AMD - Piledriver chitchat

[CAT-THE-FIFTH]

TH tests the A10 7800:

http://www.tomshardware.com/reviews/...ency,3899.html

[CAT-THE-FIFTH]

Looks like Tonga is being released on the 23rd:

http://cdn.videocardz.com/1/2014/08/...f-amd-gpus.jpg

Interesting "What's Next?" - I wonder if it means HBM??

[scaryjim]

TH tests the A10 7800

Those oscilloscope traces They're fantastic! Interesting to see how high those transient peaks go: they could easily explain why my DC-DC PSU was tripping despite being rated for more than the at wall draw of my old system. Another good reason to avoid skimping too much on your PSU, I guess....

http://cdn.videocardz.com/1/2014/08/...f-amd-gpus.jpg

Eyefinity's been around for 5 years already? Lawks I feel old....

[CAT-THE-FIFTH]

Those oscilloscope traces They're fantastic! Interesting to see how high those transient peaks go: they could easily explain why my DC-DC PSU was tripping despite being rated for more than the at wall draw of my old system. Another good reason to avoid skimping too much on your PSU, I guess....



Eyefinity's been around for 5 years already? Lawks I feel old....

They were using a 65W DC-DC PSU for the A10 7800.

[scaryjim]

They were using a 65W DC-DC PSU for the A10 7800.

Yeah I know - my DC board only supported 5A on the 12V line (so 60W) and was tripping when I turned the computer on. The total at wall draw off my really inefficient normal PSU was only ~ 60W, but looking at their figures (where peak draw can spike 20W above average) there's a good chance the 12V line was overloading and tripping the OCP. Might need to think about a DC board for my audio workstation I'm currently rebuilding....

[CAT-THE-FIFTH]

Yeah I know - my DC board only supported 5A on the 12V line (so 60W) and was tripping when I turned the computer on. The total at wall draw off my really inefficient normal PSU was only ~ 60W, but looking at their figures (where peak draw can spike 20W above average) there's a good chance the 12V line was overloading and tripping the OCP. Might need to think about a DC board for my audio workstation I'm currently rebuilding....

However it is interesting that under normal testing(not stupid Prime95 and Furmark) the AMD APUs are not that inefficient.

CPU performance looks a match for the Core i3 3220(going by the Core i3 2100 and Core i3 4330 results).

So basically Core i3 3220 level performance with an R7 250 level IGP(or better) in a 45W to 65W TDP CPU is actually not too bad at all.

Edit!!

It seems APU shipments have increased:
http://www.techpowerup.com/204282/jp...dia-slips.html

Both AMD and Nvidia have decreased discrete GPU shipments.

[CAT-THE-FIFTH]

The FX8370E looks pretty decent:

http://www.xbitlabs.com/news/cpu/dis...rocessors.html

4.1GHZ to 4.3GHZ clockspeed but only a 95W TDP!

[CountBartok]

The FX8370E looks pretty decent:

http://www.xbitlabs.com/news/cpu/dis...rocessors.html

4.1GHZ to 4.3GHZ clockspeed but only a 95W TDP!

8370E looks like quite a nice chip. Have to see what UK pricing pans out like.

[arsenalfan]

Are these new AMD CPU's worth it or should I just go with an FX 8350?

[Noxvayl]

The price drop for the FX 8350 is great value for your money to be honest. I'm not sure the new chips will be anything to be excited about though. The increased frequency is only 100MHz which all FX 8350/8320 can handle without even changing the voltage.

For stock use you won't see much improvement in performance but since it is so close to release it wouldn't hurt to see what reviews say, FX 8350 won't be disappearing.

[shaithis]

They look like re-hashes....personally I wouldn't wait unless the low power variant really tickles your fancy.

[watercooled]

That's pretty much what they are AFAIK, with process maturity affording clock bumps and/or power reductions.

It could be interesting to see how power consumption compares to the earlier samples, but AMD tend not to bin too aggressively anyway, so maybe an undervolting comparison?

[watercooled]

Haswell-E is out: http://anandtech.com/show/8426/the-i...7-5820k-tested

Not read the full thing yet but got to the die size; 356mm2. How big is the supposed 18 core going to be???
As I think I said earlier on, maybe we'll see an MCM like AMD have been doing with their larger Opteron parts?

Edit: Something else has just occurred to me though, looking back at this http://hexus.net/tech/news/cpu/72585...heat-spreader/ I'd guess that isn't a 5960X, or not a retail one at least. Considering the die size Intel have published for the 8 core, it seems about right when compared to the IVB-E six core die size, so it seems unlikely there are 12 cores present on the retail die.

It does still look like a variant of the same architecture though; you can see the cut-off line at the right of the die on both images with some blank/unlabelled area on the 8 core where the cache would be on the 12 core.

[kalniel]

Haswell-E is out: http://anandtech.com/show/8426/the-i...7-5820k-tested

Not read the full thing yet but got to the die size; 356mm2. How big is the supposed 18 core going to be???

Last gen 15 core was something like 541mm2. This should be a little larger due to the DDR4 controller and then the extra cores might add a bit too. Impressive stuff from Intel.

[watercooled]

Also the Haswell cores are bigger, but seemingly not by a huge margin going by the desktop die sizes and transistor counts. Although process maturity of course affords some more usable die space over IVB.

Intel had Tukwila at just under 700mm2 on 65nm so it's not exactly unheard of for them to do crazy die sizes for markets which really aren't all that cost sensitive.

We could easily see something over 600mm2 depending on uncore size!

Edit: Also, Beckton (Nehalem 8C) is 684 mm² on 45nm, so I guess this sort of die size isn't unusual on Tocks, just SNB was skipped so we've had two 'EX' Ticks in a row.

[scaryjim]

Hmmm, just looked at some of the Haswell E reviews, and I'm confused. I thought the whole point of DDR4 was that each DIMM was meant to have its own channel on the controller. Yet there seem to be plenty of X99 boards with 8 DIMM slots, presumably using some of the channel-sharing tech that was meant to only really be appropriate for servers. So what's the deal with that? It rather horribly suggests that perhaps when DDR4 moves to the mainstream platform there's a really possibility that we'll end up stuck with 4 DIMMs running off a dual channel controller - which isn't what I'd call progress....