AMD 64-core EPYC clock speeds leaked - The power of ROME

J

juanrga

2[H]4U
Joined
Feb 22, 2017
Messages
2,804
That is actually WAY better than I expected. I was expecting 2ghz maybe 2.5. with boost. Color me impressed.
 
Rockenrooster said:
^^^^^^^^^^^^^^
Dat base clock...............
Higher than previous 32 core flagship
Click to expand...


But the TDP jumped from 180w to 240w. So not as big an efficiency gain as you're implying.

If you take into account TDP vs core count, the efficiency gains are only 30%.
 
Teenk9 said:
What?! With double the cores?!
Click to expand...


Yes, there is an efficiency improvement, but it's not DOUBLE.

You have double the cores for 3/4 the TDP of the old architecture. It's 1/3 lower power consumption, not 1/2 the power consumption. This was expected, given the process node.

This is the reason I wasn't expecting AMD to launch Zen 2 mainstream with more than 12 cores - it would mean a totally new class of TDP above 105w. But with the 7nm refresh next year, they could find the extra efficiency in the tweaked process.
 
Last edited:
defaultluser said:
Yes, there is an efficiency improvement, but it's not DOUBLE.

You have double the cores for 3/4 the TDP of the old architecture. It's 1/3 lower power consumption, not 1/2 the power consumption. This was expected, given the process node.

This is the reason I wasn't expecting AMD to launch Zen 2 mainstream with more than 12 cores - it would mean a totally new class of TDP above 105w. But with the 7nm refresh next year, they could find the extra efficiency in the tweaked process.
Click to expand...

But the clocks also went up. Expecting same TDP while doubling core count and upping the base clocks is asking way too much. Off the top of your head, what's the last cpu that released like that?
 
defaultluser said:
Yes, there is an efficiency improvement, but it's not DOUBLE.

You have double the cores for 3/4 the TDP of the old architecture. It's 1/3 lower power consumption, not 1/2 the power consumption. This was expected, given the process node.

This is the reason I wasn't expecting AMD to launch Zen 2 mainstream with more than 12 cores - it would mean a totally new class of TDP above 105w. But with the 7nm refresh next year, they could find the extra efficiency in the tweaked process.
Click to expand...

Depends on the clocks.
 
Teenk9 said:
But the clocks also went up. Expecting same TDP while doubling core count and upping the base clocks is asking way too much. Off the top of your head, what's the last cpu that released like that?
Click to expand...

The boost clocks are identical, so base clocks is not as important as you might think, as long as the cooling is adequate, it will never run at base clocks (this is a server we're talking about, with oversized cooling and VRMs). The latest-generation boost system is a lot more aggressive than that.

Also, we're talking 2.35 vs 2.2 base clocks, which is only 5% difference. It's lost in the noise.

Stop making such a big deal out of tiny advancements of paper specs. They're really not that impressive.
 
Last edited:
ryan_975 said:
Depends on the clocks.
Click to expand...


Using the 30% reduction in TDP at the same clocks as Zen+ a a guide, that means 8 cores Zen 2 at the same clocks as 2700x is approximately 80w TDP.

That's 120w for 12 cores at the same clocks speeds, or 160w for 16 cores.

If you bump it the expected 10% faster clocks (small voltage bump required, so it's a 15% change), that jumps to 135w for 12 cores (acceptable) , and 185w for 16 cores. That's getting 16-cores dangerously close to FX 9000-series land.

The power consumption of double the cores is only acceptable at lower clock speeds, and that would never sell for the mainstream platform. AMD is going to have to wait for a tweaked 7nm node next year, or take the 9900K power consumption hit, and have it today.

But of note: the Zen 2 would deliver twice the number of cores at the same power consumption as the 9900k, with almost identical per-core performance. That's pretty embarrassing. Just saying, a whole new line of high power motherboards would be harder for fans to swallow.
 
Last edited:
defaultluser said:
it would mean a totally new class of TDP above 105w.
Click to expand...
You mean like the 9999°K? That thing don't give a fuck about the '125W' or whatever bullshit TDP rating it has when it doubles it routinely under its sanctioned turbo limits. I don't think any of the reviewers do either until AMD increases their TDP rating, then it's literally dead pandas and no icebergs because AMD sucks.

I'd gladly take a 160W TDP for 16 fast cores though. Reckon they'll try keep the 12 core 'within existing realms' and the 16 core as a HEDT-lite/enthusiast part will push higher.
 
defaultluser said:
The boost clocks are identical, so base clocks is not as important as you might think, as long as the cooling is adequate, it will never run at base clocks (this is a server we're talking about, with oversized cooling and VRMs). The latest-generation boost system is a lot more aggressive than that.

Also, we're talking 2.35 vs 2.2 base clocks, which is only 5% difference. It's lost in the noise.

Stop making such a big deal out of tiny advancements of paper specs. They're really not that impressive.
Click to expand...

You are incorrect on all your points because electricity, heat, etc, which all falls under basic physics/ohms law etc. and does not work the way you are implying, as it IS NOT LINEAR. . The only way your argument(s) would be correct, is if you where comparing identical chips, identical core count, and identical technology other than node/die shrinkage. But we are not, we are talking about double the core count with with improved/different technology. That difference alone completely shatters your 1 to 1 LINEAR comparison because physics/electricity/ohms law, resistance, etc, does not work that way.

If nothing changed, and AMD just doubled the core count (no die shrinkage, no technology improvements, nothing) all on the same chip, It would be IMPOSSIBLE to do with with even double the TDP (360w) and keep the same clock speeds as as it is not Linear. In fact, they would have a difficult time Achieving even 50% of the clock speeds (if even that) and keep it under 360w much less being able to keep the heat output under control and manageable because the IF they ran those 64 cores at the same speeds, the power draws, and heat output would multiply exponentially.. What this means is your .05% clock speed increase is exponentially much larger, and your 30% TWD decrease is also exponentially much larger when you realize that it is not a linear comparison.
 
Last edited:
defaultluser said:
The boost clocks are identical, so base clocks is not as important as you might think, as long as the cooling is adequate, it will never run at base clocks (this is a server we're talking about, with oversized cooling and VRMs). The latest-generation boost system is a lot more aggressive than that.

Also, we're talking 2.35 vs 2.2 base clocks, which is only 5% difference. It's lost in the noise.

Stop making such a big deal out of tiny advancements of paper specs. They're really not that impressive.
Click to expand...

So you can't give an example of what you're insisting AMD should do with Zen2. Got it.
 
Lol defaultluser.....
A more accurate comparason would be to measure the power diff between Epyc 7601 and Rome with the same 32 cores and clock config..... Then you have IPC to take into account, then doubled AVX..................
I'll bet it's more than 33% more power efficient...............................................

but then you already got owned by the people above me....
 
Rockenrooster said:
How does your Xeon phi compare to a epyc 7601 in like cinebench or some other widely used software? encoding, etc...
Click to expand...

getting widely used software to work on the phis is pretty difficult by itself. they are also pretty much limited to linux as windows cant handle 272 threads effectively. The most direct comparison I can find is for mining monero where a EPYC 7601 gets ~ 1750h/s and a xeon phi 7250 gets ~2100h/s but this isnt that great of a comperison as monero benifits from a high memory bandwidth and the 16gb MCDRAM on the phis blows ddr4 out of the water.

I did manage to get passmark to atleast run on one of them but it wasnt effectively utilizing the chip. Ill attach the screenshot for that below
iMt4hUo.jpg
 
You must log in or register to reply here.
Back
Top