AMD Ryzen Threadripper Breaks 5.2 GHz + X399 Boards on Display!

[VSG]

AMD did not just announce retail availability on Ryzen Threadripper today, they also had some on-site and arranged for a fun LN2 overclocking event as part of Capsaicin SIGGRAPH 2017. As always, such events are to give day one estimates on the maximum performance potential of the silicon which in turn guides end users and board partners alike on the worst case scenarios as far as power draw and cooling requirements go.

Monstru from Lab501 was kind enough to share a couple of pictures of the actual event with us while AMD followed up with a Cinebench R15 screenshot as seen below. All 16 cores of the Ryzen Threadripper 1950X were overclocked to 5.2 GHz with a x52 multiplier on a standard 100 MHz bus speed. Core VID from CPU-Z is not trustworthy at these temperatures, so presumably it was more in the range of 1.6 V than 1.16 V. They did have DDR4 RAM in quad channel but at the JEDEC base of 2133 MHz to get as high a CPU frequency without the IMC being a factor. The Cinebench R15 score of 4122 cb is very impressive, given the previous high score for a 16-core CPU was 2867 cb, and it took a 28 core CPU to beat this score before. Sure, the days of high core count overclockable CPUs is only coming now but it goes to show where we were before AMD and Intel both decided to go big this generation.



After the break we have some photographs of X399 motherboards from various manufacturers, so be sure to take a look.


AMD had also invited their board partners to showcase some of their motherboards for this platform, and we saw AsRock present two motherboards- the Taichi and the Fatal1ty Professional Gaming. AsRock has in the recent years made the Taichi and Fatal1ty Pro Gaming two sides of the same coin, if you will, and we expect both to share a lot here too. They generally have well designed boards, and if the heatpipes on the VRM heatsinks are any indication the story is the same here.


MSI and Gigabyte were on board as well with a Gaming PRO CARBON AC and AUROS Gaming 7 respectively. These are on the higher end of their motherboard ladder each, so it is good to see board partners come out with feature-rich boards right off the bat.


Rounding off the showcase was ASUS with two boards as well. They presented the PRIME X399-A which should be a more budget friendly option compared to what will be their flagship for this generation- the Zenith Extreme. Of note is the fact that both have a heatpipe from the VRM heatsink to the I/O heatsink, so thermals were definitely a consideration in the design of these boards.

All six boards on display were ATX or E-ATX in form factor, and hopefully we will see some mATX motherboards sooner than later for those who need a workhorse on the go.

View at TechPowerUp Main Site

 

[cdawall]

Give me MATX bah

 

[KarymidoN]

Sh*t look bad for intel, Geez... Those TR4 CPUS are nasty...

 

[ASOT]

Romania

 

[Nuckles56]

I want to know what the score would have been like if they could have gotten the CPU running with 3200MHz RAM as well

 

[Chaitanya]

Give me MATX bah

Highly doubt we will get any M-ATX boards in next 5-6 months, maybe after that.

 

[DeathtoGnomes]

Give me MATX bah

NO!

 

[cadaveca]

I want to know what the score would have been like if they could have gotten the CPU running with 3200MHz RAM as well

Meh. It's LN2, so it doesn't show anything about real-world usage anyway. Will be interesting to see if disabling cores, or maybe a whole chip, allows for better clocking.

 

[Durvelle27]

Interesting,

I wonder how high these can go just on water though

 

[pantherx12]

Interesting,

I wonder how high these can go just on water though

As long as your board can provide the power required and you can keep them cool these thread ripper CPUs should all clock the same as Ryzen.

I.e 3.8-4.1 GHz all cores.

It's literally a bunch of Ryzen CPUs and as it's not a monolithic design max overclocks will be the same on all product lines.

 

[ratirt]

I wonder if AMD's working on Ryzen CPUs to hit higher frequencies. 4.5 Ghz would be fine

 

[Durvelle27]

As long as your board can provide the power required and you can keep them cool these thread ripper CPUs should all clock the same as Ryzen.

I.e 3.8-4.1 GHz all cores.

It's literally a bunch of Ryzen CPUs and as it's not a monolithic design max overclocks will be the same on all product lines.

