The Ryzen "ThreadRipper"... 16 cores of awesome

[tamz_msc]

Looks like the ST score is being held down by the use of DDR-2400 memory. If you look at the individual ST benchmark scores, the Whitehaven CPU is actually beating the 1700X in almost all of the individual benchmarks outside of memory copy/latency/etc.
https://browser.primatelabs.com/v4/cpu/compare/3324737?baseline=3328483

Thats...interesting.

Also goes to show why giving 20% weight to memory scores to calculate an overall CPU score is a stupid idea.

 

[mikk]

Tell me again why ST is low?

https://browser.primatelabs.com/v4/cpu/3328483

4000 isn't low for you on a high end CPU that will cost at least 1k USD?

 

[tamz_msc]

4000 isn't low for you on a high end CPU that will cost at least 1k USD?

What is so surprising about that score? It is almost the same as a 3.4GHz Ryzen 8-core. How do you know it'll cost 1000$ or more? Even if it does, you're paying that money for up to 16 cores, single-thread performance is going to be a byproduct of whatever clocks are achievable with high core count CPUs.

 

[Shivansps]

What is so surprising about that score? It is almost the same as a 3.4GHz Ryzen 8-core. How do you know it'll cost 1000$ or more? Even if it does, you're paying that money for up to 16 cores, single-thread performance is going to be a byproduct of whatever clocks are achievable with high core count CPUs.

Well they need to price it more expensively than the lower clocked EPYC 16C ones. 1K seems about right. The 1P EPYC 7351P is 2.4Ghz base/2.9Ghz turbo at about $750, a 3.4Ghz 16C TR needs to be WAY over that or they will eat their own EPYC sales. They cant afford that when EPYC is 4 die vs 2 die TR.

 

[mattiasnyc]

Well they need to price it more expensively than the lower clocked EPYC 16C ones. 1K seems about right. The 1P EPYC 7351P is 2.4Ghz base/2.9Ghz turbo at about $750, a 3.4Ghz 16C TR needs to be WAY over that or they will eat their own EPYC sales. They cant afford that when EPYC is 4 die vs 2 die TR.

I'm not so sure that's the right way of looking at it though. EPYC is more than cores and speed. And because they won't share the same socket I don't see how they'll compete. I mean, if you're looking to build a server and need more than 64 PCIe lanes then you'll need EPYC, TR won't suffice because of that. And you can't get a TR and run it in an EPYC socket with the intention of changing later.

So I can absolutely see how they won't bump the price too high. If anything what'll decide the TR price will more likely be Intel's pricing. That's my guess anyway.

 

[tamz_msc]

Well they need to price it more expensively than the lower clocked EPYC 16C ones. 1K seems about right. The 1P EPYC 7351P is 2.4Ghz base/2.9Ghz turbo at about $750, a 3.4Ghz 16C TR needs to be WAY over that or they will eat their own EPYC sales. They cant afford that when EPYC is 4 die vs 2 die TR.

When competitively priced Xeons don't eat into HEDT sales, I think it's safe to assume that the opposite won't happen in case of Epyc and Threadripper.

Besides, sockets are incompatible, unlike X99.

 

[Topweasel]

I have said it before and the EPYC pricing showed true. AMD has all the segmentation they need by platform. They aren't worried about which variation of Zeppelin implementation as long as you are buying Zeppelin's.

Honestly if someone created a Gaming board version of EPYC and people bought cheaper 16c EPYCs for that. I don't think they care. Same thing with someone building a server with TR performance price.

There will be a price difference between TR16 and E16. Based on clocks. But we shouldn't be looking at it through the Blue colored glasses.

 

[Shivansps]

When competitively priced Xeons don't eat into HEDT sales, I think it's safe to assume that the opposite won't happen in case of Epyc and Threadripper.

Besides, sockets are incompatible, unlike X99.

Intel went long ways to ensure HEDT could never cripple Xeon sales, both in X99 and in X299, on X99 they were way too expensive.... a Xeon 8C was $440, the 6900K was $1000, second they removed server features like ECC on HEDT, we could go on but i think i made my point here.

You need to balace those things out here, AMD does not want TR to eat EPYC sales. EPYC cost more money to them, they are selling the server platform, and TR dies are kinda the cherry picked ones. And Intel pricing sure dosent help at all as well. They could charge $1200 for it and it will still be good.

If they cost $850 nobody in its right mind whould buy EPYCs 16C cpus. Clock difference and less huma overhead... less memory slots to populate as well.

 

[tamz_msc]

Intel went long ways to ensure HEDT could never cripple Xeon sales, both in X99 and in X299, on X99 they were way too expensive.... a Xeon 8C was $440, the 6900K was $1000, second they removed server features like ECC on HEDT, we could go on but i think i made my point here.

You need to balace those things out here, AMD does not want TR to eat EPYC sales. EPYC cost more money to them, they are selling the server platform, and TR dies are kinda the cherry picked ones. And Intel pricing sure dosent help at all as well. They could charge $1200 for it and it will still be good.

