Intel launches 28 core CPU , AMD says "Hold my beer" and launches 32 Core.

SirFooK'nG

Executive Member
Joined
Feb 13, 2009
Messages
8,138
How many applications can utilize all these cores? Remember when quad came out...
 

cguy

Executive Member
Joined
Jan 2, 2013
Messages
6,018
Hmmm.. Intel launched their 28 core about a year ago. It's also an actual 28 core die, and not 4 separate ones in the same package. Also, it has just about half the flops of the Intel, so meh... AMD... meh...
 

Zyraz

Expert Member
Joined
Jul 7, 2011
Messages
2,655
Last edited:

cguy

Executive Member
Joined
Jan 2, 2013
Messages
6,018
I might be wrong but as far as I know, you are wrong .
Intel only recently announced 28 core CPU.
The Xeon version has been around since July 17 (see Skylake Wiki).
 

Ancalagon

Honorary Master
Joined
Feb 23, 2010
Messages
17,020
The Xeon version has been around since July 17 (see Skylake Wiki).
This will be a lot cheaper. I doubt it performs half as well as the Xeon - I'd say it should be similar, depending on the tasks. AMD's IPC is only a little behind that of Intel. So in this case, more cores but at slightly lower clockspeeds.

I've seen benchmarks that show that AMD's chips make very cost efficient database servers.

For instance, Anandtech did a detailed review of Epyc last year that showed, dollar for dollar, it simply cannot be beaten.
 

ponder

Honorary Master
Joined
Jan 22, 2005
Messages
83,896
Hmmm.. Intel launched their 28 core about a year ago. It's also an actual 28 core die, and not 4 separate ones in the same package. Also, it has just about half the flops of the Intel, so meh... AMD... meh...
The Intel 28 core from 2017 was a Xeon server cpu, amd launched their 32 core Epyc server cpu around the same time. The 32 core Epyc processors cost about 50% of the 28 core Xeon price...

The latest 32 Core Ryzen is not a server CPU but aimed at desktop/workstation machines and I suspect it will cost substantially less than the i7-8086K.

Don't have any realword benchmarks for servers to compare cost vs performance, only seen things like cinebench etc where amd reaches 85%+ of the intel.
 

CataclysmZA

Expert Member
Joined
Apr 1, 2010
Messages
4,798
Hmmm.. Intel launched their 28 core about a year ago. It's also an actual 28 core die, and not 4 separate ones in the same package. Also, it has just about half the flops of the Intel, so meh... AMD... meh...
FLOP rates aren't really useful for evaluating CPU performance these days. However...

https://www.techpowerup.com/cpudb/1929/epyc-7551p

https://www.techpowerup.com/cpudb/2055/xeon-platinum-8180

EPYC 7551P with all-core boost of 2.55GHz and 8 DP ops/cycle yields... 652 GFLOPS

Intel Platinum 8180 with an all-core boost of 3.3GHz and 16 DP ops/cycle yields... 1.433 TFLOPS.
You are correct. I have the sneaking suspicion that AMD has raised the all-core boost on Threadripper/EPYC 2 substantially, so we'll probably see all-core boosts of up to 3.5GHz with this generation, in which case that brings a 32-core Threadripper up to 896 GFLOPS.
 
Last edited:

cguy

Executive Member
Joined
Jan 2, 2013
Messages
6,018
This will be a lot cheaper. I doubt it performs half as well as the Xeon - I'd say it should be similar, depending on the tasks. AMD's IPC is only a little behind that of Intel. So in this case, more cores but at slightly lower clockspeeds.

I've seen benchmarks that show that AMD's chips make very cost efficient database servers.

For instance, Anandtech did a detailed review of Epyc last year that showed, dollar for dollar, it simply cannot be beaten.
The Intel 28 core from 2017 was a Xeon server cpu, amd launched their 32 core Epyc server cpu around the same time. The 32 core Epyc processors cost about 50% of the 28 core Xeon price...

The latest 32 Core Ryzen is not a server CPU but aimed at desktop/workstation machines and I suspect it will cost substantially less than the i7-8086K.

Don't have any realword benchmarks for servers to compare cost vs performance, only seen things like cinebench etc where amd reaches 85%+ of the intel.
It really depends on one's metrics and applications. The Epyc is much flakier in terms of performance than the Skylake. A bunch of our code ran horribly on it, because of the 4 chip design and lack of flops. A DB typically isn't limited by FLOPS and doesn't need tight intercore communication latencies, so it's probably ideal for the Epyc.

Skylake does have a few regressions from previous architectures, but doesn't seem to have these sharp edges. Also, the $ amount is often more complex than just the chip when talking about work stations or data-centers - requiring 2x more machines and/or rack space to get the same performance (or even somewhat better) is much more expensive than the chip price difference.
 

Ancalagon

Honorary Master
Joined
Feb 23, 2010
Messages
17,020
It really depends on one's metrics and applications. The Epyc is much flakier in terms of performance than the Skylake. A bunch of our code ran horribly on it, because of the 4 chip design and lack of flops. A DB typically isn't limited by FLOPS and doesn't need tight intercore communication latencies, so it's probably ideal for the Epyc.

Skylake does have a few regressions from previous architectures, but doesn't seem to have these sharp edges. Also, the $ amount is often more complex than just the chip when talking about work stations or data-centers - requiring 2x more machines and/or rack space to get the same performance (or even somewhat better) is much more expensive than the chip price difference.
Could it be because you simply hadn't optimized your code for it? The benchmarks I have seen do have more variance than the Intel chips, but none of them were downright terrible.
 

CataclysmZA

Expert Member
Joined
Apr 1, 2010
Messages
4,798
It really depends on one's metrics and applications. The Epyc is much flakier in terms of performance than the Skylake. A bunch of our code ran horribly on it, because of the 4 chip design and lack of flops. A DB typically isn't limited by FLOPS and doesn't need tight intercore communication latencies, so it's probably ideal for the Epyc.
I certainly don't see EPYC being used for machine learning stuff just yet, Intel and NVIDIA certainly have a lead there.
 

cguy

Executive Member
Joined
Jan 2, 2013
Messages
6,018
I certainly don't see EPYC being used for machine learning stuff just yet, Intel and NVIDIA certainly have a lead there.
Yeah, agreed. That's one of the cases it has an issue (due to FLOPS), but the other case is anything that requires tight chip-to-chip communication. You have 4 groups of 8 cores. The interconnect between these chips is off-die and much slower (both bandwidth and latency). This makes little different if every thread is independent, but if they share data and/or coordinate execution, performance falls through the floor.
 

crackersa

Honorary Master
Joined
May 31, 2011
Messages
29,032
Intel launches 28 core CPU , AMD says "Hold my beer" and launches 32 Core.

Awesome headline (I’m an intel fan too)
 
Last edited:
Top