(16core added)AMD 3000 specs! 4.7 GHZ, R9 3950x, R7 3700x, 3800x.

[S w a t s o n]

Just now, Drak3 said:

Give us an actual source. Not excuses. Not hearsay.

 

I don't care that it's technically possible. That's irrelevant to the question of whether AMD was even planning on one to begin with.

 

It's unofficial information? If there was an official AMD source it would be an announced product, just stop trying to deny the 16 core dream

[leadeater]

4 minutes ago, S w a t s o n said:

It's unofficial information? If there was an official AMD source it would be an announced product, just stop trying to deny the 16 core dream

Only problem I can see is the SKU naming, everything so far goes up in 100s and we're already at 3900X. Using 3950X is possible but a bit odd. Limiting AM4 to 12 cores could be a market segmentation choice so Threadripper has enough room product line wise. I don't think AMD want to self compete so the easiest way not to do that is to not have 16 cores on AM4.

[S w a t s o n]

1 minute ago, leadeater said:

Only problem I can see is the SKU naming, everything so far goes up in 100s and we're already at 3900X. Using 3950X is possible but a bit odd. Limiting AM4 to 12 cores would be a market segmentation choice so Threadripper has enough room product line wise. I don't think AMD want to self compete so the easiest way not to do that is to not have 16 cores on AM4.

I dont think Threadripper has any problem being the 16+ core option up to 32 (48?) cores. AM4 will have a 16-core no matter what.

[leadeater]

1 minute ago, S w a t s o n said:

I dont think Threadripper has any problem being the 16+ core option up to 32 cores. AM4 will have a 16-core no matter what.

Correct but I suspect it will start at 16 cores which is where you'd have the self competing SKU. Edit: Not that I doubt AMD will do a 16 core SKU on AM4 at some point.

[S w a t s o n]

Just now, leadeater said:

Correct but I suspect it will start at 16 cores which is where you'd have the self competing SKU.

I agree, there will be a 16 core threadripper like there was an 8 core threadripper. More pcie lanes and memory channels if you want to pay up

[Humbug]

Great news- the chips will still overclock to 5Ghz!! As per Adored's highly reliable source.

 

 

LOL anyway, personally I noticed that Lisa did not talk about clockspeeds gains. That wasn't a focus clearly. No talk about how 7nm has allowed higher clocks etc... So I expect about 4.7Ghz common overlclocks without doing anything special. 100Mhz higher than the highest boosting stock part would be a reasonable guess methinks. But as 7nm matures over the next year I expect it to get better.

[leadeater]

@S w a t s o n

Also that previous demo last year of Zen 2 that was an 8 core had a single chiplet on it, where they compared it to the 9900K.

[S w a t s o n]

1 minute ago, VegetableStu said:

 

so the difference would be PCIe lane count, memory channel count, and perf/$ ._.

 

also the other thing would be if a 16-core TR 3000 part would have a 8x2 or a 4x4 core layout ._. so far we can safely guess that AMD are willing to chain 6-core and 8-core chiplets ._. (and 4-core parts hasn't been announced yet)

I suspect 4x4 could be great for temps but 2x8 makes sense to not just completely waste dies

[Drak3]

2 minutes ago, Humbug said:

the chips will still overclock to 5Ghz!! As per Adored's highly reliable source.

Given that his last prediction was 25% wrong, I now fully expect 3.75GHz.

 

 

 

 

 

 

 

 

 

 

/s

[S w a t s o n]

1 minute ago, S w a t s o n said:

I suspect 4x4 could be great for temps but 2x8 makes sense to not just completely waste dies

 

3 minutes ago, VegetableStu said:

 

so the difference would be PCIe lane count, memory channel count, and perf/$ ._. as basis for choosing between the two

 

also the other thing would be if a 16-core TR 3000 part would have a 8x2 or a 4x4 core layout ._. so far we can safely guess that AMD are willing to chain 6-core and 8-core chiplets ._. (and 4-core parts hasn't been announced yet)

Actually they went 4 dies on the 8 core TR didnt they? Maybe they dont care, yields are supposed to be good just like zen 1 70%vs80%

[leadeater]

5 minutes ago, S w a t s o n said:

I suspect 4x4 could be great for temps but 2x8 makes sense to not just completely waste dies

Edit: Nope never mind remembered wrong, IGNORE

[S w a t s o n]

Just now, leadeater said:

Pretty sure the PCIe controllers are on the chiplets so that should be a rather important factor, if they want upper SKUs to have more lanes or not etc.

IO Die?

[leadeater]

Just now, S w a t s o n said:

IO Die?

SHHHHHHHHHHHHhhhhhhhhh . Thought only the memory controllers and other stuff was on it, remembering an old diagram I saw which I half knew was wrong and forgot.

[Soppro]

It's exciting to finally see some major things happening in the CPU world

Your move, Intel

[leadeater]

2 minutes ago, VegetableStu said:

whoop whoop whoop emergency emergency

I was heavily considering buying AMD stock recently, wanted solar panels more but that may have been a bad decision financial gain wise.

[melete]

2 minutes ago, Soppro said:

It's exciting to finally see some major things happening in the CPU world

Your move, Intel

*Delays 10nm desktop CPUs again* 

[Soppro]

9 minutes ago, melete said:

*Delays 10nm desktop CPUs again* 

ಠ_ಠ 

Wouldn't be surprised if it actually happens 

[Valentyn]

Whelp, I'm looking at that 3900X, with 32GB 3600-4000MHZ RAM in my future. Maybe one of those fancy new NVMe SSDs also.

