BitsAndChips: entry-level Threadripper 16c/32t to come at the nice low price of $849

[Taf the Ghost]

When you have highly paid engineers with good leadership, they can do some pretty amazing things.  It's the leadership aspect that's normally the hardest to nail down.

 

There's also just the reality that this is the direction we've been going since the Dual Cores became standard. And it's also not like Intel isn't headed that direction with EIMB. AMD just is going for the multiple uses of a single package setup, which should be cheaper to manufacture. 

[Taf the Ghost]

4 minutes ago, Notional said:

They don't have to. Right now they have about 0% market share in the high end enterprise market and the HEDT markets. What they need to do is capture market share. That always costs money, and as such it would be fine for AMD to cut, say 20% of the profit margin, so they could be cheaper. If the prices are holding up, a 16 core TR will beat the crap out of Intel's 10 core, at a lower cost, and render ALL higher core Intel CPU's completely pointless. Having a lower profit margin but selling a lot more chips will increase the overall revenue and in the end profits as well.

Technology's profit margin also has to be calculated with the Sales Volume in mind. R&D is normally the biggest cost that has to be retired (it's the reason $110-130 being full packaging costs for TR isn't surprising), so the more sales that R&D cost is spread over, the greater the effective margin.  It makes finding the most effective Price actually somewhat difficult, as there's multiple aspects one is trying to maximize. 

[Notional]

4 minutes ago, Taf the Ghost said:

Technology's profit margin also has to be calculated with the Sales Volume in mind. R&D is normally the biggest cost that has to be retired (it's the reason $110-130 being full packaging costs for TR isn't surprising), so the more sales that R&D cost is spread over, the greater the effective margin.  It makes finding the most effective Price actually somewhat difficult, as there's multiple aspects one is trying to maximize. 

Indeed:

Now factor in price elasticity, demand change based on time of year (Q4 sales > Q1 sales), differentiated sales (consumer > b2b/vendors), etc, and of course competition (although Intel basically priced themselves out of the HEDT platform if this is true.

[Taf the Ghost]

It also gets messy because of the existence of multiple SKUs, you end up pricing some parts at less optimal levels, on the first run, as keeping your market properly aligned can move buyers to products they want at a better margin for the seller.

[leadeater]

6 hours ago, Notional said:

They don't have to. Right now they have about 0% market share in the high end enterprise market and the HEDT markets. What they need to do is capture market share. That always costs money, and as such it would be fine for AMD to cut, say 20% of the profit margin, so they could be cheaper. If the prices are holding up, a 16 core TR will beat the crap out of Intel's 10 core, at a lower cost, and render ALL higher core Intel CPU's completely pointless. Having a lower profit margin but selling a lot more chips will increase the overall revenue and in the end profits as well.

Can't price yourself too low though, there's actually a thing where if you're too cheap it puts buyers off as it puts you under suspicion of being a low quality or inferior product. You also don't want to give up significant amounts of profit just for a few percent market share, that's going to take a long time to gain so short term gain at the expense of long term R&D isn't that smart.

 

AMD has no market share in the enterprise market because HPE/Dell etc don't have servers that have their CPUs in them, you can't have what isn't there.

[Notional]

1 minute ago, leadeater said:

Can't price yourself too low though, there's actually a thing where if you're too cheap it puts buyers off as it puts you under suspicion of being a low quality or inferior product. You also don't want to give up significant amounts of profit just for a few percent market share, that's going to take a long time to gain so short term gain at the expense of long term R&D isn't that smart.

 

AMD has no market share in the enterprise market because HPE/Dell etc don't have servers that have their CPUs in them, you can't have what isn't there.

I get what you mean, but I don't think that factors into the tech market. Especially not the HEDT markets, as people who needs platforms and chips, tends to be very educated on the products and their performance. So I don't think that would factor in here. After all "quality" (performance) is very easy to measure in computing.

 

As for production costs, they are very low. We know there is an 80+% yield on 8 core CPU's which is very impressive, and we know the 16 core is not 1 die 16 core like Intel's insanely expensive chips, but rather 2 dies with Infinity Fabric. This makes a 16 core TR much much cheaper to manufacture with almost an 80% yield rate (assuming attaching 2 dies to each other will result in some defective packages), so profits should be very high. After all Intel's pricing is not based on R&D costs or manufacturing costs but rather inflated into space, due to a monopolistic market situation. This is something people misunderstand with Intel's and Nvidia's pricing. It's massive profits that are the reason you pay that insane pricing, not production cost or R&D.

 

My point is simply that lower prices can result in higher revenue and if done correctly, even higher profits. PC hardware seems to be pretty price elastic.

