Threadripper 2990x - 32c/64t

I mean, it's pretty obviously just 4 Ryzen chips slapped together. Each chip alone is somewhere between a 2700 and a 2700x in performance at 250w TDP, so you should be seeing multicore workloads scale to about 3.5-4x a 2700x's benchmarks. With high-end custom loops, you're looking at 4x or more the 2700x's raw performance.

>1x360mm radiator
>less than 50% coldplate coverage
>overclocked to just under its single core boost (read: 100-200mhz more will be doable with even just the Enermax TR4 coolers)
>beats the Intel 28-core chip under a custom water loop with multiple radiators, at just under 6100 points

I'd spitball this at better than the Intel 28-core chip at all workloads, including singlethreaded ones, at stock, given a 240-360mm radiator and full coldplate coverage. 0-10% better in ST, 10-20% better in MT. About equal (0-5% worse) in ST if it has partial coldplate coverage.

>vega having anything to do with the the CPU division
>old bulldozer meme

Attached: 1528835745426.jpg (727x480, 38K)

it'll be $2k minimum so pony up poorfag or you're going to be spending the rest of your time making bulldozer memes instead of graduating

Attached: 1447161347847.jpg (1846x1923, 607K)

I'm worried about two things, first the original one started having some crazy stability issues past 4.0ghz which in this 7nm fab should be remedied at least a little bit but the fact they also double the amount of cores is a bit scary in the thermal side. To me for example it's just unviable to go around making custom water loops the best I can do since I will use it for some research with research funding is to buy an enermax liqtech 360 radiator or that new noctua one the guys from gamernexus were showing that has like 30% higher fin density.
To me as long as it doesn't get past 70~75 celcius full load at 4.0ghz I'm fine.
Pic related the TR4 socket AIO from enermax.

>Not understanding AMD has been doing the hypetrain for a billion years.
>Not knowing that it's better to wait and see than think anything they promise will happen.
It's like you enjoy being lied to and told to wait for it.

Attached: 35-214-080-V01.jpg (640x480, 31K)

>$2k minimum
>16c/32t 1950x cost $1k at launch, just $750 now.
>1800x cost $500, 2700x cost $350, price drops universal for Ryzen 2
>expecting to pay $2k for this product
I'd say $2k is the absolute maximum, not the minimum. The minimum I'd guesstimate is $1.5k.

>>Not understanding AMD has been doing the hypetrain for a billion years.
>>Not knowing that it's better to wait and see than think anything they promise will happen.
>It's like you enjoy being lied to and told to wait for it.
>vega having anything to do with the CPU division
>old bulldozer meme

Attached: 1528926268121.jpg (960x536, 44K)

I won't buy it until I see proper testing unless you're willing to send an engineering sample and your beta drivers to my physics department marketeer kun.
I want it but I'm not going to jump the gun.

>drivers

Attached: 1529099944782.gif (200x270, 1.09M)

Plain Ryzen couldn't really scale past 4GHz, but Threadripper's cap usually rested at 4.1GHz on almost every chip. 4GHz was near-guaranteed between 1.35 and 1.45v.
TR2 is 12nm, not 7nm, same as the Ryzen 2000 series, which carry massive efficiency gains below 4.1GHz, with most 2700x's being able to do 4-4.1GHz at just 100w. I would expect the 2990x to stay below 500w anywhere from 4.0-4.2GHz.
The 240mm Enermax AIO is capable of cooling ~400w at just ~50c delta, which equates to 72c at room temperature. See the HardOCP review, one of the few 1950x's that couldn't hit 4GHz. They had 3.9 at 1.4v IIRC, which burned ~400w.
You'll be able to cool this one using a LiqTech TR4 240. 280 maximum.

Forgot pic.

Attached: 2700x 100w.png (725x724, 468K)

>250w TDP

Noctua shills

B T F O
T
F
O

Pretty sure I will have to update bios and other crap if I shove it in a taichi x399 motherboard even if it's supposed to be backwards compatible, just like some anons had to do with the 2200g etc in their AM4 mobos.

Gotta go download my CPU drivers really quick

Jesus fucking Christ, shut up and take my money if this is real.

sure pal here you go
support.amd.com/en-us/download/chipset?os=Windows 10 - 64

Thanks for the info user, so the 360 should be able to handle it then.

