-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)
- Thread starter Boomstick777
- Start date
- Status
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,335
- Location
- Planet Earth
Lets hope the R9 370X is a decent chunk under £200 - we really need some more competition in that market now.
Makes no sense for the supposed 385 to have 2GB memory when the 380 has 4...
And if the 380/X is a re-badged Hawaii, there will be no reason at all to buy one over a 2nd hand 290/X from the MM here.
I must say, the more rumours we see, the less excited I am for the 3xx series. Mid-range cards looking pretty dire, and even the top end has only 4GB. Just not good enough from AMD, esp at a time when they could really hurt nV with some good cards.
And if the 380/X is a re-badged Hawaii, there will be no reason at all to buy one over a 2nd hand 290/X from the MM here.
I must say, the more rumours we see, the less excited I am for the 3xx series. Mid-range cards looking pretty dire, and even the top end has only 4GB. Just not good enough from AMD, esp at a time when they could really hurt nV with some good cards.
No where near the same i have 5670 4.4ghz and a 3930k which will reach 4.8ghz in cinebench
Gulftown got just under 1000 points
Sandybridge-E @ 4.8ghz got just under 1200 points
So the ipc is roughly about 20% faster inculding overclocking head room ill run it @ 4.4 and you will see how much faster it is at the same clocks
Edit @ 4.4ghz got 1100 points so 10% faster at the same clocks but nearly 20% when you add in the overclock.
Last edited:
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Nothing stopping the 390x being on 28nm, the 385 would never, ever have 2GB nor 256GB/s of bandwidth. A rebaged Hawaii with 512bit memory bus isn't possible, it doesn't have a 512bit bus, if one came out with a 512bit bus, it would not be rebadged bandwidth wasn't any great limited for it at that size. It's possible that moved over to Global for a better 28nm process that they added in more bandwidth but it doesn't make a whole lot of sense beyond getting 4GB on it more comfortably, no real reason to stick with gcn 1.1 and tape out an old/new product.
Basically it's all nonsense.
As for yields, there ARE yield issues when adding things to an interposer. You can have a $200 gpu, and 4 stacks of HBM at however much they cost(more than gddr5 for the same capacity.... but you do make savings on power circuitry, pcb production, other areas, still hbm likely costs more), if one chip fails then the entire lot goes in the bin.
Yields are said to be surprisingly good for the interposer connection process, but they still diminish the more chips you have and that is why most people assume a hard limit somewhere. There is also the small fact that you have to connect each chip to somewhere on the die. A current 512 or 256bit memory controller is generally speaking split up into multiple smaller memory controllers, 4-8 of them, and memory will connect to each one. YOu can't just magic up new connections, the chip has to be designed up front for X amount of connections to memory chips, be it gddr or hbm. To connect 8 stacks of gddr you'd both get 1TB/s of bandwidth, it would cost a fair amount in power and it would cost double the amount in die size(for the memory controller) to connect twice as many chips.
Basically it's all nonsense.
As for yields, there ARE yield issues when adding things to an interposer. You can have a $200 gpu, and 4 stacks of HBM at however much they cost(more than gddr5 for the same capacity.... but you do make savings on power circuitry, pcb production, other areas, still hbm likely costs more), if one chip fails then the entire lot goes in the bin.
Yields are said to be surprisingly good for the interposer connection process, but they still diminish the more chips you have and that is why most people assume a hard limit somewhere. There is also the small fact that you have to connect each chip to somewhere on the die. A current 512 or 256bit memory controller is generally speaking split up into multiple smaller memory controllers, 4-8 of them, and memory will connect to each one. YOu can't just magic up new connections, the chip has to be designed up front for X amount of connections to memory chips, be it gddr or hbm. To connect 8 stacks of gddr you'd both get 1TB/s of bandwidth, it would cost a fair amount in power and it would cost double the amount in die size(for the memory controller) to connect twice as many chips.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
I can't say I am overly impressed if it does only have 4GB. It clearly isn't enough now for 4K. No point having the grunt with a couple of them but not having the VRAM.
I am disappoint AMD if this is true
So does it make sense to NOT release any faster card unless it's fully capable of being decent at 4k. Don't bother to release a 4GB card soon because the 2-3% of users with 4k screens might have memory problems? Even if the card is fine for 97% of users, don't make something faster, don't address the 97% of the market because the 3% aren't suited to this card. Does that make sense to you, really?
You have guys like Rroff who went 4k, then went backwards to get a more responsive lower res panel.... because 4k is almost pointless with crap slow unresponsive panels?
