[Rumor] Zen Can OC To 5GHz on Air

[patrickjp93]

1 minute ago, MageTank said:

Even more so since you are involved in them.

 

I'll show myself out.

[deviant88]

4 hours ago, patrickjp93 said:

http://www.guru3d.com/news-story/amd-ryzen-processors-reach-5-ghz-on-air,2.html

 

Nothing else really to say. If this is true (which buildzoid claims is unlikely), it will likely be NH-D15 territory only and very rarely.

 

  Reveal hidden contents

"I'll just use this comment to say this since it's relevant to me.

I don't have a Zen but I know people who do.

I continue to stand by my comment about 5G not happening on air without too much voltage. This doesn't disprove the statement of ZenOC@Air=5G however I wouldn't get your hopes up because we don't know how many cores or volts actually managed that clock. (my sources never gave me absolute max clock just that it does about X.X for daily use)

For all those wondering about X.X. Well lets just say that I am pleased with 

X.X."

 

the most important things has yet to be answerered and i see no one talking about it, maybe you know

what is the max temps allowed by zen silicon? if its the same shit as faildozer and phenoms at @62C Zen is fucked, i know how much trouble that gave me with my overheatin phenom in the summer even on a mild 3.8ghz, and it was the main reason i switched to i5 4570 instead of bullshitdozer, i sure hope for its own good it can do 90C + like intels and like everyother GPU and high quality silicon, that 62C is too fucking close to room temps in summer, doesnt leave much for the Delta,hence little to 0 OC or even underclockin if you use something like intense constant video processing @ 16Threads @100% it cant sit under 60's not a chance in hell, even at stock

[Vode]

If Zen actually hits 5GHz I'll eat my 5960x with Eggs and some Hot Sauce for Breakfast.

 

Quote me.

 

There is just no way IMO.

[patrickjp93]

1 minute ago, deviant88 said:

the most important things has yet to be answerered and i see no one talking about it, maybe you know

what is the max temps allowed by zen silicon? if its the same shit as faildozer and phenoms at @62C Zen is fucked, i know how much trouble that gave me with my overheatin phenom in the summer even on a mild 3.8ghz, and it was the main reason i switched to i5 4570 instead of bullshitdozer, i sure hope for its own good it can do 90C + like intels and like everyother GPU and high quality silicon, that 62C is too fucking close to room temps in summer, doesnt leave much for the Delta,hence little to 0 OC or even underclockin if you use something like intense constant video processing @ 16Threads @100% it cant sit under 60's not a chance in hell, even at stock

Bulldozer and Vishera can go into the 90s just fine.

[patrickjp93]

1 minute ago, Vode said:

If Zen actually hits 5GHz I'll eat my 5960x with Eggs and some Hot Sauce for Breakfast.

 

Quote me.

 

There is just no way IMO.

You should have qualified that with "on air" b/c now when the LN2 guys do it, I still have this.

[Humbug]

I hope this is true.

 

BUT we already got confirmation that the top end Zen model (8 core-16 thread) will have a base clock of 3.4Ghz with still undecided boost clocks.

 

If they can crank up to 5Ghz on air wouldn't it make sense for AMD to up their stock base clock to around 4GHz? That combined with the strong IPC of zen would make the product a winner...

 

is there a reason why AMD would leave 1.6GHz of OC headroom? Power consumption concerns?

[Bouzoo]

17 minutes ago, patrickjp93 said:

What, actual intelligent conversations are weird on LTT threads?

 

Oh...

We went from AMD to Intel to blowing your horn in LGBT thread to LoTR memes. Now we only need fake Rick Astley links and Giorgio Tsoukalos to tell us about aliens and we have everything. 

You may insert Terry Crews because why not. 

[Vode]

2 minutes ago, patrickjp93 said:

You should have qualified that with "on air" b/c now when the LN2 guys do it, I still have this.

Damn, you're right. *on air

[patrickjp93]

4 minutes ago, Humbug said:

I hope this is true.

 

BUT we already got confirmation that the top end Zen model (8 core-16 thread) will have a base clock of 3.4Ghz with still undecided boost clocks.

 

If they can crank up to 5Ghz on air wouldn't it make sense for AMD to up their stock base clock to around 4GHz? That combined with the strong IPC of zen would make the product a winner...

