AMD's Future Chips & SoC's: News, Info & Rumours.

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Infinity Fabric is just another way to interconnect stuff; they could be using it as a parallel connection scheme to PCIe lanes. IBM with nVidia are doing that with nVLink. They don't use PCIe to communicate stuff they need to have moving around "fast". It would also be interesting to see if AMD is actually going to use "IF" for consumer products, since nVidia will bring nVLink.

Cheers!
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985


Thanks for that mate...
Another thing that leaps to mind is that overclocking the RAM will increase the speed the Fabric talks to everything in the system including the GPU, this is another reason to use the fastest RAM that our budget allows..
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965


I was thinking more along the lines of AVX. Ryzen demonstrated poor performance compared to Intel with these instructions.
 


The issue is there's a lot of cases where PCI-E is a bottleneck; this is a noticeable problem in multi-GPU configurations and why both NVIDIA and AMD rely on bridges for card-to-card communication, rather then the PCI-E bus.

16GB/Sec bandwidth sounds like a lot...but not for cards that ideally want to work on a .16 second interval with 8GB VRAM on die to play with. Suddenly, 16GB/Sec seems downright slow when you consider the amount of data GPUs want to play with.

I'm not saying go the AGP route and make a dedicated high bandwidth capable bus just for GPUs...but we're going down the road where we need something an order of magnitude faster then PCI-E, especially considering how compute is fast becoming a thing.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
@goldstone77
Makes lots of sense alright... hopefully this is not just implemented in the server market.

@gamerk316
Would be nice to have a faster interconnect especially when it comes to communication with the graphics card... There's rumour's of a dual GPU Vega... not sure if we would have the bandwidth for a consumer version.
 

0ldsch00l

Notable
May 9, 2017
288
0
810
The question is and yes all new game engine AAA titles can take advantage of 16 cores, question is WHO NEEDS IT FOR GAMING? Same thing was said in 04 with dual cores who needs them, back then that was blue sides claim to fame to discredit AMD, now more and more is indeed multithreaded, but if games and yes if AMD want to market their 16 core for games then I will talk about games, if they are barely taking advantage of an 8 core vishera, tell me why we need 16 cores? Unless well all get 5k together and all of us build mnster rigs with 4 X dual gpu vega cards for VSR 5k rez with 8 x AA, what need is there for 16 cores? A 1700 heck a 6 core Ryzen will not bottleneck any setup no matter how many monster cards.

Now if I were AMD Id market the 16 core to users more then little gamers, not server market, but intense CPU use, audocad and all that, maybe forensics experts for CPU cracking tools?
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985


Hey 0ldsch00l how are you mate... Welcome to tomshardware btw !!

The 16 core is marketed as a high end desktop part... so more or less as you say...
It's aimed at the "enthusiast consumer space" so I'm sure that will include professionals and lots of "full nerds"... that love having beastly hardware... :)

I have to say I find it quite interesting myself... I can't help wondering about pricing..(an about a 12 core)..
Also the x390 platform supports quad channel memory... nice..

It's pretty likely there will be a 12 core / 24 thread part released for x399... it would be very interesting to see what the pricing is like.. hopefully not too expensive... The specs look an sound savage.
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
Maybe they want compete with Intel on every level to offer true competition. That would be my guess, because of the R3,R5, and R7 lines do exactly that. 12 and 16 core would with 44 pci-e lines and quad channel are definitely good for content creation professionals.
 

0ldsch00l

Notable
May 9, 2017
288
0
810

Wait X399 isnt Am4? Can you put say a Ryzen 1400 in this chipset?!?!?!

Im assuming its same socket just better VRM then 370 for more power to a 16core....

Hey dont get me wrong I used to be into beastly hardware and will build a decent Ryzen, but all that epeen stuff uhh Im not 20 anymore.... and hey Im not a nerd I guess Im a dweeb :D

I mean for bragging rights and yeah future proofing but I just dont see any gaming system needing 16 cores for at least 3 years.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
@oldschool
No it's a different socket mate..
AM44 or AM8 we don't exactly know yet.. It will be a bigger slot I guess with more CCX's on the die taking up real estate.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
AMD Raven Ridge 3.0 GHz APU Engineering Sample With Quad Zen Cores Spotted – Features 704 Vega NCUs Clocked at 800 MHz

Check it out Raven Ridge engineering sample spotted:
http://wccftech.com/amd-raven-ridge-ryzen-apu-vega-gpu-leak/
 

0ldsch00l

Notable
May 9, 2017
288
0
810


Good ok that will be my spinal cord then for next system, I always get monster board even if its a say lowest tier 8300 when there will only visheras, I must have had 20 visheras in the past 2 years, lowest tiers happened to. despite being labled bottom bin 95W, OCed better then 8350s even, 8300s were golden. I think I put a 1700 in a X399 :D
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985


It's not really by choice.. a 16c Ryzen would have four CCX's with 4 core in each one.. The 8 core has 2 CCX's.. It will be just physically bigger end of story... More cores, more real estate.. so bigger socket.
 

0ldsch00l

Notable
May 9, 2017
288
0
810

VEGA!!!!!!!
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
3483rj6.png

http://wccftech.com/sk-hynix-lists-hbm2-gddr6-amds-vega-nvidias-volta-latest-memory-databook/
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
AFAIK it was supposed to be 256Gb/s per 4Gb stack.. but it appears (if the leak is right) their going with the slower stuff for now anyway.. Looks like Sk Hynix is letting them down with the 2.0 Gb/s HBM2... Shame, it's gonna affect benchmarks, hopefully it wont be too long before Vega can be paired with the faster HBM 2.

 


I'm starting to get the sense that the lack of orders for HBM/HBM2 is going to be it's demise (less orders = higher cost = less orders), especially since it looks like GDDR6 is on the horizon.
 


Let the best [strike]man[/strike] spec win! :p

I don't really mind; both specs are well suited for their specific target needs. Cheap usually triumphs technically better most of the time though.

Cheers!
 


Historically, the cheaper option that addresses immediate concerns wins over superior technical solutions.

My manager has a quote, which is true when you think about it:

"Faster, Cheaper, Better; pick two out of three" [Faster refers to time of deployment, not performance]

Can GDDR6 get out the door before HBM2 price drops make it more attractive? We'll see soon.
 

TRENDING THREADS