Why should the 97% who game at sensible resolutions pay more so the card works great at 4k? I don't want to, most of the 97% don't want to. Fact is IF it has 4GB only and no 8GB version of the card it won't be a choice AMD made to not offer a 8GB card, it will be a technical limitation of HBM. So again, because HBM doesn't offer the capacity to suit the tiny portion of 4k gamers, AMD shouldn't offer a card that is monumentally faster for all other users as a result? Nonsense.
AMD must surely see that people are clamoring for 8GB cards right now, especially with 4K on the rise, and with the current crop of 8GB 290X cards selling so well. Unless they simply CANNOT produce a card with this much memory due to technical reasons, it's a huge missed opportunity IMO if they don't release until Q4. Makes the current deals on the 8GB cards even MORE appealing if you'll be waiting almost a year for a new 8GB card to hit the market, which will most likely be twice the price of the ones available today anyway!
I have to agree to some extent, but why haven't NV done it already, both AMD and NV have been touting 4K for a while and AMD turned around and said ok we dont have a new card but here is an 8GB version of an existing card that does not rip your wallet apart, the 980 should have been the 8GB part by now.
We need 4,6 and 8GB cards and 16GB for the duals
Last edited:
What part of "I" did you not get? I don't think for everyone and give what my thoughts are. If you want 4GB and are happy, sweet but "I" want to go back to 4K and "I" know that 4GB isn't enough.
The other 97% can think what they like and I accept their thoughts (even if it doesn't coincide with mine). It doesn't make 4GB any better for me though now does it?
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Yeah, except that doesn't gel with what you posted. YOu didn't say you were disappointed in this card, or disappointed it wouldn't work for you, you said you were disappointed in AMD. Disappointed AMD would make a 4GB card which MIGHT not be great at 4k(I'll place money that it will spank, utterly spank the 290x/980/780ti/titan in 9 out of 10 4k games) as if that was a deciding factor in AMD's product decisions.
You can only be disappointed directly in AMD for a product if you think it's a bad product.
Were you disappointed in Nvidia for releasing the 750ti? Did you buy one, was it going to be any good for you and your level of gaming? Or were you not disappointed in Nvidia, just disappointed in that specific card because it wasn't the card you wanted.
Soldato
- Joined
- 27 Feb 2012
- Posts
- 6,586
I'm disappointed in AMD. If these specs are true then we're looking at stuttery games for proper enthusiasts, the market they claim to be aiming for.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
No, it's not, 4GB = 4GB, HBM won't change that. However it's a made up specification where the guy making this stuff up hasn't got the slightest clue what he's talking about. You match bandwidth and memory to the amount of bandwidth required on a card and the amount of memory the power level is aimed at. With over 3k cores there isn't the slightest chance in hell of pairing it with less bandwidth than a 7970 has and only 2GB of memory.
2GB and 256GB/s of memory would limit the 7970 with 2048 cores, there is no chance at all AMD would use 2GB/256GB/s of memory for a card with another 50% cores.
Why people continue to be led by two sites nonsense in particular I don't know, videocardz and the other one, wcftech or whatever it is, both entirely useless. Not only is every other thing they post contradictory, they are constantly wrong and show a basic lack of understanding of the technology involved in these products.
Last edited:
Don't turn this into a fanboy war, as that won't cut it with me and this is exactly what I said.
Notice the last bit where again I say "I" am disappoint AMD if this is true. Nowhere does it say "AMD have messed up for everyone and they should be releasing 8GB cards now does it?
Try reading what "I" type and seeing it from my perspective and not the opinion of everyone. I don't have to start putting a disclaimer into my posts now do I?
Please note that these are my thoughts and not the thoughts of everyone.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Because HBM is in it's infancy and there might simply not be a realistic possibility of putting 8GB on a card right now.
If no one makes an HBM product, HBM production doesn't start. If HBM production doesn't start, it doesn't improve, doesn't ramp up and stops being invested in. To get the next HBM product, you need to get them to make the first one, meaning you have to take what can be produced. Which looks currently like 128GB/s of bandwidth in 1GB stacks. There are yield(and therefore significant cost implications) of going beyond a certain number of stacks, meaning 4GB is doable, maybe even past the ideal number of stacks and 8GB could become effectively impossible at this point.
If there is no 8GB option(and considering 90% of the article is made up) it won't be because AMD actively chose not to offer one, it will be because it's not viable yet. But making HBM products today brings the date the 2-4GB stacks closer, making the next gen 8-16GB products more viable.
Less total to access but a much higher bandwidth. The way the specs are written out for HBM it looks like there's no "bus" as we traditionally know it so It might be that it can just swap huge chunks of data out very quickly and so needs a smaller "pool" to access.
Or maybe the bandwidth doesn't help and you do need the higher total. we wont know till they get benched...properly.
- Status