Soldato
Fair comment.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
*** AMD ThreadRipper ***
- Thread starter Boomstick777
- Start date
Fair comment.
separate entities or not, It makes no difference. Go high enough in the chain and you get to one person running the show.
Yep, they only mentioned the Broadwell architecture by name.
Yep, 2.1-3.5 GHz is their optimum range in terms of performance per watt, so almost all of their chips are within this region. Only really the R7 1800X sits outside of this range, aside from boost clocks.I know we're on an enthusiast's forum, but if we ignore performance desktop, the process is actually really, really good for the two biggest markets namely servers and APUs where the efficiency at around 3.0GHz is actually almost ideal. Yes, the process was designed for ARM SOCs but with a good efficiency curve up to around 3.0GHz it is actually very good for server chips as well.
(from The Stilt's Strictly technical thread over on AT).
Will be interesting how the 7nm process works out (those rumours of 5GHz+), but if the higher clock damage the efficiency a lot, AMD may have to do two designs for Zen+ one for servers and APUs and one for desktop and HEDT. Hopefully, by then they will have a bit more cash and volume to be able to afford this and for it to make sense.
AMD managed not only to bring a competitive product but also happily for AMD it came to market at pretty much the worst possible moment in the product cycle for Intel meaning Intel has to either go to massive expense and write off a lot of hardware or delay in response (or both). Hence resorting to dirty tactics.
Complacency has pretty much bitten Intel in the rear.
Soldato
If they achieve reasonable voltages at 5GHz, then surely when they underclock they will have reasonably good efficiency at lower clocks
Seems more like ignorance to me in terms of xerox.Complacency. What killed Xerox also
However judging by the Xeon product stack, it seems the something is happening to Intel.
Soldato
It was a very large dose of both if my memory is working properly. I think though that a much better match for what is happening to Intel is what happened to Eastman Kodak. A monolith in the photographic world and in real terms at the time, much bigger than Intel is now. They, just like Intel, ignored the competition (Fuji in Kodak's case) believing that Kodak's customer base would never shift. It did shift and it shifted very fast. Even when that shift was as clear as daylight to anyone, they still did next to nothing to counter it. Even the move Kodak made into digital photography was 6 or 7 years too late.
A lesson that Intel would do well to learn.
Sounds like arrogance goes hand in hand with complacency.
Soldato
Yep, spot on. Not only did Kodak's attitude mirror that of Intel at the moment but the way it mirrors Intel's is uncanny. Fuji launched a transparency film called Velvia (think Ryzen here). Kodak rubbished it saying the processing used on kodachrome was far superior and gave a better projected image. What Kodak completely ignored was that Velvia could be processed in any high street lab using the universal E6 process. This meant that the cost plummeted for Fuji users (no need to send the film to Kodak at an inflated price for them to develop and mount the slides). The rubbishing assault that Kodak made on Fuji was also a complete pack of lies.
Any of the above sound familiar to anyone ?
Soldato
- Joined
- 19 Oct 2008
- Posts
- 6,022
Film and image a little different from CPU's and a completely different approach to CPU design 
. It's interesting times that's for sure. Remains to be seen if there are any downsides to the infinity fabric approach especially as the core count increases which we'll be seeing soon. I wouldn't write off Intel - their chips are still mighty good just that their approach means high manufacturing costs for more cores.
. It's interesting times that's for sure. Remains to be seen if there are any downsides to the infinity fabric approach especially as the core count increases which we'll be seeing soon. I wouldn't write off Intel - their chips are still mighty good just that their approach means high manufacturing costs for more cores.It's more about the corporate attitude, that we are looking at.Film and image a little different from CPU's and a completely different approach to CPU design
Soldato
- Joined
- 19 Feb 2011
- Posts
- 5,849
+1 you don't have to look hard for stories of Intel employees spilling the beans on the company and how things are there currently.
Caporegime
Intel got caught with their pants down. The market for more cores is there, it's the future both for HEDT and enterprise.
Intel are doing this the wrong way. Huge cores, hard to cool, expensive to make.
Intel are doing this the wrong way. Huge cores, hard to cool, expensive to make.
Soldato
Film and image a little different from CPU's and a completely different approach to CPU design
The SkylakeX & Skylake Xeon mesh topology, is more or less similar to Infinity Fabric, and ironically all software requires exactly the same optimisation for those products with Zen.
However while Intel does it inside a single die, AMD just makes it span and scale evenly over multiple CPUs on the same PCB, cutting costs in the process.
FYI the Skylake/Kabylake (non X or xeon parts) still using the old ring topology.
Also if it was bad design it wouldn't be scaling 100% nor both Intel & Nvidia having announced last week, that they are working to produce similar product design, in 4-5 years.
The only drawback for Infinity Fabric, is that we are still stuck on DDR4 after gazillion years, and have to wait 2018-9 for DDR5. Which will greatly improve the speeds and performance of such design due to the double speed of the ram, than now. So what we see here today, is AMD being ahead 4-5 years.
And still Infinity Fabric has many unknown factors. For example how it will behave with a RX Vega GPU in the same system, sharing the HBM2 as fast cache? (as per Raja's video)
How would react with 2 Vega cards, all CPU & GPU interconnected with Infinity fabric, sharing 16GB of HBM2 cache?
Last edited:
I though the current latency problem with Ryzen wasn't caused by being bandwidth limited.
Soldato
Is not the bandwidth, is the speed because the parts communicate between each other at 50% the ram speed.
And the gains will be massive.
https://community.amd.com/community...emory-oc-showdown-frequency-vs-memory-timings
have a look here
3200-3466 is 10% almost!!!
Last edited:
Is not the bandwidth, is the speed.
https://community.amd.com/community...emory-oc-showdown-frequency-vs-memory-timings
have a look here
Just checking bandwidth is how much data you can move. Speed is how fast that data moves?
Whats the name for the combination of both speed and bandwidth?
Soldato
Bandwidth is how much data you can move..... eg Throughput in MB/s
The speed of the IF modules communicating with each other is 50% of the ram speed.
Sonic.
As it is for Zen's Infinitiy fabric both are important, but the actual Speed, which affects the bandwidth is what's important.
Since the I.F runs at 50% RAM speed, which directly affects the bandwidth between the CCXs.
With high speed RAM Infinity Fabric runs from 22GB/s all the way to 50GB/s, so you can see why it's so important.
I.F can scale to 500GB/s according to AMD, which is why they're also going to use it for APUs, and Navi/GPUs.