They should have designated the top of range chip the 1999P and called it the Prince of Threads. It'd certainly be party time for number-crunchers.
They should have designated the top of range chip the 1999P and called it the Prince of Threads. It'd certainly be party time for number-crunchers.
i can see the threadripper helping me with my dvd editing. handbrake and dvdfab maybe.
$100 Less vs. Intel versions tops for pricing. Anything more than that, assuming it wins most apps, loses games, just like current ryzen chips, management should be fired for being too stupid to make money (4 more years of no profits again). They need to make money, and pricing too low is how you lose money (for decades BTW, about 8B over the last 2 decades of losses at AMD). I'm guessing they already pissed away any vega profits (2 times delayed due to HBM2 etc, Nvidia already answered, etc). You haven't made a dime with a whole Quarter of selling ryzen. It would appear ONLY servers and HEDT can save your bottom line, so for the love of GOD, charge what they are worth, not what people WANT to pay. Stop screwing your products (console crap instead of Vega/Ryzen etc all out 2-5yrs ago), and stop screwing the pricing on those products when you actually have GREAT products that WOULD fetch much more money if you had a management team with at least SOME idea of what a quarterly report should look like.
It's pretty clear Intel isn't after a massive price war since they just put ships on top of older ones (all the way up to $2100 for the 18 core probably). So price just under each of theirs or even the same price if you have 2 more cores vs. Intel's chip. Quit giving stuff away so cheap or get used to another 4yrs of LOSSES even though you have a great GROUP of products this time. I won't be too shocked if a year from today you still haven't made a profit for more than a quarter. Surely they should be able to get 1 out of 4 quarters right with all the new products they have coming.
I did. In the comments to the EPYC article.
It looks like these are using the same physical socket as the NaplesEPYC server chips, in which case the package has to support up to 4 die MCMs, which is going to make the fully assembled processor huge. It's one of the reasons I was dubious about that rumour, but it appears to have turned out to be true (like most of the others, which is why I'm hesitant to call BS on the 10 & 14 core processors*).
I'm also intrigued that AMD are talking about the X399 platform, rather than chipset, and pushing the up to 64 PCIe lanes. That's straight from their EPYC server playbook, where there is no chipset and everything just hangs off a huge PCIe cluster on the CPU. I strongly suspect we'll see around 40 PCIe lanes actually plumbed out to PCie slots on most motherboards (a couple of x16s plus some x4s and x1s), with the others used for directly attached storage (a couple of x4 NVMe, maybe) or to connect something like the X370 chipset (in fact, they could easily just slap the X370 chip on the motherboard - it connects to the CPU using an x4 PCie link and is basically just a port multiplier)...
*EDIT: just had an epiphany. I've been working on the basis that cores have to balanced, but what if they only need balancing across the CCXes within a die? so a 14C TR doesn't need to be 7+7 across dies - it can be 8+6 (i.e., 4+4 & 3+3). A 10 core can be 6+4. Balanced on each die, asymmetric between dies on the MCM. I'm suddenly a lot more willing to consider 14 and 10 core TR chips
Last edited by scaryjim; 02-06-2017 at 12:22 PM.
There are currently 1 users browsing this thread. (0 members and 1 guests)