>He doesn't know about how fucked up people who bought the 2200G and 2400G were.
>He doesn't know that they had to ask AMD for an upgrade kid where they used a shitty AM4 cpu to boot update the bios then would send the CPU back.

user you have no idea how much of a shitshow that "backwards" compatibility was.
youtube.com/watch?v=t_te-ksbGXE

That's the cooler I have on my 1950x. My chip will do 4.2ghz boot but will fall on its face past the login screen and I'm afraid to throw more voltage at it.

At 4ghz, with a push pull fan setup it will stay under 55c load temps with only Prime 95 breaking the 55c mark. At idle, 4.1ghz it can still get as low as 19c not doing anything with the fans kicked on.

The Liqtech also isn't a real closed loop, it has fill ports on the side of the block so you can actually replace the coolant inside.

Overall I'm very happy with mine. It will cool my TR passively with all fans shut off and pump at 100% at 3.4ghz, under 50c

user I need a link for my CPU drivers not my chipset you silly goose
They should've installed their CPU drivers

The 360mm radiator only has ~1-2c delta over the 240. Really, if you put Noctua's new Sterrox fans on the 240, keep the pump running at full speed no matter the CPU temp, and ramp up the fan speed to 100% at ~65c, you should be able to keep this chip below 75c with a single 240. On top of that, it's likely not worth it to lock this chip to anything below 4.2GHz, since XFR2 is so damn good. If you can't lock it to 4.2GHz, it's probably best to leave the CPU at stock (undervolt by offset if possible).

>high-end custom loops will be able to hit 4.3-4.5 all-core
citation needed
8 core ryzen 2 is a lucky chip if you can get 4.3 on water, 4.5 is entirely impossible (unless cooling sub-ambient / way into the negative) due to voltage walls similar to what we saw with Ryzen 1

Attached: 1922563191567884209.jpg (739x605, 74K)

Yeah I don't plan to go past 4.0 anyways because to me that already sounds like a nice OC from stock. Not too high not too low.
If it can keep the 1950x at 55c then it can probably hold the 29xx series at 70-ish even with double core count.
Hopefully I mean.
I'm still considering that new noctua one they showed on computex air coolers seems like a better leave it and forget to me than the AIO since I'm worried some shit could happen with the pump. (had like 3 different corsair ones die on me in the past 7 years but on another build at home)

>le damage control.
Post some smug lolis too user gotta win that internet fight.

Honestly, what did you expect? That the bios updated itself with magic or something?
AMD always stated that there would be backwards compatibility with a new bios released by the mobo manufacturers. If the mobo is using the first Gen chipsets it needs to be update.
You don't really sound like you know what you're talking about.

do you have any idea what the fuck did you post?
the review nigger says on the video that the motherboard has issues, like the wireless card, gpuz bsods when probs bios memory addresses, 3dmark installation corruption which indicates that the south-bridge frequencies are wrong on that motherboard...
literally zero problems about the APU itself...
just another shitty MSI motherboard.

Not if you do PBO overclocking + undervolting. Tell XFR2 you have headroom, and it'll push everything past 4.3GHz when it's needed. If you give it 750w of cooling, it'll push voltages past 1.45 and clocks as high as you let it.

So you're basically saying I could just use the auto overclock and call it a day? Ok then although I'm worried about one thing, auto overclocks always push the shit out of the voltage for no reason because they want easy fast OCs that work with as many chips as possible so I may end up forcing the VRMs too much over time, no?

Also didn't knew about the 240 radiator one, good to know this may help a bit with the case choice.

The noctua TR4 coolers would work if you left it at stock. If they update the 120mm cooler with the Sterrox fans and increased fin density, it'll probably get you to 4GHz without too much of an issue.

not bad

Attached: 1700.jpg (2351x1730, 657K)

Let's not forget the same thing happened with Intel.

Wait no, let's forget.
>
3

0/10 and go fuck yourself

I'm projecting a similar temperature to the 1950x on that AIO, given the smaller die manufacturing process and lower base clock speed.

My 1950x usually stays below 50 doing anything other than benchmarks, my average temps while its being used for games, Photoshop, Lightroom and CAD are between 28c and 35c. It will only touch 50 if doing Passmark or Prime95, so you'll be fine with that.