 

is there a reason why AMD would leave 1.6GHz of OC headroom? Power consumption concerns?

Tuning of GloFo's process could be another, but I'm in the heavy doubters camp.

[patrickjp93]

3 minutes ago, Bouzoo said:

We went from AMD to Intel to blowing your horn in LGBT thread to LoTR memes. Now we only need fake Rick Astley links and Giorgio Tsoukalos to tell us about aliens and we have everything. 

You may insert Terry Crews because why not. 

[Bouzoo]

5 minutes ago, patrickjp93 said:

 

It's weird. 

[Lord Nicoll]

53 minutes ago, patrickjp93 said:

Piledriver was a very loose-time (low IPC) architecture on a very leaky/power-inefficient node(32nm SOI bulk planar). Intel's own Sandy Bridge processors can reach 5+ with a fair amount of frequency. As timings get tighter for higher IPC, higher clocks become more and more difficult (the actual circuits on the CPU fire much faster than the CPU's actual shown clock speed), and FinFET also hampered Intel's upper clock speed ceilings. To expect it to not have a similar effect on AMD is a bit ridiculous imho.

I'm fully aware how the CPU works, I had to spend a month learning and then realised it had nothing to do with what I wanted to do. Obviously as we shrink the accuracy of the cuts and circuits, it will get harder to overclock but average clocks and yields get better as we're able to make them better, I'm sure if either one of the companies really wanted they'd be able to make a massively power architecture that has very high speeds and IPCs, but they wouldn't make any money on it from the massive cost and they'd probably price it well over say, Intel top of the range CPU E7-8890 v4 at $7174.00 for just one. 

[patrickjp93]

2 minutes ago, Lord Nicoll said:

I'm fully aware how the CPU works, I had to spend a month learning and then realised it had nothing to do with what I wanted to do. Obviously as we shrink the accuracy of the cuts and circuits, it will get harder to overclock but average clocks and yields get better as we're able to make them better, I'm sure if either one of the companies really wanted they'd be able to make a massively power architecture that has very high speeds and IPCs, but they wouldn't make any money on it from the massive cost and they'd probably price it well over say, Intel top of the range CPU E7-8890 v4 at $7174.00 for just one. 

Cough *IBM Power8* Cough

 

Also, the 8890A is 200MHz faster and is $7600.

[Lord Nicoll]

5 minutes ago, patrickjp93 said:

Cough *IBM Power8* Cough

 

Also, the 8890A is 200MHz faster and is $7600.

IBM's power8 is pretty shit being RISC. Intel's E7 8890 V4 in a server with 8 CPUs is basically 24*8*2 for logical processors, of glorious Intel architecture, with in the region of like 24 TB of RAM on some supported systems. I didn't find any mention of an 8890A, so either that's something else or there's missing info.

[patrickjp93]

6 minutes ago, Lord Nicoll said:

IBM's power8 is pretty shit being RISC. Intel's E7 8890 V4 in a server with 8 CPUs is basically 24*8*2 for logical processors, of glorious Intel architecture, with in the region of like 24 TB of RAM on some supported systems. I didn't find any mention of an 8890A, so either that's something else or there's missing info.

Uh, not in some cases. The SAP benchmarks are the big place IBM wins hands down. The many threads per core, with high clock speed, and fewer cores has actually produced a superior analytics architecture, particularly because of the reduced overhead for cache coherency.

 

Sorry, was thinking of the 2699A. MY mistake.

[Lord Nicoll]

1 minute ago, patrickjp93 said:

Uh, not in some cases. The SAP benchmarks are the big place IBM wins hands down. The many threads per core, with high clock speed, and fewer cores has actually produced a superior analytics architecture, particularly because of the reduced overhead for cache coherency.

 

Sorry, was thinking of the 2699. MY mistake.

Oh I thought you meant one of the Power PC ones. Thing is broadwelL EP is clock for clock more powerful then Power8, and a lot more diverse and useful, I use two ancient Xeon X5670's for ESXi and, I don't think there is a cheaper was to virtualise so many VMs as I do, and you just can't do that on Power8, in enterprise terms it blows, and when compared to super computer uses, it still gets destroyed by the vastly more powerful X86_64 nature, especially when you've got a couple million logical processors anyways.

