• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

FEATURED AMD ZEN Discussion (Previous Rumor Thread)

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
My assumption(yes assume this is complete conjecture) is that infinity fabric is an extension of the tech that AMD bought out several years back. They bought some large tech lab that was doing research on high speed network tech to be used in super computers. To me the Infinity Fabric is probably the consumer level implementation of the evolution of the patents they acquired back then. Nonetheless, for them to be used in the way helgaiden suggests still relies on a separate bus to carry the information between the GPUs so I highly doubt we will see an implementation like this anytime soon. However, in theory it is possible to eventually have scalable GPU through a similar type of tech with a few IO design tweaks.
 
Bluefalcon13 said:
From what I understand, it's just a fancy bus name. I'm probably wrong. As with any bus type, it's a combination of it's clock speed(think speed limit on a highway) and width (number of Lanes on highway).
Click to expand...




Well yeah, that i get. My message was more about the technology behind it. If there was a technical name for what it is (other than "interconnect" or something, thats too broad) then i'd prefer to use that term as "infinity fabric" sounds too buzzwordy for me. Is what it is though. I imagine whatever it technically is, AMD probably gonna keep it secret or something. Unless theres a patent on it that makes its details publicly available? i dont know. Either way, whatever the infinity fabric actually is, if the they could adpapt it for use in multi-GPU graphics card to operate as a single GPU, that could make for some interesting times. If infinity fabric currently shows great gains with faster DRAM, i wonder if the higher throughput of HBM or GDDR5X or GDDR6 if adapted could maintain the scale of those gains...
 
Here's AMD's take on it https://community.amd.com/thread/211126

Infinity Fabric is a coherent implementation which means that cache coherency is maintained across multiple processors externally and scaling up cores, in a CPU or a GPU, is not a problem and only limited by the bandwidth of the transport itself
Click to expand...
 
The infinity fabric is what I elude to in the Vega thread. Vega has HBM which is capable of working as a coherent cache system between the CPU and GPU. So you get to combine multiple GPUs, and CPUs with a nice I/O bus. Great for these top end systems that require it. Yet to prove its worth in gaming/benching.
 
What do y'all think of Ryzen 3 and its potential impact on the budget market? I could see it being an even bigger deal than Ryzen 5, to be honest. I mean think about it, for the past 6 years you COULD get a quad core chip for the price of an i3, but it'd either be an APU, an Athlon, or an FX 4xxx chip, all based on the steamroller/piledriver architectures. Now that AMD has released a new core to replace these, that's actually competitive with Intel, I think we could see a real big shift in budget market share.

The cheapest Ryzen PC I can build right now, that won't fall apart in less than a year, is $498. Ryzen 3 should bring that down to $450 or cheaper. For not that much of a performance hit over an i5. That's just incredible to me.

 
Something looks wrong to me. How could the Threadripper 1950X 16C/32T @ 3.4 GHz only score 24,723 on the multi-threaded test vs. 23,188 for the Ryzen 7 1800X 8C/16T @ 3.6 GHz. Double the threads for only a 6% bump in the score? Even allowing for the clock being 6% slower that's an anemic increase for twice the core count. All this while the Threadripper's single core score of 4216 at the lower clock rate is actually higher than the Ryzen 7's 4208. Odd.

And while I'm not familiar with Geekbench, it claims to be "Multicore Aware" and is endorsed by AMD. They state that "whether you're running Geekbench on a dual-core phone or a 32-core server, Geekbench is able to measure the performance of all the cores in your system".
 
DaveB said:
Something looks wrong to me. How could the Threadripper 1950X 16C/32T @ 3.4 GHz only score 24,723 on the multi-threaded test vs. 23,188 for the Ryzen 7 1800X 8C/16T @ 3.6 GHz. Double the threads for only a 6% bump in the score? Even allowing for the clock being 6% slower that's an anemic increase for twice the core count. All this while the Threadripper's single core score of 4216 at the lower clock rate is actually higher than the Ryzen 7's 4208. Odd.

And while I'm not familiar with Geekbench, it claims to be "Multicore Aware" and is endorsed by AMD. They state that "whether you're running Geekbench on a dual-core phone or a 32-core server, Geekbench is able to measure the performance of all the cores in your system".
Click to expand...

