Speculation: Ryzen 3000 series

[Topweasel]

Fair enough. Sorry, I was pretty snarky yesterday because Vettel lost the Canadian Grand Prix due to a stupid call by the race stewards.

Damn, not your fault. But surprising place to get spoiled about the race. Was/am going to watch it tonight after work. Sucks. Ferrari really needs a win right now.

 

[Atari2600]

Yeah, disgusting that the wee spiteful pr**k Hamilton is closing in on Schuey's records.

Vettel is now useless, can't handle the pressure anymore.

 

[DrMrLordX]

Vettel is now useless, can't handle the pressure anymore.

I'll bet he has a DiY PC with an A320 board, too.

 

[Atari2600]

I'll bet he has a DiY PC with an A320 board, too.

I have it on good account he got himself a 7740X...

 

[Topweasel]

Yeah, disgusting that the wee spiteful pr**k Hamilton is closing in on Schuey's records.

Vettel is now useless, can't handle the pressure anymore.

Yeah he is closing on Schumi's original retirement age. Hoping that he pops out before that. But he seems to be totally absorbed with breaking these records considering how he has handled a couple of them. Which means I don't think he hangs up the helmut till he does.

 

[Paratus]

Good question! We don't exactly know what is the minimum bandwidth for proper operation of a 8c Matisse, so it's difficult to estimate what a 16c monster will need before it undergoes starvation.

Maybe someone with a 3800x at fixed clocks can try running single-channel and double-channel mode to help us find the actual bandwidth requirements for the platform. Cutting channels in half is the only easy way to reduce bandwidth without notably affecting latency.



Heh woops. Hope the boss didn't notice.

Not sure this helps but I ran memory benchmarks on my TR1900X 8C/16T

Going from 2133 loose timings to 3333 tight timings does improve performance depending on benchmark.

Gen1 threadripper has some cache improvements over Zen 1 but performance does improve with bandwidth higher than 2 channel DDR 4266 (4 channel DDR 2133).

Hopefully the new cache structure and memory controller of Zen2 on AM4 will minimize the performance penalty of the lower bandwidth from 2 channel memory.

 

[VirtualLarry]

Also, if you want a 3950x, it may be time for you to get your feet wet. I don't see these things doing very well on air. They'll be stuck at or near base clocks during most operations that actually utilize all those cores.

Whoa, whoa, whoa. Let's not start the rumor mill that "Larry's never tried water-cooling". I said I've never done custom water. I've used AIO WC (these past two years or more), for my Ryzen rigs, some of them, anyways. My R7 2700 @ 4.0Ghz / 1.35-ish V is on a 240mm CoolerMaster AIO, for example.

I'm just afraid of custom water leaking all over my system and killing it, due to some incompetence on my part. (*Yeah, I know, they say to "leak-test it first, I guess. Not quite sure how I would go about that.)

 

[Ajay]

Not sure this helps but I ran memory benchmarks on my TR1900X 8C/16T

It is an interesting data point. What would be more instructive would be a 2950 (16 core) on two and four channels, IMO.

 

[zinfamous]

Fair enough. Sorry, I was pretty snarky yesterday because Vettel lost the Canadian Grand Prix due to a stupid call by the race stewards.

yeah, I saw that. Was quite ridiculous.

 

[Ajay]

I'm just afraid of custom water leaking all over my system and killing it, due to some incompetence on my part. (*Yeah, I know, they say to "leak-test it first, I guess. Not quite sure how I would go about that.)

I used to put a little sealant on the back end of the barbs because I was paranoid. Oh, and used hose clamps as well. The real bitch is annual maintenance.

 

[jpiniero]

Videocardz says the 16 core is being released in September.

 

[exquisitechar]

Videocardz says the 16 core is being released in September.

There's also this.

 

[DrMrLordX]

Not sure this helps but I ran memory benchmarks on my TR1900X 8C/16T

Going from 2133 loose timings to 3333 tight timings does improve performance depending on benchmark.

Gen1 threadripper has some cache improvements over Zen 1 but performance does improve with bandwidth higher than 2 channel DDR 4266 (4 channel DDR 2133).

Hmm, can you switch from DDR4-3333 quad channel to DDR4-3333 dual and single channel?

Whoa, whoa, whoa. Let's not start the rumor mill that "Larry's never tried water-cooling". I said I've never done custom water. I've used AIO WC (these past two years or more), for my Ryzen rigs, some of them, anyways. My R7 2700 @ 4.0Ghz / 1.35-ish V is on a 240mm CoolerMaster AIO, for example.

I'm just afraid of custom water leaking all over my system and killing it, due to some incompetence on my part. (*Yeah, I know, they say to "leak-test it first, I guess. Not quite sure how I would go about that.)

I don't know if AiO is going to cut it for 16c overclocking. Yeah it's got that 105W stock TDP, but still. Most AiO are going to crap out in the 250-300W territory, if they even get there.

There's also this.

That seems to imply that DDR4-3733 is as far as they can go with 1:1 IF ratio.

 

[inf64]

I hope we get some slides that cover PBO+XFR on the new cores. 4.5Ghz all core clock on 3950X via one click on premium WC?

 

[Markfw]

Whoa, whoa, whoa. Let's not start the rumor mill that "Larry's never tried water-cooling". I said I've never done custom water. I've used AIO WC (these past two years or more), for my Ryzen rigs, some of them, anyways. My R7 2700 @ 4.0Ghz / 1.35-ish V is on a 240mm CoolerMaster AIO, for example.

I'm just afraid of custom water leaking all over my system and killing it, due to some incompetence on my part. (*Yeah, I know, they say to "leak-test it first, I guess. Not quite sure how I would go about that.)

Build the entire water system WITHOUT THE MOTHERBOARD OR PSU INSTALLED, then use an external power source (I can link it on Amazon) to fire it up and test it Run it for quite a while, and use TP to look for leaks (since it soaks it up so quickly). Once its bullet-proof with no leaks, then you install the motherboard. The only thing is make sure the hose lengths are good on the water block, maybe leave a couple of inches extra.

 

[Fanatical Meat]

Build the entire water system WITHOUT THE MOTHERBOARD OR PSU INSTALLED, then use an external power source (I can link it on Amazon) to fire it up and test it Run it for quite a while, and use TP to look for leaks (since it soaks it up so quickly). Once its bullet-proof with no leaks, then you install the motherboard. The only thing is make sure the hose lengths are good on the water block, maybe leave a couple of inches extra.

Leap frog them all and do this @VirtualLarry

 

[VirtualLarry]

The real bitch is annual maintenance.

Oh, yeah, that too. I'm way too lazy for that.

 

[zinfamous]

Leap frog them all and do this @VirtualLarry

What a brilliant case. I bet that thing could double as a sous vide oven!

 

[Fanatical Meat]

What a brilliant case. I bet that thing could double as a sous vide oven!

It’s a nerd dream of mine to build a submerged PC on a one piece glass tank.

 

[.vodka]

It's started

 

[Atari2600]

No 16C. That would have been the point to talk about it.

 

[Zucker2k]

Gaming tho............




Yeah, no trolling tho.....................


esquared
Anandtech Forum Director

 

[Abwx]

If they can do this at 10% average lower frequencies then it s litteraly the end of the game for Intel as far as perf crown goes, and that s in games wich was their last stronghold thanks to clocks..

 

[thigobr]

Seems ok numbers but probably that new 9900KS will be few % faster.
I will wait for the 16C part before upgrading...

 

[Markfw]

So, not counting "ties" Ryzen wins 3 and 9900k wins 2 if I am correct. About what I expected, but those ARE AMD slides. We still have to wait for independent benchmarks.