FEATURED AMD ZEN Discussion (Previous Rumor Thread)

[mackerel]

It certainly is, pretty much two Ryzen CPUs stuck together and Ryzen is already big in the CPU world

Is it big? In some earlier semiconductor conference there was a comparison showing a Ryzen core is smaller than a modern Intel core, so we're not including things like the iGPU here. I think that was in part due to the decision to not match FP performance with Intel.

 

[Johan45]

The IHS is any way when compared to skylake type intel IHS

 

[mackerel]

Ok, you're talking about package? I thought you meant silicon.

 

[Johan45]

Even the die itself would be nearly double that of the kaby i7 ( 8c vs 4c), that's why the IHS is so large

 

[mackerel]

Have another rumour: "entry level" 16c part claimed to be US$849. Originally from BitsAndChips twitter, it's spreading around the usual rumour sites now. The other rumour I posted previously claiming to be the lineup only had two 16c parts differentiated by clocks. I wonder if this would put the upper model at $999 or would there be a bigger jump for the top of the tree?

 

[Johan45]

If there's only two 16c and IF $849 is accurate for the lower tier , I'd place the XFR version $899-$949

 

[Dolk]

Pcie lanes... meh. So few need more than 28...its not a selling point (to me).

PCI-E may not be required in Gaming/Benching, but for everyone else that fully loads up servers/workstations/cryptocurrency/F@H/SETI setups this will be a huge advantage over Intel. No longer requiring shared bus for the 3rd or 4th PCIE slot for all x16 PCI-E 3.0.


BTW, Servers really like the idea of more PCI-E. Reasons being for <NDA>, <NDA>, and <NDA?>. Server market is looking to shift gears around 2018-2020 timeframe, and X86 CPUs that have the most PCI-E will probably win. For instance, NVME devices are starting to port over to PCI-E. Come PCI-E Gen 4.0, this will outright beat SAS or SATA. The more PCI-E that comes back to the CPU will allow for higher I/O. Or maybe it will be NVMe -> FPGA -> X86

 

[EarthDog]

Like i said, so few need it when looking at the overall market. Again, amd is ahead of its time for few current payoffs. Lets talk in 3-5 years when it may be more necessary for the masses.

Servers, not part of the discussion, really...different space.

 

[Dolk]

We all have different perspectives, but my point is that the F@H/SETI guys will really like the threadripper setup. If consumer grade mobo designers focus on PCI-E expansion rather than USB3.1/C/A and other junk, than they will get a nice little bump in performance.

 

[Alaric]

I'm lucky (!?) that I'm staying on Windows 7, leaving me at Skylake for my top available processor. An HEDT would be a complete waste of money for me, as the only reason I would build a Windows 10 rig would be for just gaming. For that, IPC is king. I don't see more than 6c/12t being necessary for that for a long time. Heck, it will be quite a while before my current chip can't keep up gaming. I'm a fan of AMD, and I hope Ryzen and Threadripper put the hurts on the Big Blue Giant, but Intel has me for the foreseeable future because they do what I want better than Team Red.

 

[Woomack]

PCI-E may not be required in Gaming/Benching, but for everyone else that fully loads up servers/workstations/cryptocurrency/F@H/SETI setups this will be a huge advantage over Intel. No longer requiring shared bus for the 3rd or 4th PCIE slot for all x16 PCI-E 3.0.


BTW, Servers really like the idea of more PCI-E. Reasons being for <NDA>, <NDA>, and <NDA?>. Server market is looking to shift gears around 2018-2020 timeframe, and X86 CPUs that have the most PCI-E will probably win. For instance, NVME devices are starting to port over to PCI-E. Come PCI-E Gen 4.0, this will outright beat SAS or SATA. The more PCI-E that comes back to the CPU will allow for higher I/O. Or maybe it will be NVMe -> FPGA -> X86

Problem is only that before new AMD will hit server market and people start to trust it enough to buy it in larger quantities then Intel will have competitive platform. Threadripper can be popular in home computers for anything that needs more cores and is better priced than Intel like F@H/SETI but on the other hand graphics cards don't really need high PCIE bandwidth for all that. Most computers for mining bitcoins are using mainly PCIE x1.
Theoretical win is on AMD side looking at threadripper but in real it's only theory and I doubt that large part of the market will move to AMD side because of that.
NVME in servers won't be popular for a long time and I mean like really long time. Recently I see that server SSD are moving to SATA/SAS MLC while they were mainly SLC in the past. All because of costs while NVME in most cases is waste of money.
Whatever now seems like a great piece of hardware will start to be popular in servers in 2-3 years. Then we can start to talk what side is better. Now we have a bunch of rumours and technical documents. Most are more like marketing documents.

 

[mackerel]

https://community.amd.com/message/2796982

Just saw there are growing reports of unexplained errors when compiling on Linux. Not something I do, but thought I'd mention it. I'm not worried about it. Intel also have had various bugs which are usually fixed or worked around in bios updates later on.

 

[Vishera]

So what do you guys think about threadripper? Personally, I think AMD hit a pretty sore spot for Intel. Ryzen was a little lacking compared to Kaby lake, but it was on par or better than Broadwell-E. Now, they've caught Intel with their pants around their ankles, and Intel has hurried to put together a plan for an 18 core CPU. Anybody wanna place bets on how sloppy that launch is gonna be?

 

[Bluefalcon13]

I think it's a bit overkill. Nice shot across the bow from AMD though. I just don't see a need for 16 cores myself. I guess I could fold on it, but I'm quite happy with Ryzen right now

 

[Alaric]

What retail sales calls a "loss leader". Threadripper will generate buzz and competitive numbers for the marketing people. It will sell far more lower cost chips (they hope).

 

[mackerel]

What retail sales calls a "loss leader". Threadripper will generate buzz and competitive numbers for the marketing people. It will sell far more lower cost chips (they hope).

I understood "loss leaders" to be products sold below cost to get customers in, who then buy other stuff at the same time to offset those losses and make an overall gain. I don't see that with AMD, especially not ThreadRipper. What we have is a "hero product" fight. You need to show off a leading product that is so wow and great, people associate the positive feelings with the whole range, including the lower models they buy. What Intel have done with the 18 core announcement is to say, their hero product is better than AMDs hero product. It doesn't matter if they never sell any, it is purely about perception of the high end rippling down.

 

[Johan45]

AMD's EPYC SOC, 32 core 64 thread, 128 PCI lanes on one chip. They're hoping to claw back some business http://www.eetimes.com/author.asp?section_id=36&doc_id=1331764

 

[DaveB]

So it's virtually four Ryzen 7 8C/16T cores tied together via AMD’s Infinity interconnect which then runs at the DDR4 memory speed. And Threadripper is two Ryzen 7 8C/16T cores, so not really much different. I can see how it was very simple for AMD to quickly scale up from Ryzen to HEDT and server chips by going this way. We'll just have to see how well they run and how much OS tweaking will be required to optimize performance. Again, not relevant to anything I need but good to see competition on the high end.

 

[Helgaiden]

How revolutionary is the infinity fabric interconnect if when all is said and done, it can be used to scale this sort of stuff up or down and maintain performance?

Could it be transplanted into the radeons? could we see multi-gpu radeons with infinity fabric finally be treated as a single GPU and not at the mercy of Crossfire?

 

[Bluefalcon13]

How revolutionary is the infinity fabric interconnect if when all is said and done, it can be used to scale this sort of stuff up or down and maintain performance?

Could it be transplanted into the radeons? could we see multi-gpu radeons with infinity fabric finally be treated as a single GPU and not at the mercy of Crossfire?

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).