AMD's EPYC CPU'S with Benchmarks & Pricing!

[TechFanboy]

4 minutes ago, The Benjamins said:

Man in this live blog they are making Intel Xeons look so bad lol.

HAHAH! Not just bad.. but Insanely EXPENSIVE! HAHAHA 

[cj09beira]

2.7ghz not bad

[VerticalDiscussions]

They support ECC memory right?

 

I havent seen an y reference to that yet.

[cj09beira]

Just now, VerticalDiscussions said:

They support ECC memory right?

 

I havent seen an y reference to that yet.

all ryzen cpus support ecc memory 

[TechFanboy]

Just now, VerticalDiscussions said:

They support ECC memory right?

 

I havent seen an y reference to that yet.

I am sure they will support it.. Just wait for the announcement.  

[The Benjamins]

Time to order my 64c128t 128GB ram 1050ti RX 550 gaming rig.

 

EDIT: sorry fixed the GPU

[ravenshrike]

Just now, The Benjamins said:

Time to order my 64c128t 128GB ram 1050ti gaming rig.

I wasn't aware anyone had made a copy of Dwarf Fortress that scales with cores available.

[NumLock21]

1 hour ago, PCGuy_5960 said:

>Server CPUs

>Best for your wallet

Wut? Who buys 32 core CPUs for desktop use?

ME!

 

[TechFanboy]

Just now, The Benjamins said:

Time to order my 64c128t 128GB ram 1050ti gaming rig.

HAHAHA! I am pretty sure once they've fixed the virtualization issues with AMD.. Linus will redo the 7 Gamers 1 CPU.  

[cj09beira]

Just now, TechFanboy said:

HAHAHA! I am pretty sure once they've fixed the virtualization issues with AMD.. Linus will redo the 7 Gamers 1 CPU.  

its already fixed in agesa 1.0.6

[TechFanboy]

Just now, cj09beira said:

its already fixed in agesa 1.0.6

Oh Cool! But what about Unraid????  

[AnonymousGuy]

Conspicuously neglecting to include any latency numbers.

 

Also looks like AMD is fucking with the benchmark numbers:

 

Quote

It bears mentioning that Intel publicly posts its SPEC benchmark data, and AMD's endnotes indicates that it reduced the scores used for these calculations by 46%. AMD justified this adjustment because they feel the Intel C++ compiler provides an unfair advantage in the benchmark. There is a notable advantage to the compiler, but most predict it is in the 20% range, so AMD's adjustments appear aggressive.

Oh gee let's just exclude a MASSIVE reason to use IA in the first place (software and compilers being superior) and handwave it away.

[pyrojoe34]

1 hour ago, SC2Mitch said:

Oh, my point still stands, does any manfuacture do that kind of nuts amount lol and what do you need 2tb of DDR4 memory for, Chrome Machines?

A lot of computational work, machine learning, neural networks, and deep learning applications will chew through that data fast. The biggest bottleneck with running massive datasets is data availability (speed) and being able to read and process the data at speeds upwards of 60GB/s and latencies orders of magnitude lower than drives can massively cut down run times and allow for larger datasets to be used without severely bottlenecking the algorithms. A while back I wrote a simple machine learning algorithm that ran through some pretty extensive datasets (250k+ subjects with 40 variables per subject) that chewed through the server's 256GB of RAM in no time. I had to start caching temp data to the drives which slowed down the process substantially. My problem was almost certainly poorly optimized code (e.g. multiple copies of the same data stored across the RAM during merge functions) but when you're talking about people making deep learning applications that process terabytes of data you don't want the processor waiting around for the next data block to load from drives at a measly 500MB/s (even a PCIe drive is a slow 3GB/s compared to RAM).

[DXMember]

12 minutes ago, AnonymousGuy said:

Conspicuously neglecting to include any latency numbers.

 

Also looks like AMD is fucking with the benchmark numbers:

 

Oh gee let's just exclude a MASSIVE reason to use IA in the first place (software and compilers being superior) and handwave it away.

using benchmark results but then adjusting them due to compiler optimizations not being fair???

ridiculous.... unlocking a new era in marking

[DXMember]

1 minute ago, pyrojoe34 said:

My problem was almost certainly poorly optimized code (e.g. multiple copies of the same data stored across the RAM during merge functions)

Hardware is cheaper than dev time, MORE CORES !!!

[NumLock21]

 

NVMe as well as SAS up to 12Gb/s

[DXMember]

4 minutes ago, NumLock21 said:

 

NVMe as well as SAS up to 12Gb/s

24 NVMe devices... image the likes of the databases you could run on that

[grimreeper132]

these CPUs look like beasts I can see them going down well with companies, pity though that they dinny have quad or octa socket  support as that would of been awesome (yes most companies couldn't use it but the ones who could would most certainly use it)

[grimreeper132]

2 hours ago, SC2Mitch said:

2TB of memory rofl, does manufacture even sell upto that amount in 8 sticks?

I think it's 16 sticks, like how dual channel has 4 slots, so it would be 32 sticks if your on a dual socket motherboard and that's definitely possible

[cj09beira]

32 cores for 3400 dollars, Wow

 

[cj09beira]

32 core for 2100 bucks if you dont need 2 socket support WOW

[NumLock21]

Anand's Live Blog

Quote

• EPYC 7601 offers 73% perf per dollar over E5-2699v4
• 24-core from $1850, 16-core from $650, 8-core from $475
• Single socket pricing: 16-core at $750
• So there's your ThreadRipper pricing, probably

 

 

[JoeCoke]

Am I understanding this right?

Intel: 9000$ for 24 cores @ 2.4ghz

AMD: 2000ish$ for 32 cores @ 2.7ghz

 

WOW. If that's right, then AMD really came through. Like, REALLY came through.

[cj09beira]

GPU TIME :-)

[cj09beira]

12.3 Teraflops fp32

24.6 fp16

1500mhz core clore confirmed