Got a 360mm custom loop waiting on my TR chip

 

[Vayra86]

As long as your board can provide the power required and you can keep them cool these thread ripper CPUs should all clock the same as Ryzen.

I.e 3.8-4.1 GHz all cores.

It's literally a bunch of Ryzen CPUs and as it's not a monolithic design max overclocks will be the same on all product lines.

Seeing and knowing this I am so very tempted to get a TR just to OC the shit out of it and see how far it goes.

Sixteen cores at 4.0 is just glorious for what it is, who needs a workload.

 

[badboy87]

Romania

Ce bine-i sta la Monstru printre atatia straini. Ma mandresc ca avem persoane cu adevarat entuziaste.

 

[[XC] Oj101]

I want to know what the score would have been like if they could have gotten the CPU running with 3200MHz RAM as well

Memory can go *significantly* higher, but at the cost of core frequency.

 

[pantherx12]

Seeing and knowing this I am so very tempted to get a TR just to OC the shit out of it and see how far it goes.

Sixteen cores at 4.0 is just glorious for what it is, who needs a workload.

I'm in the same boat man, I've not built a desktop for myself in years but thread ripper is that huge jump in power I've been waiting for. 4ghz on all cores oils just be gravy, I remember feeling chuffed when I got my FX 8120 going at 5ghz all cores for a few bench mark runs. This is going to make me giddy with the power ha ha. Certainly going to compile and run all kind of simulations on the thing

 

[buildzoid]

Memory can go *significantly* higher, but at the cost of core frequency.

Nope on LN2 Ryzen's memory controller tends to crap out. Not usually to levels of 2133MHz bad but on early BIOSs and boards 2133 might actually be the limit of what the IMC can deal with when around -190C. Will be intresting to see what it can do given more time but I imagine it will probably top out around 2900-3400 on LN2 depending on the silicon lottery.

 

[[XC] Oj101]

Nope on LN2 Ryzen's memory controller tends to crap out. Not usually to levels of 2133MHz bad but on early BIOSs and boards 2133 might actually be the limit of what the IMC can deal with when around -190C. Will be intresting to see what it can do given more time but I imagine it will probably top out around 2900-3400 on LN2 depending on the silicon lottery.

Re-read what I said

 

[buildzoid]

Re-read what I said

Under LN2 dropping core clock won't make the mem go any faster. Though I guess you meant the indirect trade of running the CPU at positive temps leading to a loss of about 1GHz core clock to get 3000+ mem.

 

[EarthDog]

I wonder if AMD's working on Ryzen CPUs to hit higher frequencies. 4.5 Ghz would be fine

Zen 2...i wouldnt hold my breath for that much more on a refresh.

 

[ratirt]

Zen 2...i wouldnt hold my breath for that much more on a refresh.

Has AMD already said zen2 will be 7nm? ALso it won't be the zen+ made in 7nm but improved 14nm I suppose. Who knows if the zen+ won't reach 4.5 and same socket AM4. That would be great.

 

[[XC] Oj101]

Under LN2 dropping core clock won't make the mem go any faster. Though I guess you meant the indirect trade of running the CPU at positive temps leading to a loss of about 1GHz core clock to get 3000+ mem.

 

[Massman]

Nope on LN2 Ryzen's memory controller tends to crap out. Not usually to levels of 2133MHz bad but on early BIOSs and boards 2133 might actually be the limit of what the IMC can deal with when around -190C. Will be intresting to see what it can do given more time but I imagine it will probably top out around 2900-3400 on LN2 depending on the silicon lottery.

Funny story: it's not the actual temperature that is causing this, but the reported temperature.

You can use SenseMI skew to drop the reported temperature and see the memory become more unstable

 

[cadaveca]

Funny story: it's not the actual temperature that is causing this, but the reported temperature.

You can use SenseMI skew to drop the reported temperature and see the memory become more unstable

Many devices change refresh rates according to temperature, and memory is most definitely one of those devices. Hardly surprising, honestly, that a memory controller might do the same.

 

[buildzoid]

Funny story: it's not the actual temperature that is causing this, but the reported temperature.

You can use SenseMI skew to drop the reported temperature and see the memory become more unstable

Yet another fine example of why power all fancy power management tech needs a nice big easily accessible disable switch.