"forget it" and it'll cool itself, if it wants to live anyway otherwise it gets the hose

The voltage spikes XFR2 produces are entirely safe; they're not long enough to overload your cooling OR the VRMs, and they don't ever push the power consumption past levels determined by the BIOS. If you do PBO-style overclocking like a lot of 2700x users are doing, then the motherboard tells the CPU it can give it basically infinite power, and the CPU responds accordingly. It won't drive voltages high enough for long enough to damage the chip, though; that much is basically guaranteed by the algorithm and dozens of sensors.

tl;dr: leave Ryzen 2000 series at stock. You'll get better performance in almost every case.

>>le damage control.

Attached: Delusion.png (595x514, 481K)

I didn't expect anything I just told the user that if he sent me an engineering sample he would have to send me the beta chipset drivers and bios since he was trying to sell me something that we don't have yet in the stores.

Why are you adamantly defending something like that? A lot of people bought motherboards and that CPU and found out they had to send an email to AMD receive an upgrade kit so they could finally boot up their systems to be able to update the bios for the new cpu.
I'm not saying it's AMD only fault but the companies should have worked on those problems before launch.
Stop being butthurt and defending the mistakes of companies.

Didn't they increase the fin density by like 30%? should be good enough for at leat 4.0 I would say.
at least I hope but user said the 240mm AIO should do the job flawlessly also so maybe that one is not a bad deal since it can fit easily anywhere.

Yep. 2000 series has absurdly good efficiency gains. Compared to your 1700, the 2700x gets +300MHz and -10W.

most x470 boards went overboard with the VRM in anticipation of MOAR COARS for the incoming 7nm CPUs next year.
Just make sure you don't buy a garbage cardboard thing with 4 phases or whatever

>Didn't they increase the fin density by like 30%?
Yes, but only on the standard coldplate. The TR4-compatible coldplate version has much worse performance for Threadripper, even if it's better overall for other CPUs. If you get any cooler for Threadripper, the part you put thermal paste on should cover 100% of the Threadripper heatspreader. Otherwise it'll have bad performance.

How Intel will recover? With a Xeon Phi overclocked to 10GHz, using Helium II?

Attached: 1529363082799.png (512x415, 58K)

I've noticed my 2700x gets fairly higher temperatures than my 1700x did (inb4 why did you upgrade), it essentially all written off. Where as my 1700x got around 58c in cycles renders at 3.8ghz 1.3v ambient 74f, my 2700x gets around 65-68c in cycles renders at 4.2ghz 1.3v 76f ish ambient. These were both under AIO too so most likely the clockspeed jump is causing temperature differences.

valid.x86.fr/qwzl5d
stock cooler.

We got an asrock taichi x399
So it's a better idea to change it with the CPU then? That's starting to get past my poorfag budget but I can think of something.

I'm not planning to buy an intel cpu I wanted the threadripper 2 so none of this matters to me.
How many times intel ever made some backwards compatible shit? they always make a new socket you nigger there's no "yo can use your old mobo on this new shit here son" kinda deal.

>tl;dr: leave Ryzen 2000 series at stock. You'll get better performance in almost every case.
Most of what I will do is vector math wouldn't having higher clock at least as a little bump from 3.4 to 4.0 help?

Guess the enermax AIO seems more fail proof then even though I'm worried about the pump as a point of failure compared to just a tower heatsink.

Anyways thanks for all the info, anons.

help

Attached: wtfarethesedefaultvoltages.png (1051x918, 204K)

Yeah, at same voltages, higher clocks will produce higher temperatures; the real efficiency gain in Ryzen 2 is being able to run higher clockspeeds at the same voltage, as you demonstrate here with a +400MHz boost at the same voltage, leading to ~15% increased power consumption. If you ran it at 4GHz it'd likely have lower temperatures than your 3.8GHz 1700x.

It's weird that you need a full 1.332v to run at 4GHz, most late-production Ryzen 1's would run 4GHz at 1.34v.

