AMD Posts EPYC Benchmarks, Teases Vega Frontier Edition

Not open for further replies.


Well, their benchmark for Rysen were true...

Trolls (BUT LOOK AT THE PERFORMANCE AT 1080P WITH A 1080 GTX ROLFCOPTER)... sure, a totally unpractical and unprobable bench...

1. You don't bundle a 500$ CPU with a 1080 GTX to play at 1080p
2. You don't bundle a 1600x to play at 1080p with a 1080 GTX

Thanks Toms for generating people like XFRGTR.


Jan 31, 2009
You say "That leaves the margin-rich data center as AMD's final target. Intel and Nvidia are both titans in the data center, with Intel holding 95%+ of the global server CPU sockets and Nvidia holding the lion's share of the burgeoning AI segment with its lineup of GPUs."

and then:

"...gain share in the data center, especially in light of the conservative nature of risk-averse data center operators who are slow to adopt new architectures."

I would think yes and no. Are google and amazon really such pussys? Datacenters have the skillsets for workarounds. Cheaper means more affordable redundancy, & hence, safer....

Whatever, objective comparisons with epyc are very embarassing for intel atm.

But the MCM/fabric story has just started.

True to form of lucrative pro models first, i expect vega GPU inclusive variants of the epyc MCM before raven ridge APUs release (they are variants of the same theme), and all processors inter connected via the 128 lane fabric (? I asume, as thats the interconnect for the 2 socket epyc servers) , with absurd speed and efficiency, independent of the legacy beholden system board, and its busy shared bus.

an exciting image IMO:

it grapically shows, as we know from vega specs, a path for gpuS to be linked w/ cpuS on the same fabric.

The fabric can even mount its own ssd raid 0 arrays for onboard storage or virtual memory, and on fabric ~500GBps hbm2 ram, as per vega specs.

Clearly, using storage as virtual memory, aint what it used to be, given the above scenario, and amd make much of their advanced memory pooling, prioritising and layering technology, and tho inevitably slower than vram, it allows ~unlimited (512TB) virtual ram and vram. NB also, its concurrently adding a chunk of radically fast vram to the hierarchical memory resource pool

Its a wet dream for cgpu server guys.

The big picture is that an inactive player in the ~server market, has come from nowhere and trounced the ~100% share incumbents metrics. The minnow cant digest that 100% share anyway, but they can sure prosper on over 30% of intels river of gold, and that seems very doable.

John Wittenberg

Mar 9, 2014
"1. You don't bundle a 500$ CPU with a 1080 GTX to play at 1080p"

Except when the best video card on the market can't play that shiny new game at 4K with the eyecandy turned up and not dip below 40FPS (I can't stand less than 55, and my TV doesn't have Gsync ya know). Cough, Ghost Recon Woodlands, cough.

So yeah, my $700 (used) E5-1660 V3 is paired with my $700 1080 Ti to play some games at 1080P.

Totally realistic scenario, champ.
Not open for further replies.