I'm not surprised at all.
Ryzen has been out for less than a year and Threadripper isn't even released yet.
There has been little if any progress in optimization for any benchmarks on the new AMD products.
 
Mr.Scott said:
I'm not surprised at all.
Ryzen has been out for less than a year and Threadripper isn't even released yet.
There has been little if any progress in optimization for any benchmarks on the new AMD products.
Click to expand...

Sorry, I'm still surprised since the Ryzen 7 bench looks right compared to the Intel 7700K and it's a recent update to the benchmark. And the release notes state Geekbench 4.1 includes improved support for Ryzen processors. Threadripper is just basically two Ryzen 7 8C/16T cores tied together via AMD’s Infinity interconnect so if the bench sees the extra cores/threads it should just use them. Again, I'm certainly not an expert in Geekbench but Threadripper is not something completely different, it is just an evolution of Ryzen 7.
 
Most new benchmarks have Ryzen support so if there was anything to improve then it was already improved. I mean most popular benchmarks like all from Futuremark series, AIDA64, Geekbench etc. Maybe there is something to improve on the BIOS side what will cause threadripper perform better but as Dave said, it's like 2x Ryzen 7 in one package. Maybe the way how internal connection works is causing some delays in the core communication, hard to say. If yes then there is a chance it will be fixed with new BIOS. It wasn't even released so I guess that motherboard manufacturers had no time to prepare fully stable BIOS with all fixes.
 
Don't mean me wrong but this list is a joke. Most popular 3000+ memory kits are not on the list and are working without issues. They should test higher frequency memory kits as nearly all 2666 and lower memory kits are working without issues on all motherboards.
New HyperX FURY were designed for Ryzen and have SPD=XMP what is "cheating" auto settings so all these kits should work. The same all 2133-2400 kits should work as most of them run at JEDEC profiles. There is nothing new what really helps users. It would be really bad if 2133 kits at JEDEC settings were not working.
List of vendors and link to Guru3D article. They are not even trying :p
All this makes me wonder who and how tested memory kits on this list as not all memory kits made for Ryzen are working without issues on all motherboards like 3200 FlareX.
 
All this makes me wonder who and how tested memory kits on this list as not all memory kits made for Ryzen are working without issues on all motherboards like 3200 FlareX
Click to expand...

I was under the impression AMD provided the list. Not so?
 
Check their document. Their data base on other sources, there are links and one is to guru3d which posted tests made on ADATA memory ( by ADATA ). Next thing is that AMD is not manufacturing motherboards. Still there is no info what "officially validated by AMD" means. They just provided data based on memory manufacturers tests. At least it's what I see after browsing their documents.
So my example is G.Skill FlareX 3200 which is marked as compatible because G.Skill performed tests but on some motherboards it's not working at 3200. It's simply compatible but with these boards on which G.Skill made tests.
 
Woomack said:
Check their document. Their data base on other sources, there are links and one is to guru3d which posted tests made on ADATA memory ( by ADATA ). Next thing is that AMD is not manufacturing motherboards. Still there is no info what "officially validated by AMD" means. They just provided data based on memory manufacturers tests. At least it's what I see after browsing their documents.
So my example is G.Skill FlareX 3200 which is marked as compatible because G.Skill performed tests but on some motherboards it's not working at 3200. It's simply compatible but with these boards on which G.Skill made tests.
Click to expand...

Do you have a link for AMD document sources?
 
on the top of AMD link were vendors links and about what is there is also on this list ... this PDF is as official as random post on the forums :) ... just published by AMD
 
You must log in or register to reply here.

Similar threads

mackerel
HEDT is back - Zen 4 TR pro and non-pro
Replies
7
Views
488
TerranBrackiatt
TerranBrackiatt
magellan
Low core count Rome gen. AMD EPYC CPU's: some questions
Replies
9
Views
1K
EarthDog
EarthDog
Overclockers.com
FRONTPAGE AMD Ryzen 9 7900X and 7700X Review: The Dawn of ZEN 4
2 3 4 5 6
Replies
108
Views
6K
TerranBrackiatt
TerranBrackiatt
Zerileous
Ryzen 5500 OC Potential
Replies
6
Views
5K
freeagent
freeagent
D
Recommended memory for AMD Zen 2/3
2
Replies
26
Views
2K
Woomack
W
Back