 

Will be a massive upgrade over my 5820K.

 

12 minutes ago, leadeater said:

I was heavily considering buying AMD stock recently, wanted solar panels more but that may have been a bad decision financial gain wise.

I should have bought more, during the last big drop when Ryzen 1 , and Vega launched. Although they ensure the next free PC upgrades for years to come

[CarlBar]

48 minutes ago, VegetableStu said:

had another table done, this time to include R5 and the related R3 rumor

 

Adored			AMD
R3 3300			R5 3600
	6c12T?			6c12T
	3.2/4.0 Ghz			3.6/4.2 Ghz
	50W			65W
	$100			$200
R3 3300X			R5 3600X
	6c12T?			6c12T
	3.5/4.3 Ghz			3.8/4.4 Ghz
	65W			95W
	$130			$250
R5 3600			R7 3700X
	8c16T?			8c16T
	3.6/4.4 Ghz			3.6/4.4 Ghz
	65W			65W
	$180			$330
R5 3600X			R7 3800X
	8c16T?			8c16T
	4.0/4.8 Ghz			3.9/4.5 Ghz
	95W			105W
	$230			$400
R7 3700
	12c24T?
	3.8/4.6 Ghz
	105W
	$330
R7 3700X			R9 3900X
	12c24T?			12c24T
	4.2/5.0 Ghz			3.8/4.6 Ghz
	105W			105W
	$330			$500
R9 3800X
	16c32T?!
	3.9/4.7 Ghz
	125W
	$450
R9 3850X
	16c32T?!
	4.3/5.1 Ghz
	135W?
	$500

 

 

one of the 8c16T part seems to be spot on. maybe because THAT's what the CPU was during the CPU teaser demo at CES

 

You've got the 12c on the right in the wrong space, move it up one, it's a perfect match. I never personally expected AMD to release 5Ghz base clocked chips, though i belive they could if intel had pulled a rabbit out of their hats, (i actually suspect the info he gets is official- unofficial leaks where they knowingly leak him a mix of good and bad info, he's got early info on too many things not to have a source inside AMD, but the error's are schoolboy enough that i kinda doubt they're by accident).

 

 

As far as OC headroom. We know from the data AMD has provided officially today that a 16 core with a 4.4Ghz boost clock should be around 130w, and we know from GN that it can pull upto 300w on a water loop. Even with terrible scaling that adds upto a lot more headroom than people are assuming. Yes Zen1/Zen+ didn't have much headroom, but that also meant max OC'd power wasn't especially amazing ethier, (on ambient obviously). 300w is 2.5 times the power draw. A mere 14% frequency uptick will get them to 5Ghz. less than 14% frequency from a 150% extra power is so diabolically bad i have trouble believing it.

 

Forget details about the process node, those power numbers alone indicate a fair bit fo headroom.

[CarlBar]

1 minute ago, VegetableStu said:

'DOH

 

Don't sweat it. I'm only still awake because i'm i'll so struggling to drop, god knows what mistakes i'd have made if i'd made that.

[CarlBar]

2 minutes ago, VegetableStu said:

 

Perfect description :).

[porina]

Random initial thoughts:

 

Nice boost to clock, not the 5 GHz hype. If we look at history, the wall usually isn't much further than the highest production boost. Based on this hitting 5 GHz will probably be a challenge for extreme cooling.

 

Expected FP upgrade (but we knew that from Rome reveal)

 

32MB of L3 cache? Yes please! For my uses this should adequately mitigate the crippling limitations of dual channel ram. 

 

The claim for IPC will have to be tested to be understood. I stand by my earlier comments that I would only expect a large increase like the 15% claimed if it is blended workload that includes the massive FP upgrade. If FP is excluded, it'll probably be much smaller.

 

So what do I buy for test? I'm a bit torn between the 3700X, 3800X, 3900X. For competitive benching, I wonder if 3800X has the edge on clock potential, but more cores on 3900X. Do I want to deal with two chiplet scaling yet? If building a general purpose non-OC system, 3700X is more logical due to power.

 

I hope this will finally shut up Adored and people will stop reposting that crap. Any other random person on the internet would have got as much right by chance as he did.

[melete]

31 minutes ago, VegetableStu said:

whoop whoop whoop emergency emergency

 

 

lol, this video includes Linus recounting MSI's advice to motherboard buyers for overclocking the 16 core CPU people still aren't sure exists. Around 7 minutes in.

[Rocktopus306]

33 minutes ago, VegetableStu said:

whoop whoop whoop emergency emergency

 

Hehehe
Bought some on Friday.

[CarlBar]

4 minutes ago, porina said:

I hope this will finally shut up Adored and people will stop reposting that crap. Any other random person on the internet would have got as much right by chance as he did.

 

You didn't do very well at statistics did you?

 

Seriously even using a bit of brain power to limit things to reasonable ranges the number of possibble options for each item is huge. The odds of getting even one of them right are painfully low. getting chiplets existence right so far in advance and then 3 sets of hardware specs, (out of 6 so a 50% hit rate), is way outside of probability. And getting the full combination of clocks and TDP right for each spec set he was right about in 3 seperate cases and getting none of the rest right is way outside of any sane statistical probability. Possible in theory but the odds are so long as to be nigh impossible.

 

Now anyone who hasn't assumed he's going to get even some hard data type stuff wrong, and that anything subject to change will change. Well thats on them.