The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
Besides the Xbox, if big Windows and all its ecosystem runs on x86 chips, and the second source of x86 chips (AMD) is stumbling, then isn't it in Microsoft's own interest to prop it up so they don't have to rely on a single source (Intel)?
 
Microsoft does not use Intel as a supplier, so there is no direct cost being passed onto Microsoft as a result of a lack of competition.
Even if there were, AMD's dwindling market presence and Intel's actions show that it no longer presents that much of a competitive pressure across a vast range of products.
Intel's competition with the older versions of its products users already own is a stiffer problem, and a more notable downward pressure on prices in the consumer space.

Workstation and server parts also have this, although a lot of workloads there have not stagnated as much as the PC realm.
AMD has no appreciable presence there, per its own admission since Zen is meant to be its re-entry. If Microsoft were concerned about that, it's years too late.
 
Sure, but despite its size if it wasn't for AMD we would be all using Itanium today.
I've always thought that the upper hand in the Wintel duopoly belonged to Microsoft, not Intel. And a world without AMD increases Intel's power relative to Microsoft. Even if Bulldozer sucks today, I think AMD should get the chance to release Zen. If Zen also suck then it's over.
It seems that it doesn't really matter that new Intel CPUs aren't much faster or better than their older CPUs. Intel will price its new CPUs just as high and because of that we still have things like overpriced dual-cores in this day and age.
 
Sure, but despite its size if it wasn't for AMD we would be all using Itanium today.
That might have been some kind of argument for a thank-you fruit basket a decade ago, but that doesn't make a case now.

I've always thought that the upper hand in the Wintel duopoly belonged to Microsoft, not Intel. And a world without AMD increases Intel's power relative to Microsoft.
AMD has very little power today, perhaps less than the power Microsoft has with an extra 3-5 billion dollars not spent.

It seems that it doesn't really matter that new Intel CPUs aren't much faster or better than their older CPUs. Intel will price its new CPUs just as high and because of that we still have things like overpriced dual-cores in this day and age.
Intel's manufacturing prowess and tight integration mean it can produce its chips at a cost level that AMD cannot match, since it must pay the margins of foundry partners, so it's not a clear case that an AMD with a larger market share would actually lead to significantly lower prices. Intel's troubles trying to clear its own inventory have had more of an effect than AMD. AMD's numbers for IPC improvements for Zen and its high-end 8-only core offering for 2016, show that while it may clear the low bar of Bulldozer's architecture, it doesn't look likely to be much faster than Intel's older CPUs nor play in the markets that really have the high prices.
 
I think AMD won't be doing Intel strategy with i5 and i7.

And high-end Zen would have 8 cores (16 logical) and middle-end 4 (8 logical)
vs Intel's 4 (8), 4 (4)

this could give big advantage to AMD.
 
8 cores per socket in 2016 may be a lot in the desktop space, but you can buy up to 18 core Xeons from Intel today, with Broadwell-EP/Skylake expected to go to 22/28 cores per socket.
 
Question is tho - is 8C/16T AMD Zen's performance anywhere near Intel's upcoming 4C/8T performance...? Historically, this would not exactly be a bet you would put all your money on, so to speak.
 
Question is tho - is 8C/16T AMD Zen's performance anywhere near Intel's upcoming 4C/8T performance...? Historically, this would not exactly be a bet you would put all your money on, so to speak.

In multi-threaded applications, Vishera (FX-8350) is within 10% of Haswell (i7-4770) so it's a reasonable assumption. Games and other somewhat serial applications are a very different story, although DX12 may help with that.
 
I was talking about desktop.

Sorry - I read "high-end" to mean server class. Still, I'm not sure how AMD can gain an advantage this way, considering that you can buy an 8 core Broadwell Xeon D 1540 today for ~ $600. Yeah, it's a server part with base clock of only 2GHz (and just a 45W TDP), but it's hard for me to believe that it would be difficult to turn it into a high end desktop SKU by simple TDP adjustment.
 
In multi-threaded applications, Vishera (FX-8350) is within 10% of Haswell (i7-4770) so it's a reasonable assumption. Games and other somewhat serial applications are a very different story, although DX12 may help with that.

True that. The problem with Bulldozer is not so much that it is a bad architecture. It is not bad.

The problem with Bulldozer is Microsoft and the other software it's running on.
 
The problem with Bulldozer is Microsoft and the other software it's running on.

And slow L2 cache.. and low floating-point performance because of having 1 FP unit for every module with 2 integer units, and low memory bandwidth, and not very effective branch-prediction units...
From the top of my head.

Regardless, any 3/4 module bulldozer should be sufficient for games, if DX12 and Vulkan are everything we've been promised so far.
 
And slow L2 cache.. and low floating-point performance because of having 1 FP unit for every module with 2 integer units, and low memory bandwidth, and not very effective branch-prediction units...
From the top of my head.

Regardless, any 3/4 module bulldozer should be sufficient for games, if DX12 and Vulkan are everything we've been promised so far.

Remember also that Bulldozer has never received any manufacturing process improvements.

Imagine Excavator on Intel's 14 nm!

Can you imagine that it could be a different story?!
 
Can you imagine that it could be a different story?!

Yes, I imagine it'd be a smaller chip, giving AMD better profits, and it'd probably consume less power.
I don't think it'd help absolute performance that much, because the maximum frequencies for CPUs have stagnated with newer processes.
 
Manufacturing processes and "Microsoft" aside, the Bulldozer family remains plagued by poor per-thread performance.
 
And the Surface non-Pro is the first to use Intel's Cherry Trail.
 
Status
Not open for further replies.
Back
Top