If they cost $850 nobody in its right mind whould buy EPYCs 16C cpus.

If all you wanted was moar cores, then Xeons gave you that at almost half the price-point of HEDT chips, never mind ECC. Yet we have AnandTech claiming that the 6950X might have actually been the best-selling Broadwell-E HEDT chip.

Like I said before, sockets are incompatible, PCIe lane count between the two is different, software features are missing. The only thing TR might have going for itself is better ECC support, and perhaps 2TB RAM.

 

[Topweasel]

Intel went long ways to ensure HEDT could never cripple Xeon sales, both in X99 and in X299, on X99 they were way too expensive.... a Xeon 8C was $440, the 6900K was $1000, second they removed server features like ECC on HEDT, we could go on but i think i made my point here.

You need to balace those things out here, AMD does not want TR to eat EPYC sales. EPYC cost more money to them, they are selling the server platform, and TR dies are kinda the cherry picked ones. And Intel pricing sure dosent help at all as well. They could charge $1200 for it and it will still be good.

That's not the case. If it was AMD wouldn't produce Ryzen at all and leave all dies for EPYC. They have ASP goals. But I doubt they care at all since it's the exact same die on all three lineups. TR eating EPYC and EPYC eating TR means nothing.

They have a currency it's lots of cores at a good price. Pick your cores. Pick your platform. Done. Add on to the fact that a 16 core TR is actually cheaper to manufacture than a 16c EPYC only adds to the point AMD selling TR16 top clock, selling at the same price as EPYC 7351 would mean more margin for AMD. Though the selling of the 7351 gives much better margins than selling 4 1400s. Though even there the difference is small.

 

[moinmoin]

Registered ECC will only be available with Epyc. Also Epyc is four dies/NUMA nodes whereas TR is two. So given the same core count, most desktop software not being prepared to work across multiple NUMA nodes will make Epyc the worse pick for that use case. TR is a HEDT chip, Epyc a server chip.

 

[lobz]

Again I am going to point out lack of overlap. There isn't an 8c server variation.

Time is a tricky mistress

Sent from my VTR-L09 using Tapatalk

 

[Topweasel]

Time is a tricky mistress

Sent from my VTR-L09 using Tapatalk

Lol yes it is. I'll eat crow on that one. I was not expecting a 1c per CCX option. Hell I wasn't even expecting the 16c option but that makes some sense at the time. In retrospect I should have seen it coming the chip is less a CPU and more of a facilitator for the 128 PCIe lanes and 8 channels of memory.

But Ryzen 1700 isn't going to cannibalize EPYC's sales and AMD's pricing for that EPYC also proves another point that they don't feel the need to really pump up the price because "server". With both TR and the 1700, even if priced lower than the EPYC counterpart, AMD would make more off the sale of the other chip (well except the 8c EPYC some of those Dies would be scrap otherwise). A $800 TR16 HC would be worth more to AMD than a EPYC 16c HC at the same price. People getting the EPYC want the memory channels and RDIMMS with an array of PCIe connection type option. People wanting the TR want cores, clocks, and PCIE 16x ports. There isn't an overlap and keeping the prices similar makes sure people get the proper CPU and they don't have to worry about people trying to get a CPU to do something it wasn't meant for. Intel does what they do because they can dictate market prices and therefore have chosen to price everything without balance in terms of silicon differences and the lack thereof. Intel had to limit Xeon's working in x99 because they choose to use the same Silicon for the server and HEDT, yet limit choices, and then price a server part less than the HEDT market.

Why would AMD care if someone thought about an EPYC but decided on TR. If the prices are similar people will buy the platform they need. So if someone wants faster clocks is fine with 64 PCIe lanes, doesn't need 8 channels of memory nor registered dimms, why would they care? Intel cares because they are dominant and think that they can dictate pricing and market. AMD just wants people to buy stuff.

 

[ub4ty]

> Deleted link to Bench leak from wccftech..
Numbers appear to be bogus

 

[HurleyBird]

I wouldn't take the leak as authoritative. In this era of turbo boost, it's more difficult than ever to tell what the frequency of these ES chips is compared to what will eventually ship. Ryzen 7 came out much better than the various ES leaks suggested it would, for example.

 

[ub4ty]

I wouldn't take the leak as authoritative. In this era of turbo boost, it's more difficult than ever to tell what the frequency of these ES chips is compared to what will eventually ship. Ryzen 7 came out much better than the various ES leaks suggested it would, for example.

Yeah, I just deleted it. The numbers appear to be highly bogus. Thanks : wccftech

 

[HurleyBird]

Well, I wouldn't go that far. The leak is probably valid, it just may or may not be indicative of final performance.

 

[tamz_msc]

It's the usual deal with Geekbench - doesn't detect NUMA nodes properly. Search for Xeon Gold - you'll find some interesting results.

That, and the fact that there's now way of discerning Turbo clocks makes comparisons meaningless.