>If you ran it at 4GHz it'd likely have lower temperatures than your 3.8GHz 1700x.
Yeah did that experiment, pretty amazing how they did all this shit in a simple CPU refresh.

better sell all that mustard gas you're now producing

what did you do friendposter

Attached: 1522614024435.jpg (323x250, 14K)

>peak speed 4.35
I only have a x370 so I don't get those kinds of gains with XFR, I'll only see that number in single core, which surprisingly my 3rd core is my best

>Most of what I will do is vector math wouldn't having higher clock at least as a little bump from 3.4 to 4.0 help?
Ryzen will run at the maximum clockspeed (up to XFR2 boost) your cooling has the capability to cool at stock. Overclocking will generate a lot more heat for not much performance benefit, unless you have a system capable of cooling it locked to XFR2 boost clocks all-core. If you can't get to that, you're most likely going to see your performance degrade.

1.55v is safe for

offset it lower on your bios.

XFR is an algorithm essentially, it's finding what works and what doesn't. It will randomly send a shot of high voltage through your CPU at times but it's literally nothing to even worry about. Only if it's above 1.45 constantly then you should maybe go manual but Ryzen motherboard love to overvolt when left stock due to mostly incompetence from manufacturers.

it hovers between 1.45-1.55 for no reason at idle unless i set a custom profile

> How will Intel EVER FUCKING RECOVER
Muh games.

1.35v is the highest safe 24/7 voltage for Ryzen 1 and 2. Anything past that and your CPU will degrade over time (under sustained loads). At 1.35 it'll hit exactly the same clocks until your motherboard gives out.

>at idle
what motherboard? You shouldn't be getting those voltages when idling. For now I'd recommend setting 4.0 @ 1.35v locked in the BIOS to be safe.

Shit man, anons from
Scared the shit out of me now.

I think I will go with the tried and tested manual OC with lots of benchmarking and trying to get the lowest voltage possible while keeping it stable aka not getting over 1.35v
I don't want to kill a CPU like that in my situation.

See If you're getting anything over 1.35v at idle, you're killing your CPU. 1.55v is totally safe if it only goes there for

thats normal. if you have an offset mode in bios, offset it lower. im running mine -0.15V offset + precision boost override enabled.

its precision boost override at 10X, so very aggressive. also, the voltages move from 1.25~1.47 and shit, so dont trust it too much.

asus x470f

i dont see an easy way of setting a lower voltage without locking it and disabling precision boost in which case might as well just use ryzen master and keep it at base at a sane voltage

67 posts and no one is asking how these cores are wired up to the DIMM slots and PCI-E. Frankly that's really what matters and am still awaiting details about this. If the two new activated dies have no PCIE or DIMM access and have to be fed by proxy through the two full dies, this is going to be far less appealing.

The current 1950x is almost perfect in that the die parts are all fully active. You slap two more dies into the mix and it's a headscratcher as you don't get any more PCIE slots or DIMMS. If you're doing something hardcore, this is a head scratcher.
> Current 1950x owner
> 67 posts about fucking Overclocking a 250W TDP processor
> Brainlets who obviously can't afford this shit and have no use for it rattling on about customization, stupid ass water cooling, and power inefficient idiocy
> No posts about how the dies are wired up because brainlets
> smfh...

Update your BIOS and reset all CPU settings to stock. Let your CPU idle, use HWINFO64 to measure voltages. Check average. Sometimes XFR2 will boost at idle to help background processes, so occasional 1.55v when idling isn't totally out of the picture. Just check to see if the average is below 1.35v.

No sir, see 2=2 and 8>3

>two posts in one

Anandtech says the new dies get proxy DIMM/PCI-E access. However, Ryzen has massive caches; there PROBABLY won't be too much of a performance deficit for most tasks, assuming you run 3200MHz+ RAM.

i've done that and it does lower the average voltage but then it sets my memory to cl16 even though im running gskill flare-x

this is precision boost override enabled with 102 on bclk. very aggressive imo.

Attached: Capture4.png (1139x1403, 543K)

>max 65c
what benchmark did you run?

Yeah, fuck this then.
The latency is massive as you'll have to traverse :
DIMM>DIE #1=>CCX=>DIE #2
Everytime dimm #2 needs to access RAM
Also, if Die #2 needs to access any PCIE resources or SATA you'll have to :
Resource>DIE #1=>CCX=>DIE #2
The reverse for writes and an even bigger nightmare for cache coherence. The Memory management unit on one die will be getting slammed handling two dies worth of workload.

