AMD Ryzen Threadripper 2990WX

[scaryjim]

... I don't think AMD really made TR4 to take a 4 chip CPU. ...

I think it is more likely the 2990WX exists because AMD wants a halo product to beat Intel but with a lower price,as the 18 core chips are nearly 500MM2 in size and any higher core count Intel chips will be even larger.

I think this is pretty much spot on. The workstation market was one Intel were climbing all over, be it with the Core i HEDT chips or the workstation Xeons.

What's amazed me is just how well AMD have executed in that market - if you look back to the 1950X review it was beating the 7900X by similar margins in MT productivity - Intel responded by bringing its next line of Xeon dies down into the HEDT workspace (hacking off ECC support and a couple of memory channels on the way), and AMD just hammered them again. And they're doing all this on a shoestring budget, with minor core and process improvements .... what's going to happen next year when they jump a process node and release a second-gen Zen core?!

[Biscuit]

What I don't understand is why Game mode is 8 core 16 thread? IIRC There are four 8 core CCX's in the package. Two of the CCX's have a dual channel memory controller.

Why not have game mode disable the 2 CCX's with no memory controllers and disable hyperthreading so each core has double the L1, L2 and L3 cache? Same 16 threads as well.

Even better, can this be fixed by setting processor affinity for games? If so, then no reboot required.

Sorry to be pedantic, but there are 4 x die on an interposer, each die has 2 x 4 core CCX.

[philehidiot]

I'm getting one for gaming. That the Pifast performance is slower than my relatively ancient i5 is not relevant. I am inadequate in the trouser department and as a result I have a motorcycle and now need a 32 core CPU. It is not acceptable to not have under 8 cores these days and this is clearly the way to improve COD frame rates.

And that the cooler is RGB is obviously proof it's aimed at my demographic. And better. Just glow in the dark better.

The e-pr0n here is clearly exceptional and totally gaming related.

[scaryjim]

Sorry to be pedantic ...

Are you really sorry?!

Amused by the "low power" bit of that second slide … Anandtech did some detailed power measurement and found that the uncore uses ~ 1/3 of the total power budget! Some of that will be memory and PCIe, mind you, but it's still a massive proportion of the power not being used for compute...

[CAT-THE-FIFTH]

I think this is pretty much spot on. The workstation market was one Intel were climbing all over, be it with the Core i HEDT chips or the workstation Xeons.

What's amazed me is just how well AMD have executed in that market - if you look back to the 1950X review it was beating the 7900X by similar margins in MT productivity - Intel responded by bringing its next line of Xeon dies down into the HEDT workspace (hacking off ECC support and a couple of memory channels on the way), and AMD just hammered them again. And they're doing all this on a shoestring budget, with minor core and process improvements .... what's going to happen next year when they jump a process node and release a second-gen Zen core?!

As long as 7NM does not have any gremlins,AMD might be in a decent position next year too.

Also,looking at this Win10 against Linux comparison,it seems Linux works better with the 2990WX:

https://www.phoronix.com/scan.php?pa...-windows&num=1

[Tunnah]

Could do with getting a higher-core Intel chip to see how that performs too. A lot of the remarks against it were due to software not being able to keep up, but is this solely an AMD issue or will an Intel chip with 28 cores have the same problem ?

The review just reads very unfairly. This isn't a criticism in the slightest, just an observation. It feels like the part was reviewed for usage it isn't designed for, and it does a disservice to such an incredible achievement. I mean..32 cores for less than 2 grand!

Ain't technology grand eh

[stoatwblr]

either my Google-fu is letting me down or no ones ever ran gaming benchmarks on high core count Xeons.

Considering the cost of the high core count Xeons, getting gaming benchmarks might be difficult (and would probably be disappointing). I have a 2 socket * 28core system arriving in the next few days though (the cpus were _half_ the cost of the 32 core ones - and saving $8k was worth the core count reduction) - only problem is the chassis isn't setup for GPUs.

[ik9000]

Read more.

@Hexus - Can you add power draw when in gaming mode to the charts? Be interesting to see how that changes with half the cores gated off (or whatever it is the switch does).

[Muaadib]

32 cores for $1800. Competition does smell very nice.

[Corky34]

Amused by the "low power" bit of that second slide … Anandtech did some detailed power measurement and found that the uncore uses ~ 1/3 of the total power budget! Some of that will be memory and PCIe, mind you, but it's still a massive proportion of the power not being used for compute...