 

[nad-]

Well, GB isn't the only number there, sandra was there as well.

http://ranker.sisoftware.net/show_r...d4ecdde9dde5d6f082bf8fa9cca994a482f1ccfc&l=en
http://ranker.sisoftware.net/show_r...d4ecdde9dde5d7f183be8ea8cda895a583f0cdfd&l=en

Same 16 core TR, 5% different results and higher clocked one actually does worse, though I consider sandra clocks to be bogus, the power numbers seem interesting though, probably also bogus.

Here's a better overview:
http://ranker.sisoftware.net/top_de...cee880bd88aed6ebdafc99fcc1f3dde8cebd80b0&l=en

Personally I think 16 core X model will end up with 1700X clocks, all core turbo and all but I'd like to be pleasantly surprised with a 2+ core XFR of 4.1GHz+, but who knows.

 

[Aenra]

To the person wondering about the PCIe slots spacing.. am not good at tact, so simply put, think about it a bit more

You'll be fine and able to fit the exact same cards/GPUs, in number or type, with either mobo; the one having them all equally spaced out, or the one that doesn't.. think about it a bit more (and spare me a wall of text typing the obvious).

 

[Atari2600]

You'll be fine and able to fit the exact same cards/GPUs, in number or type, with either mobo; the one having them all equally spaced out, or the one that doesn't..

Indeed. The different motherboards with different PCIe layouts (and, M.2) offer tangibly different products for different users who want different things.

3x M.2 drives in native slots? Yep, can do that on some boards.
4x PCIe x16? Yep, can do that on pretty much all boards.
2x PCIe x16 and 3 or 4x PCIe x1/x4/x8? Yep, other boards offer that.


My main reason for wanting to pay out for threadripper over Ryzen 7 is not CPU performance, its board I/O and platform longevity.

 

[TheGiant]

But Ryzen 1700 isn't going to cannibalize EPYC's sales

I think otherwise. Ryzen 1700 is cannibalizing AMD itself. IMO they should release the 1700X as 1700 and sell it for minimum 400USD and the actual 1700 shouldn't even exist. I bet it wouldn't hurt their sales. The one who wanted to buy ryzen would buy it even for 400 USD without problems.
ThreadRipper should be made by AMD as something really exclusive, the product itself doesn't sell. Even if it is really good.

 

[mattiasnyc]

Indeed. The different motherboards with different PCIe layouts (and, M.2) offer tangibly different products for different users who want different things.

3x M.2 drives in native slots? Yep, can do that on some boards.
4x PCIe x16? Yep, can do that on pretty much all boards.
2x PCIe x16 and 3 or 4x PCIe x1/x4/x8? Yep, other boards offer that.


My main reason for wanting to pay out for threadripper over Ryzen 7 is not CPU performance, its board I/O and platform longevity.

My thoughts exactly.

 

[mattiasnyc]

I think otherwise. Ryzen 1700 is cannibalizing AMD itself. IMO they should release the 1700X as 1700 and sell it for minimum 400USD and the actual 1700 shouldn't even exist. I bet it wouldn't hurt their sales. The one who wanted to buy ryzen would buy it even for 400 USD without problems.

I'm pretty sure AMD did their market research. If they thought they could get away with what you propose I'm sure they'd done it. They're not that clueless.

The way I see it the issues are first that we don't really know if there's a difference in quality between the 1700 and 1700x chips we buy. Suppose they're all tested, and the 1700 chips tend to do less well at higher frequencies as far as power consumption goes. It could be entirely reasonable for AMD to then rebrand those as lower powered lower speed 1700 chips, and keep the ones that can run a higher default frequency at lower power draw 1700x. In other words do we really know they're all "the same"? From what unscientific impression it seems that the 1700 chips aren't overclocking as high with the same power consumption as the "x" chips.

Secondly, not everyone wants to tinker with OC. So, there might be a case to make for making a tier of chips that run at different speeds by default, and by default then charging different amounts, regardless of the previous possible issue.

Lastly, you could look at it from a different perspective; instead of the 1700 being underpriced for what it is perhaps it's the "x" chips that are overpriced. Perhaps AMD did their research and found out that most people would totally consider buying a 1700 for X dollars, and that some people then would happily pay more for an "x" chip even though that was a less good deal. And in turn they might have found that people actually would not buy as many chips if the pricing was as you suggested as a whole, and that they'd lose revenue that way.

 

[tamz_msc]

I think otherwise. Ryzen 1700 is cannibalizing AMD itself. IMO they should release the 1700X as 1700 and sell it for minimum 400USD and the actual 1700 shouldn't even exist. I bet it wouldn't hurt their sales. The one who wanted to buy ryzen would buy it even for 400 USD without problems.
ThreadRipper should be made by AMD as something really exclusive, the product itself doesn't sell. Even if it is really good.

There are lots of people, myself included, for whom a 65W TDP 8-core CPU is far more appealing than a higher clocked CPU that would undoubtedly have a higher TDP. Basically people who'll run multithreaded workloads most of the time. For them the lower TDP can be very appealing. Not everybody cares about overclocking, and the extra few megahertz doesn't make any difference to them.