CACHE 120W - P/N: 100-000000078 (2 chiplet)​

EPYC ROME 16-CORE 7302 3.3GHZ/SKT SP3 128MB CACHE 155W - P/N: 100-000000043 (4 chiplet)​

EPYC ROME 16-CORE 7302P 3.3GHZ/SKT SP3 128MB CACHE 155W - P/N: 100-000000049 (4 chiplet)​

EPYC ROME 24-CORE 7352 3.2GHZ/SKT SP3 128MB CACHE 155W - P/N: 100-000000077 (4 chiplet)​

EPYC ROME 24-CORE 7402 3.35GHZ/SKT SP3 128MB CACHE 180W - P/N: 100-000000046 (4 chiplet)​

EPYC ROME 24-CORE 7402P 3.35GHZ/SKT SP3 128MB CACHE 180W - P/N: 100-000000048 (4 chiplet)​

EPYC ROME 32-CORE 7452 3.35GHZ/SKT SP3 128MB CACHE 155W - P/N: 100-000000057 (4 chiplet)​

EPYC ROME 32-CORE 7502P 3.35GHZ/SKT SP3 128MB CACHE 180W - P/N: 100-000000045 (4 chiplet)​

EPYC ROME 32-CORE 7502 3.35GHZ/SKT SP3 128MB CACHE 180W - P/N: 100-000000054 (4 chiplet)​

EPYC ROME 32-CORE 7542 3.4GHZ/SKT SP3 128MB CACHE 225W - P/N: 100-000000075 (4 chiplet)​

EPYC ROME 48-CORE 7552 3.35GHZ/SKT SP3 192MB CACHE 200W - P/N: 100-000000076 (6 chiplet)​

EPYC ROME 48-CORE 7642 3.4GHZ/SKT SP3 192MB CACHE 225W - P/N: 100-000000074 (6 chiplet)​

EPYC ROME 64-CORE 7702 3.35GHZ/SKT SP3 256MB CACHE 200W - P/N: 100-000000038 (8 chiplet)​

EPYC ROME 64-CORE 7702P 3.35GHZ/SKT SP3 256MB CACHE 200W - P/N: 100-000000047 (8 chiplet)​

EPYC ROME 64-CORE 7742 3.4GHZ/SKT SP3 256MB CACHE 225W - P/N: 100-000000053 (8 chiplet)​

It's even more interesting.

Source:

https://www.reddit.com/r/Amd/comments/c31d4z

 

[Markfw]

that 225 watt 7742, is lower tdp than my 32 core 2990wx, and twice the cores and 400 mhz clocked higher ! Impressive !

 

[Tuna-Fish]

and 400 mhz clocked higher

Those *have* to be boost clocks.

Mostly because if they aren't, AMD just knocked this so far out of the park the spectators are looking backwards and expecting the ball to come back around on it's second orbit.

No, seriously, if the chips can maintain those speeds for any length of time, AMD didn't just beat Intel, they utterly trounced them. If the long-term average speed for the 64-core is ~2GHz it's still a ridiculously good product.

 

[Hitman928]

that 225 watt 7742, is lower tdp than my 32 core 2990wx, and twice the cores and 400 mhz clocked higher ! Impressive !

Those speeds, if true, have to be boost clocks, not base.

 

[Markfw]

Those speeds, if true, have to be boost clocks, not base.

Thats higher than my boost clocks, at twice the cores and less TDP ! Oh, they knocked it out of the park OK.......