[SSL]

2 hours ago, patrickjp93 said:

Yes I blow my own horn from time to time

 

That's the understatement of the century.

[patrickjp93]

6 minutes ago, Lord Nicoll said:

Oh I thought you meant one of the Power PC ones. Thing is broadwelL EP is clock for clock more powerful then Power8, and a lot more diverse and useful, I use two ancient Xeon X5670's for ESXi and, I don't think there is a cheaper was to virtualise so many VMs as I do, and you just can't do that on Power8, in enterprise terms it blows, and when compared to super computer uses, it still gets destroyed by the vastly more powerful X86_64 nature, especially when you've got a couple million logical processors anyways.

I meant PowerPC Power 8 specifically still kicks Broadwell's behind in scale-up workloads. Anandtech has the benchmark scores somewhere.

 

And no, Power8 is not that far behind in IPC.

 

Is that why the most powerful DOE supercomputer is Power 9 and Volta based instead of Skylake and Volta based?

 

http://www.anandtech.com/show/10435/assessing-ibms-power8-part-1/10

 

Yup, Broadwell is destroyed.

[Prysin]

3 hours ago, patrickjp93 said:

It's not an AMD announcement.

no this is a Patrick announcement

[patrickjp93]

1 minute ago, Prysin said:

no this is a Patrick announcement

No, my announcement is this:

 

Bow before your new moderator (as of tomorrow).

[Lord Nicoll]

1 minute ago, patrickjp93 said:

I meant PowerPC Power 8 specifically still kicks Broadwell's behind in scale-up workloads. Anandtech has the benchmark scores somewhere.

 

And no, Power8 is not that far behind in IPC.

 

Is that why the most powerful DOE supercomputer is Power 9 and Volta based instead of Skylake and Volta based?

The Sunway TaihuLight, being the most powerful, is build using sunfire, but mostly because US government blocked China's attempts to buy more Xeon CPUs for an upgrade to Tianhe-2 and to build TaihuLight, so they had no choice but to design, and in the case of this one, license Sunfires systems and use RISC instead. Tianhe-2 has 768,000 Logical Intel processors and 2,736,000 Xeon Phi cores, conversely TaihuLight has 10,649,600 RISC cores, and is only 3x as powerful. If Tianhe-2 had 10,649,600 logical processors it would smash past TaihuLight, but powerdraw, cooling, size and cost would prohibit that. 

[patrickjp93]

1 minute ago, Lord Nicoll said:

The Sunway TaihuLight, being the most powerful, is build using sunfire, but mostly because US government blocked China's attempts to buy more Xeon CPUs for an upgrade to Tianhe-2 and to build TaihuLight, so they had no choice but to design, and in the case of this one, license Sunfires systems and use RISC instead. Tianhe-2 has 768,000 Logical Intel processors and 2,736,000 Xeon Phi cores, conversely TaihuLight has 10,649,600 RISC cores, and is only 3x as powerful. If Tianhe-2 had 10,649,600 logical processors it would smash past TaihuLight, but powerdraw, cooling, size and cost would prohibit that. 

And the DOE's planned for this upcoming year is far more powerful than TaihuLight.

[Lord Nicoll]

Just now, patrickjp93 said:

And the DOE's planned for this upcoming year is far more powerful than TaihuLight.

That is unlikely as the sheer size and cost of being far more powerful is, well, impractical. 

[patrickjp93]

1 minute ago, Lord Nicoll said:

That is unlikely as the sheer size and cost of being far more powerful is, well, impractical. 

No it isn't.

 

http://www.nvidia.com/object/exascale-supercomputing.html

 

300 Petaflops at just 10 Megawatts, using LESS power than TaihuLight.

[Lord Nicoll]

1 minute ago, patrickjp93 said:

No it isn't.

 

http://www.nvidia.com/object/exascale-supercomputing.html

 

300 Petaflops at just 10 Megawatts, using LESS power than TaihuLight.

That's because it's GPU accelerated, which I don't think will be running off RISC machine due to PCI-e and compatibility, I'm guessing it will be using Xeons for that reason.