Not disagreeing but 2pj per bit is pretty darn good for an interconnect, IIRC Intel's solution of a mesh of half-rings uses quiet a lot more.

I think Anandtech hit the nail on the head when he said the big battles are going to be fought over interconnects in the future, as yourself and he pointed out they can take up large portions of a total power budget.

[3dcandy]

This is pretty much a because we can product in many ways, but also shows that AMD are on the right track as they can scale from just a few cores right up to 32. Only problem is how many will move away from an Intel ecosystem even if the gains are big when Intel have owned the market for so long...
I'd be interested to see if they can break the stranglehold at all with this. I'm pleased they've forced change and it's great to have 2 seperate companies providing this much power at a relatively small outlay

[Biscuit]

Not disagreeing but 2pj per bit is pretty darn good for an interconnect, IIRC Intel's solution of a mesh of half-rings uses quiet a lot more.

I think Anandtech hit the nail on the head when he said the big battles are going to be fought over interconnects in the future, as yourself and he pointed out they can take up large portions of a total power budget.

I wonder how long it will take before intel has a multi-die+interposer system, certainly not in 2019. I foresee them making some losses on monolithic chips.

[Dashers]

Perhaps we'll see multi-socket motherboards appearing in the "gaming" market. Buy a n cores today, and more in 6 months.

I'm really, really, turned off by game mode. But on the workstation front, I'd definitely opt for one of these over an Intel for my work computer which does no gaming at all.

[scaryjim]

… It feels like the part was reviewed for usage it isn't designed for ...

Depends on perspective, I guess - AMD has launched it into an HEDT platform, and from a platform point of view it's being tested appropriately. I thought Hexus did a reasonable job of pointing out that this was an outstanding workstation chip, but if you were looking for an HEDT platform for gaming you should look elsewhere.

Not disagreeing but 2pj per bit is pretty darn good for an interconnect, IIRC Intel's solution of a mesh of half-rings uses quiet a lot more. ….

Read up the Anandtech review. The Intel uncore uses a lot less energy overall (although I'm not 100% sure on the j/bit). Although one thing I have noticed on re-reading the AT review is that they miss the very obvious, glaring fact that doubling the number of die and adding die-to-die interconnects doesn't even double the uncore power draw, when it should double it 'plus a bit'. So increasing the number of IF links seems to improve their efficiency somehow...?

I wonder how long it will take before intel has a multi-die+interposer system, certainly not in 2019. ...

They had an MCM in the consumer market in 2006. All Core 2 Quads were an MCM of 2 Core 2 Duo dies. It worked very well for them, but makes their current line on Threadripper a bit hypocritical...

Perhaps we'll see multi-socket motherboards appearing in the "gaming" market. Buy a n cores today, and more in 6 months. ...

Also been done before, again with Core 2 (and I think AMD had a similar product - the marketing name Quad FX is screaming in my hind brain). The monolithic core/thread counts are now so high that I don't see many games ever swamping them. No doubt you could write a game that would hammer 16 threads, but I doubt it would be an efficient way to code - many tasks just don't parallelise that well...

[CAT-THE-FIFTH]

Not disagreeing but 2pj per bit is pretty darn good for an interconnect, IIRC Intel's solution of a mesh of half-rings uses quiet a lot more.

I think Anandtech hit the nail on the head when he said the big battles are going to be fought over interconnects in the future, as yourself and he pointed out they can take up large portions of a total power budget.

Which does indicate the Zen cores are very low power. It might also explain why IF is tied at 1:2 rate to memory frequency as it might be down to power - Vega also uses IF too.

[Corky34]

Read up the Anandtech review. The Intel uncore uses a lot less energy overall (although I'm not 100% sure on the j/bit). Although one thing I have noticed on re-reading the AT review is that they miss the very obvious, glaring fact that doubling the number of die and adding die-to-die interconnects doesn't even double the uncore power draw, when it should double it 'plus a bit'. So increasing the number of IF links seems to improve their efficiency somehow...? .

I read it, and i think it's a bit yes and no. It seems Intel's interconnect use less power when the active core count is low, as the active core count goes up so does the interconnects overall share of the total power usage, it starts off using less power but it doesn't seem to scale very well.

It seems with IF that's the other way around, you pay more in power terms on low active core counts as you have to light up all the interconnects, but as the active core count goes up the percentage of total power that the interconnect consumes goes down and then remains fairly constant.