[leadeater]

3 minutes ago, Notional said:

I get what you mean, but I don't think that factors into the tech market. Especially not the HEDT markets, as people who needs platforms and chips, tends to be very educated on the products and their performance. So I don't think that would factor in here. After all "quality" (performance) is very easy to measure in computing.

It does in the enterprise market when you have to write business cases for purchase and finance people have to approve it who have zero understanding of tech and would ask the question, it's their job to make sure you don't make bad purchases.

[Notional]

Just now, leadeater said:

It does in the enterprise market when you have to write business cases for purchase and finance people have to approve it who have zero understanding of tech and would ask the question, ti's their job to make sure you don't make bad purchases.

Well, the enterprise EPYC chips should be priced higher than TR chips though, so I don't think it would be a problem. Besides most competent finance/business people would take the cheapest option as long as it lives up to the demands. 

[Dylanc1500]

1 hour ago, leadeater said:

It does in the enterprise market when you have to write business cases for purchase and finance people have to approve it who have zero understanding of tech and would ask the question, it's their job to make sure you don't make bad purchases.

This is very great point.

 

Constraints of scale can be a big factor as well. If it takes 5 cabinets with Intel, and 9 cabinets with AMD in order to get X amount of work done.

[Masada02]

10 hours ago, leadeater said:

I think AMD should sell their TR CPUs at the highest possible price that consumers would find acceptable and buy in decent volumes. Get some of that sweet cash coming in to invest in Zen+ etc.

Need to balance that with volumes though. The lower the consumer price the more will sell, and the more that sell the the more production costs decrease. 

 

I'm betting they are pricing for volume. If they can get the volume then we all get cheaper chips. 

[Misanthrope]

800 to 1000 would be really damn good starting price but might seem unrealistic depending on when they plan to launch.

 

Also I simply cannot let another Threadripper thread pass without another theme song, this might be the best one so far though:

 

Spoiler

 

 

[leadeater]

14 minutes ago, Dylanc1500 said:

This is very great point.

 

Constraints of scale can be a big factor as well. If it takes 5 cabinets with Intel, and 9 cabinets with AMD in order to get X amount of work done.

Depending on workloads and requirements AMD servers could use less rack space. They can support more NVMe SSDs off the CPU, more GPUs and have more cores in similar configurations. The more cores is pretty much the only thing in question as to will it perform as well as the fewer Intel ones.

[THE_maverick]

4 hours ago, leadeater said:

You also don't want to give up significant amounts of profit just for a few percent market share, that's going to take a long time to gain so short term gain at the expense of long term R&D isn't that smart.

2

Have you ever heard of Tesla, Amazon or Google?  Don't make statements that are not true, 'cause that is just plain ol' trolling.

[leadeater]

Just now, THE_maverick said:

Have you ever heard of Tesla, Amazon or Google?  Don't make statements that are not true, 'cause that is just plain ol' trolling.

Yes, but what is your point?

[THE_maverick]

Just now, leadeater said:

Yes, but what is your point?

That them alongside, for example twitter and youtube, have never turned a profit.  SSo companies in this day and age value marketshare and gross above net profit.  #TheMoreYouKnow 

[leadeater]

Also Tesla is the demonstration of my point, they are not undercutting anyone, only have a very small market share, are very careful about making sure there is long term viability of the company.

[leadeater]

16 minutes ago, THE_maverick said:

That them alongside, for example twitter and youtube, have never turned a profit.  SSo companies in this day and age value marketshare and gross above net profit.  #TheMoreYouKnow 

Wow that is an over simplification of everything, they can take the extra profit from the higher price of the product and put in back in to R&D for Zen+ and have a net zero yearly profit if they so wish .

 

Or they could pay off debt.

 

Edit:

Also Amazon does make a profit, $912 million over the last year.

http://phx.corporate-ir.net/phoenix.zhtml?c=97664&p=irol-reportsother

[Lays]

On 6/2/2017 at 7:41 AM, tom_w141 said:

CCX latency has a very small effect on performance so if enough performance is there then it will be very hard to justify paying double for the Intel. Price is key here.

who said intel's competing product will be 2x the price?

 

We already know the intel product will have more single thread performance, not sure why people are talking about comparing the 18 core intel chip to a 16/32 lower clocked + lower single thread performance chip...  

[leadeater]

Just now, Lays said:

We already know the intel product will have more single thread performance, not sure why people are talking about comparing the 18 core intel chip to a 16/32 lower clocked + lower single thread performance chip...  

Which is why it's worth more of course. We don't know if it's clocked higher though but it'll almost certainly boost higher.

[Taf the Ghost]

53 minutes ago, Lays said:

who said intel's competing product will be 2x the price?