> However, Ryzen has massive caches;
Doesn't matter. Cache misses happen often in L1/L2/L3 on a number of workloads and you're still ignoring the hits for cache synchronization, PCI-E access, Chipset access. All of this gets loaded on only two dies now for 4 dies worth of cores + you have to squeeze it through CCX. The performance impact is substantial and will vary wildly depending on workloads. 24/32 would be better suited for a whole new mobo w/ 16 dimms. Go E-ATX and don't nigger it like ASUS did and get the cores properly fed. Maybe introduce enterprise level PCI-E riser tech to allow them all to be accessible.
> there PROBABLY won't be too much of a performance deficit for most tasks
No sorry.. there is a deficit for all tasks when you have no DIMM access..
> assuming you run 3200MHz+ RAM.
You obviously know nothing about DIMM access latency + crossing NUMA.
You're talking about a 50% hike in latency

same settings precision boost disabled

Attached: Capture5.png (941x1342, 476K)

BUT 28 CORES 5 GHZ

Attached: 1527629778452.jpg (679x758, 54K)

>10GHz
Not even with liquid nitrogen.

Is ringbus absolutely finished?

Attached: 1488432385535.png (1920x1200, 655K)

but. . . but . . . 28 (((fantasy))) cores . . . 5 nigga. . . hurrt!!

Attached: 1505147990486.jpg (329x329, 57K)

>Why are you adamantly defending something like that? A lot of people bought motherboards and that CPU and found out they had to send an email to AMD receive an upgrade kit so they could finally boot up their systems to be able to update the bios for the new cpu.
I'm not even that guy, but this is like buying a smartphone with USB-C, noticing that you can't charge it with your old Micro USB charger and then, somehow, blaming the manufacturer for your incompetence, even though they explicitly stated that you would need to be DONGLED to do that.
Read this post again, AMD always stated that you would need a BIOS update for backwards compatibility.
There was never a "shitshow", only stupid fucks like you that don't know what the fuck they are doing.
Also, you talk about AMD sending kits so retards like you could update their BIOS without leaving their houses as being a bad thing. It's not.

>250W

>doesn't understand what architecture redesigned means
>>doesn't go above 75°c overclocked
>doesn't realize the closest CPU Intel has to offer needs to be cooled with a 1 horse power AC compressor
>complains about thermals when the 7700k runs near 80 on a slight 4.7ghz overclock thanks to shitty IHS plumbing culking cooling the chip.

Attached: 1528769762602-pol.gif (380x290, 2.99M)

>Checkers in the Task Menu

How awesome is that?!

Attached: 1508108137217.png (1400x1000, 986K)

>Game of life in task manager

vs 1500w Xeon - Gold™ 28core 5.0GHz 1 horsepower liquid nitrogen meme CPU

How does your intestines smell because you're pretty far up there to believe any fur coming from Intel.

You are either a dim child or a paid $hill.

Attached: 1518298886815-r9k.jpg (117x125, 3K)

but will 1771 watt cooler be enuff for daily use and internet browsing ?

>user I need a link for my CPU drivers not my chipset you silly goose
What do you think he meant by CPU drivers you raging autist. The chipset drivers allow the CPU to operate properly.

Attached: 1529364677572m.jpg (1024x1022, 111K)

Looks like whatever theyre doing is is excluding virtual cores there. Neat.

Yes, the process was optimized.

No need to be so hostile, dude. Even if this is the case, we already see in the link that the CPU has pretty damn good performance (6400cb), which means that it's either less of a problem than you think, or they've found a workaround that makes it more viable. Maybe to keep the 4-channel memory TR limit, they'll limit the memory controller of each die to one channel. I dunno, man, I'm not an expert on these things, but everything we've seen so far shows really good performance.

Lol amd shills this desperate
Their module "32c" cpu will barely match intels 10core let alone 28c cpu
4.5ghz? Keep on dreaming, amd barely hits 4.3 causing a housefire, what will happen if they go further?

>get cpu
>it works as a room heater
nice deal

Attached: electric-room-heater-250x250.jpg (250x250, 13K)