Alleged AMD EPYC 7003 128-Core 2P Zen 3 System Cranks Insane Cinebench R23 Score

Then live with it and stop complaining. Even changes in chip arch like SSE2 ect can do the same thing.

I'm not complaining and I'll keep running Intel. I haven't had to turn off a VM to move it since before I brought up my ESX 3.5 clusters and there's no reason to change now.
 
Not only that, a full-fledged server/workstation product where TDP and noise doesn't really matter
On that I am not sure how much it is true, maybe thermal but Epyc type of solution seem to be a big part about a thermal and efficiency challenge (I imagine why they run so much slower than a 5800x hz wise and do not come close to fully scale).

A Mac Mini consume about 20 watt at full load it seem (close to a Ryzen 4700U full system that is powering a display at the same time, so imagine removing the screen they would look really close):
https://cdn.arstechnica.net/wp-cont...ple-M1-Mac-Mini.power-consumption-640x480.png

And peak at 30 watt.

An 64 core Epyc cpu seem to peak under 350 watt (for 8 or 16 time the core depending how the 4+4 play out power wise), the new 128 core mylan is announced with the same TDP, i.e. could end up below 16 time the peak consumption of an M1 system even with the ram and motherboard on and not too different of 16 time the average during heavy load.
 
I'm not complaining and I'll keep running Intel. I haven't had to turn off a VM to move it since before I brought up my ESX 3.5 clusters and there's no reason to change now.
I just tried it and of the 3 I tested only one would boot after, the others tossed all sorts of errors.

Edit:
All 3 are now booting but my old ass Accounting software is throwing a hissy fit. It may have to live on Intel hardware but if it's the only one that is manageable, sent off info requests to the support company for it to see what they recommend in this case.
 
Last edited:
Very happy with our Rome cluster - 9 x Dell R6515, just under 10K each.
1607969249424.png

No issues with migrations between Intel & AMD here, just power down the VM, vmotion it, boot it back up (no steps in-between, esx6.7u3). Literally a second or two plus shutdown/bootup time.
 
On that I am not sure how much it is true, maybe thermal but Epyc type of solution seem to be a big part about a thermal and efficiency challenge (I imagine why they run so much slower than a 5800x hz wise and do not come close to fully scale).

A Mac Mini consume about 20 watt at full load it seem (close to a Ryzen 4700U full system that is powering a display at the same time, so imagine removing the screen they would look really close):
https://cdn.arstechnica.net/wp-cont...ple-M1-Mac-Mini.power-consumption-640x480.png

And peak at 30 watt.

An 64 core Epyc cpu seem to peak under 350 watt (for 8 or 16 time the core depending how the 4+4 play out power wise), the new 128 core mylan is announced with the same TDP, i.e. could end up below 16 time the peak consumption of an M1 system even with the ram and motherboard on and not too different of 16 time the average during heavy load.

The 128 cores is a 2 processor system, each with 64 cores
 
So your comments really serve no purpose other than to let everyone know that you don't shut your vms down and your not changing anything. Got it. Thanks. :rolleyes:

No, the point is you don't shut down VMs when you can just vMotion things. To suggest having to do so when you need to move a VM to different host doesn't work in most environments where you need to put change controls in for outages. If you're doing a greenfield deployment, sure, go Ryzen but other than that, you're creating a world of headaches for yourself that you can easily avoid just by sticking with the same architecture.
 
I don't do this crap but I know there are some CPU-ID stuff you can do to mitigate problems during movement of VMs up and down the compatibility tree. It really sounds like you're just lazy or unwilling learn.
 
I don't do this crap but I know there are some CPU-ID stuff you can do to mitigate problems during movement of VMs up and down the compatibility tree. It really sounds like you're just lazy or unwilling learn.
That's the whole point, when you enable EVC, you enable it for either Intel Hosts or AMD Hosts and then you set the mode once you've selected one of the two.
 
The 128 cores is a 2 processor system, each with 64 cores
Make more sense, I would still expect the watt by core to be very low on that system under load around 4 to 6 would not surprise me (2 to 3 by thread), dual epyc socket system seem to often achieve under 500 watt with heavy load), not in a different world than an M1 mac mini.
 
Last edited:
AMD is the only megacorp keeping x86-64 alive, and with the continued results I am seeing, and if they can keep up this momentum, then I retract my statements about x86-64 being all but dead within a decade.
Don't think you'll need to retract your statement. Not IMO.

The projected Apple 1Mx 32 core chip would likely yield a Cinebench 23 score of around 90,000, besting the score above. It would be cheaper to produce and use less energy. This is what is expected for the next MacPro. If Apple builds a server machine...it will be very, very interesting.
 
Last edited:
Its still cores though, not threads. Using the bigger cores would make the die even bigger and power usage would go up some.
adding 4 small cores take waaaaaay more die space than adding SMT.
4+4c vs 8c is more of a valid comparison than 4+4 vs 4c/8T any day...

No one here talked about space. SMT is more efficient regarding die space, but big-little is more efficient regarding power consumption.

We are here talking about performance. 4 big + 4 small cores is similar to 4C+SMT regarding performance. The claim that EPIC is "16 times the cores" is misleading if you want to compare the performance of each core.
 
Mega6 it is not complaining, it was explaining exactly why they can not just up and migrate to AMD based nodes. And a legit concern for AMD if they want to enter to the server market smoothly. A major point of running virtualized is to avoid any downtime, when companies spend millions on clusters, and are then told 3 years later, sorry, we have to shut down every VM through a rolling outage to migrate to our shiny new fast AMD systems..it nevers goes over well.
 
M1 was a poor choice to compare with. It over preforms as a 4+4 and under performs as an 8. It is possibly the furthest point of reference sans a mid range quad core. A 4800u scores ~10k which is ~8x the ~80k+ result. (cinebench nice scales)

But I can see why the latest buzz needs to be compared to the latest buzz.
 
  • Like
Reactions: kac77
like this
An
No one here talked about space. SMT is more efficient regarding die space, but big-little is more efficient regarding power consumption.

We are here talking about performance. 4 big + 4 small cores is similar to 4C+SMT regarding performance. The claim that EPIC is "16 times the cores" is misleading if you want to compare the performance of each core.
No it's not.
Cores, big or small are still cores...
SMT is not a core.
It would only be a problem if it couldn't use all 8 cores at the same time, but since it can, it actually is a valid comparison, you just have to note the big.little config.
Not sure where you got your reasoning, you are smarter than this...
Now if you wanted a more interesting comparison I would do the 4 big vs a 4 core Ryzen (with SMT).
I mean it's not AMDs fault they did a 4+4 config. Soon enough they'll have a 8c config too.
 
Last edited:
SMT is not a core.

SMT adds logical cores.

It would only be a problem if it couldn't use all 8 cores at the same time, but since it can, it actually is a valid comparison, you just have to note the big.little config.

Not sure where you got your reasoning, you are smarter than this...
Now if you wanted a more interesting comparison I would do the 4 big vs a 4 core Ryzen (with SMT).
I mean it's not AMDs fault they did a 4+4 config. Soon enough they'll have a 8c config too.

Once more and just moving away from this discussion. 4C+4C is a better representation than 8C when discussing performance, because the M1 is a heterogeneous design.

No one here said anything about being AMD fault. The fault is on Hothardware for the technical reasons explained above.
 
Back
Top