 

We already know the intel product will have more single thread performance, not sure why people are talking about comparing the 18 core intel chip to a 16/32 lower clocked + lower single thread performance chip...  

Odds are the top level Ryzen chip will be a lot higher clocked than the top Intel chip.  We'll see if it's enough of a difference to overcome the slight IPC difference. Which is also why Intel hasn't released the clock specs of the 12c to 18c parts. Those parts are repurposed Xeon chips, which normally run in the low 2 Ghz range.  Single Thread Turbo is normally not that important in workloads where you want all of those cores.

[Dylanc1500]

4 hours ago, leadeater said:

Depending on workloads and requirements AMD servers could use less rack space. They can support more NVMe SSDs off the CPU, more GPUs and have more cores in similar configurations. The more cores is pretty much the only thing in question as to will it perform as well as the fewer Intel ones.

I thought there was an issue with support for more than two socket configurations per board. Maybe I misunderstood. Space constraints on the system boards due socket size could add up as well. 

 

I suppose we will see how things go.

[leadeater]

28 minutes ago, Dylanc1500 said:

I thought there was an issue with support for more than two socket configurations per board. Maybe I misunderstood. Space constraints on the system boards due socket size could add up as well. 

 

I suppose we will see how things go.

Two socket maximum yes but that is by far and large the most common server configuration used. There are also hybrid blades that can take 4 dual socket nodes in the rear of the server and either 12 LFF or 24 SFF disks in the front.

https://www.supermicro.com/products/system/2U/2028/SYS-2028TP-HTR-SIOM.cfm

 

In a 2U configuration you can get 8 sockets which if you are using the upcoming AMD CPUs that's 256 cores, and in a rack with 38U usable (switches at top etc) that's 4864 cores.

[Lays]

2 hours ago, leadeater said:

Which is why it's worth more of course. We don't know if it's clocked higher though but it'll almost certainly boost higher.

hi *waves*

1 hour ago, Taf the Ghost said:

Odds are the top level Ryzen chip will be a lot higher clocked than the top Intel chip.  We'll see if it's enough of a difference to overcome the slight IPC difference. Which is also why Intel hasn't released the clock specs of the 12c to 18c parts. Those parts are repurposed Xeon chips, which normally run in the low 2 Ghz range.  Single Thread Turbo is normally not that important in workloads where you want all of those cores.

 Why do you think a chip based off a bunch of chips that could barely do 4 ghz with less cores is going to magically clock through the roof? Even if the AMD chip is slightly higher clocked, we've all seen how horrific the OC headroom is on the Ryzen chips so far, it's pretty bad lol.  In terms of SKX clocks, all you really have to do is look at normal Skylake chips and how they clock, and translate it to the extreme parts lol, it's been pretty similar for years now.

[AnonymousGuy]

5 hours ago, leadeater said:

Depending on workloads and requirements AMD servers could use less rack space. They can support more NVMe SSDs off the CPU, more GPUs and have more cores in similar configurations. The more cores is pretty much the only thing in question as to will it perform as well as the fewer Intel ones.

You're missing a whole laundry list of problems that AMD's architecture has in scaling up.  The weird multi-die multi-socket arrangement is going to create a lot of coherency scaling allocation blah blah blah problems that you don't get on a monolithic "real" server part.  Then you've got IPC, software being unanimously optimized for IA, and the cost of revalidation on an AMD platform.  Politically it'd be pretty hard to recommend AMD as well when Intel can drop 20 people to manage your account and AMD has one guy named Jeff.

 

"no one ever got fired for recommending IBM".  Except Intel in this case.

[leadeater]

8 minutes ago, AnonymousGuy said:

You're missing a whole laundry list of problems that AMD's architecture has in scaling up.  The weird multi-die multi-socket arrangement is going to create a lot of coherency scaling allocation blah blah blah problems that you don't get on a monolithic "real" server part.  Then you've got IPC, software being unanimously optimized for IA, and the cost of revalidation on an AMD platform.  Politically it'd be pretty hard to recommend AMD as well when Intel can drop 20 people to manage your account and AMD has one guy named Jeff.

 

"no one ever got fired for recommending IBM".  Except Intel in this case.

I wouldn't make any of those judgments until products are actually out. HPE seemed happy enough with the parts and are doing internal performance testing now for Gen10 servers.

 

You're making guesses at those issues since none of that information exists out in the public domain yet. I have no doubt there is going to be optimizations required for these CPUs and configurations but it's not like HPE, VMware, Microsoft etc don't already have these systems now. 

 

8 minutes ago, AnonymousGuy said:

Politically it'd be pretty hard to recommend AMD as well when Intel can drop 20 people to manage your account and AMD has one guy named Jeff.

You talk to HPE/Dell not AMD.